AMD: Zen will offer 40% faster performance per clock than Carrizo

[gamerk2]

L3 is expensive though; do remember AMD actually needs to make a decent amount of profit on their chips.

That's AMDs real problem: what little OEM support they had died with the BD arch, simply because of thermal/power constraints. And now, Intel has the stronger iGPU, making it even harder for AMD to get into the OEM market. No matter how Zen performs at the top, AMD won't sell enough to make a return on investment if 99% of the market never purchases their product.

 

[Pieter3dnow]

L3 is expensive though; do remember AMD actually needs to make a decent amount of profit on their chips.

That's AMDs real problem: what little OEM support they had died with the BD arch, simply because of thermal/power constraints. And now, Intel has the stronger iGPU, making it even harder for AMD to get into the OEM market. No matter how Zen performs at the top, AMD won't sell enough to make a return on investment if 99% of the market never purchases their product.

You are getting 2 things mixed up. Igpu means nothing to Zen because Zen will be CPU only. Intel had the power thing because they were producing on lower nm FinFET.

Where your expectation of Zen needs to be a mass selling product mine would be a step in the right direction. The fatalistic thought pattern of sell or bust does not make sense in this case. AMD needs those contracts but Intel locks up the OEM market you would guess that AMD has other plans than that for Zen as well...

 

[Zarathustra[H]]

Adding the FP execution units will certainly help in FP performance, but that by itself doesn't do much to move the needle on Integer based performance.

That's the primary reason I'm VERY skeptical of 40%; sure, FP might see that much improvement because you remove the bottleneck, but but for Integer workloads, I don't see much movement on performance.

I agree with what your saying here, but I am less skeptical.

I feel like the shared FPU unit has been blamed way too much for the bulldozzer architecture's poor performance. After all it is a 256bit FPU that can split itself into two 128bit FPU's when needed to support each integer core.

Bulldozer DOES have two REAL INTEGER CORES per module. They are just very poorly performing real integer cores.

The root of Bulldozers performance is the very basics of the integer core. it's slow as all hell and can't keep up.

Now, the reason why I am less skeptical about their 40% claims is because ZEN is supposed top be a ground up redesign. They aren't just highlighting, copying and pasting a bulldozer core out of it's CMT home into a new SMT home. If they were going to do that, they wouldn't have needed Mr. Keller.

The 40% values are probably optimistic (these early estimates usually are) but I think we will see a significant improvement over Bulldozer.

 

[Mr. Bluntman]

The root of Bulldozers performance is the very basics of the integer core.

Don't forget about the fact that the integer pipeline has just as many if not more stages than Prescott and each integer core is only 2-wide.

 

[defaultluser]

Don't forget about the fact that the integer pipeline has just as many if not more stages than Prescott and each integer core is only 2-wide.

Right, I've discussed in detail on previous threads how Bulldozer was supposed to ship at 4.5-5 GHz (L3 cache latency is a clear clue what speed they were targeting), and missed that target by a mile. They hit the power wall way before they hit their desired frequency.

They thought they could do the same thing as the Pentium 4, except this time somehow get it right where Intel failed.

This is why I believe they could do 40% higher IPC. They were already halfway there with Stars.

 

[Zarathustra[H]]

This is why I believe they could do 40% higher IPC. They were already halfway there with Stars.

Yeah, people seem to forget that Bulldozer was a step backwards from an IPC perspective...

Right, I've discussed in detail on previous threads how Bulldozer was supposed to ship at 4.5-5 GHz (L3 cache latency is a clear clue what speed they were targeting), and missed that target by a mile. They hit the power wall way before they hit their desired frequency.

They thought they could do the same thing as the Pentium 4, except this time somehow get it right where Intel failed.

That does seem kind of arrogant IMHO.

Intel with it's army of engineers and best-in-industry fabs couldn't pull it off, but we can?

 

[Darakian]

L3 is expensive though; do remember AMD actually needs to make a decent amount of profit on their chips.

L3 isn't expensive, space on silicon is expensive and L3 takes up a large amount of die. This is less now than when BD launched for two reasons
1) We're looking at 14/16nm products which means 1MB of cache is smaller than it used to be
2) 8MB seems to be a sweet spot for most applications

That's AMDs real problem: what little OEM support they had died with the BD arch, simply because of thermal/power constraints.

I dunno. They got sony and microsoft to buy a large number of chips. Nintendo is expected to buy a good number as well with their next console.

And now, Intel has the stronger iGPU, making it even harder for AMD to get into the OEM market. No matter how Zen performs at the top, AMD won't sell enough to make a return on investment if 99% of the market never purchases their product.

Intel does not have the stronger gpu. They have a gpu with faster ram that costs twice what the AMD chip costs.

 

[Mr. Bluntman]

Yeah, people seem to forget that Bulldozer was a step backwards from an IPC perspective...

So much so that Phenom II X6 parts still command insane premiums on ebay for anything past 2.6GHz.

That does seem kind of arrogant IMHO.

Intel with it's army of engineers and best-in-industry fabs couldn't pull it off, but we can?

Arrogant indeed. If Intel can't do it, who can? Certainly not AMD in their current (or even past) state. Makes me wondering what kind of good shit they're smoking in the labs.

 

[Mr. Bluntman]

I dunno. They got Sony and Microsoft to buy a large number of chips. Nintendo is expected to buy a good number as well with their next console.

Intel does not have the stronger gpu. They have a gpu with faster ram that costs twice what the AMD chip costs.

Which are based on their Jaguar low power/low performance cores that are designed to be in an SoC, not Bulldozer. No word about what NX will have (hopefully Zen cores).

eDRAM is a hell of a thing, and it's the only reason Intel has the lead, but as you said, at double the cost of the competition. AMD has experience with eDRAM designs (Flipper for GameCube, Xenos for Xbox 360) and I'm perplexed as to why they haven't implemented it other than maybe package size constraints.

 

[variant]

I agree they should have just moved forward with the K10 instead of designing Bulldozer. It would have been a lot cheaper and probably kept them more competitive in the budget market.

 

[Darakian]

Which are based on their Jaguar low power/low performance cores that are designed to be in an SoC, not Bulldozer. No word about what NX will have (hopefully Zen cores).

Given nintendo's history I would be quite surprised if they had zen cores in the NX. My guess is that nintendo wants arm cores, but that's just a guess (pm me if you'd like my reasoning as it's quite off topic).

eDRAM is a hell of a thing, and it's the only reason Intel has the lead, but as you said, at double the cost of the competition. AMD has experience with eDRAM designs (Flipper for GameCube, Xenos for Xbox 360) and I'm perplexed as to why they haven't implemented it other than maybe package size constraints.

I think you hit on exactly why they haven't put it on a apu. When AMD wants to sell a cpu they do some calculus on how much it will cost to make (ProdCost), how much they can sell it for (RetailCost) and how many they can sell (Volume).

That gives you an (overly simplified) equation
Profit = Volume * (RetailCost - ProdCost) - FixedCosts
With the relation that volume drops as production cost rises. Also if they doubled the retail cost of their APU they would probably sell far fewer than half what they're selling now so we have a non-linear relation. The numbers probably just don't work out.

 

[NaroonGTX]

You are getting 2 things mixed up. Igpu means nothing to Zen because Zen will be CPU only. Intel had the power thing because they were producing on lower nm FinFET.

Initial Zen products will be CPU-only, but there will be Zen-based APU's coming in 2017.

40% IPC increase is pretty realistic. Zen was a design that was done by a team that was allowed to do whatever they wanted to do, and were given as much time as they needed to do it. Compare and contrast with Bulldozer, which was in the works from as early as 2005 and went through several canned iterations before it shipped in 2011. 45nm Bulldozer was so bad it got scrapped outright. They still couldn't fix the issues by the time the 32nm Bulldozer came in 2011. Bulldozer had a bunch of little problems that when combined together, became a big problem and that was why we had the lackluster performance we saw with it. Not to mention the fact that Bulldozer itself was a design that was designed and implemented ages ago, when the software ecosystem was drastically different than what it was a decade ago and not quite what they expected it to be when the product finally launched.

None of that was the case with Zen, and when one of the lead designers (Suzanne Plummer) goes on record promising a huge increase in not only performance, but power efficiency, I'm inclined to believe her. Just how much, we're gonna have to wait and see when silicon starts appearing. We're probably gonna see some results in Sisoft and F@H soon since test chips will be making their rounds not long from now. Bulldozer could've been much better if it didn't have such terrible L2$ latencies and just a much more efficient cache subsystem altogether. L1$ as well. Then there's the front end which was gimped since even though it could decode four instructions per clock like Sandy could, it couldn't service both int units at the same time, so you'd get 2 instructions per clock per core peak, and it was servicing the cores every other cycle... Just a really weird and out-of-the-box concept (CMT) that didn't quite work out the way they expected it to. At least Excavator looks decent, but it's so far along in the making. It's what Bulldozer should've been from the get-go.

 

[NaroonGTX]

I agree they should have just moved forward with the K10 instead of designing Bulldozer. It would have been a lot cheaper and probably kept them more competitive in the budget market.

Stars was already at the end of its lifespan. There wasn't much more they could've done with it. See the Husky (Llano) core revision which took Stars to 32nm, wasn't anything bad but wasn't groundbreaking either. K10 was just an updated evolution of K8, which itself was an evolution of K7. Pretty long-in-the-tooth uarch all things considered.

 

[Pieter3dnow]

I do not subscribe to certain problems with Bulldozer even tho it was not a true consumer cpu it still performs good to very well under things as Mantle and DX12 as long as you are pushing batches beyond 60K+ .

What the problem is that IPC is still king when that stuff is not going to be fixable in the Bulldozer architecture.
What Zen is bringing are 2 things Jim Keller and lower manufacturing process unless 2 of these things absolutely fail it should do well. I' can't see AMD and Jim Keller cpu team fail so badly, what i can imagine that the manufacturing process will need some extra attention.

Coming back to Bulldozer when this architecture would be important would be where you have rely on things as Mantle/DX12/Vulkan, the cpu allows to push a lot of data to the card which is important rather then relying on the cpu as it is now with DX11/9.

If Zen has the same core count of 8 (or higher) then were going to have a good time for gaming...

 

[razor1]

I do not subscribe to certain problems with Bulldozer even tho it was not a true consumer cpu it still performs good to very well under things as Mantle and DX12 as long as you are pushing batches beyond 60K+ .

What the problem is that IPC is still king when that stuff is not going to be fixable in the Bulldozer architecture.
What Zen is bringing are 2 things Jim Keller and lower manufacturing process unless 2 of these things absolutely fail it should do well. I' can't see AMD and Jim Keller cpu team fail so badly, what i can imagine that the manufacturing process will need some extra attention.

Coming back to Bulldozer when this architecture would be important would be where you have rely on things as Mantle/DX12/Vulkan, the cpu allows to push a lot of data to the card which is important rather then relying on the cpu as it is now with DX11/9.

If Zen has the same core count of 8 (or higher) then were going to have a good time for gaming...

Bulldozer was supposed to have 50% increased performance due to increased IPC, its about the ability to traslate that directly to performance, not just what's there.

Keller might be good, but I don't think that will be enough, with a 40% increased IPC and that is if its all of it translates over it still wouldn't match Intel's current line up, it might catch up with Sandy Bridge, which is 3 gens behind.

Process wise AMD will always be behind Intel, by the time AMD comes out with 14nm, Intel will be on 10 nm......

Yeah, telling people to use a different API because your CPU has overhead due to it being worse in overall performance to an Intel counter part and more overhead through graphics drivers isn't a great way to go......

 

[drescherjm]

Bulldozer was supposed to have 50% increased performance due to increased IPC

I say that part was BS from the marketing team at AMD.

Bulldozer was actually designed to have a lower IPC than Phenom II but at a higher frequency. Remember that they took 125% of the transistors of a single Phenom II core and made that into a 2 core Bulldozer module with each core having less processing units than they had in Phenom II.

Process wise AMD will always be behind Intel, by the time AMD comes out with 14nm, Intel will be on 10 nm......

This will not happen since 10nm is delayed by at least 1 year maybe more. AMD will have 14nm next year. 10nm Intel is probably late 2017 to early 2018.

 

[Pieter3dnow]

Bulldozer was supposed to have 50% increased performance due to increased IPC, its about the ability to traslate that directly to performance, not just what's there.

Keller might be good, but I don't think that will be enough, with a 40% increased IPC and that is if its all of it translates over it still wouldn't match Intel's current line up, it might catch up with Sandy Bridge, which is 3 gens behind.

Process wise AMD will always be behind Intel, by the time AMD comes out with 14nm, Intel will be on 10 nm......

Yeah, telling people to use a different API because your CPU has overhead due to it being worse in overall performance to an Intel counter part and more overhead through graphics drivers isn't a great way to go......

That 50% was something the server guy heard (JF_AMD on this board), he is no longer with AMD, what he thought he understood what was something he didn't understand fully and drew that conclusion.
If Keller is such a bad person why did Apple hire him he did something right there? And beside that AMD wasted enough money and time on things that just did not work and it seems that was not the plan for Zen,

Telling people to use an API which uses all cores to send data to the videocard instead of the current crop of crippled API which have severe limitation does not make sense ?

I'm one of the few persons on this webboard that is certainly not impressed by Intel. I'm laughing my ass of at Intel really all they can do is beat AMD , thats the equivalent of stealing lollipops from a baby. Then all of those android devices with ARM just runs away with a new market while Intel is to busy telling everyone how great they are with their better process for making cpu and did they already rub it in they are beating AMD ...
While having to stoop to these levels : https://semiaccurate.com/2014/12/29/intel-funnel-contra-revenue-funds/ .

And their 10nm process is going to do what exactly beat AMD again ?

 

[razor1]

That 50% was something the server guy heard (JF_AMD on this board), he is no longer with AMD, what he thought he understood what was something he didn't understand fully and drew that conclusion.
If Keller is such a bad person why did Apple hire him he did something right there? And beside that AMD wasted enough money and time on things that just did not work and it seems that was not the plan for Zen,

I don't think keller is a bad person, if you want to take what I stated as such, maybe you are already thinking he isn't good, in my view he is good, but is he going to "save" AMD is the issue. Intel all they did was make their Core M into a desktop part and iterations of it and it demolished AMD, since then their iterations didn't need to get too much better performance wise because AMD has been inconsequential in the CPU front, the only competition Intel has had was Intel itself.

Telling people to use an API which uses all cores to send data to the videocard instead of the current crop of crippled API which have severe limitation does not make sense ?

When your hardware is not competitive a band aid doesn't alleviate the actually problem, because that band aid actually helps the competition too.

I'm one of the few persons on this webboard that is certainly not impressed by Intel. I'm laughing my ass of at Intel really all they can do is beat AMD , thats the equivalent of stealing lollipops from a baby. Then all of those android devices with ARM just runs away with a new market while Intel is to busy telling everyone how great they are with their better process for making cpu and did they already rub it in they are beating AMD

I'm not inpressed with Intel's recent chips but and Nehalem and prior yeah you don't come out in one year and just shut out AMD for 13 years of being even remotely competitive unless you have a good product/s

While having to stoop to these levels : https://semiaccurate.com/2014/12/29/intel-funnel-contra-revenue-funds/ .

business is war, if you want an AMD processor for your system you can still get them at discounted prices, are those discounted prices worth it, nope, market says that.

And their 10nm process is going to do what exactly beat AMD again ?

The 10 nm process comes along with Intel's next gen CPU's after Skylake (2017), which is when Zen is coming out. It will include some very nice features for consumer brands like AVX 512 which AMD probably will not have, in essence it will create a eco system for compute for Intel CPU's without getting a knights landing (phi).

Intel also has much more control over their process and libraries vs AMD now because AMD is no longer in control over its fabrication process, this is another areas where AMD can't compete in and this areas is vast, everything from cost to customization of their silicon. Actually no one can compete with Intel when it comes to CPU fabrication, everyone is 2 +years behind. Intel's R&D into their fabs is unmatched by anyone. To ignore that is quite......

 

[razor1]

This will not happen since 10nm is delayed by at least 1 year maybe more. AMD will have 14nm next year. 10nm Intel is probably late 2017 to early 2018.

Zen is slated when? we don't have hard time lines for it, but 2nd half of 2016 sounds good right? 10nm is supposed to early 2017 not late. The delay was from 2016 to 2017, hence why Intel tick tock for Skylake was changed to just Tick and then the next processor is Tick again.

 

[SickBeast]

We have seen this from AMD before. They always hype up their products prior to launch and then they always fail. Bulldozer was the best example of this.

That being said, if AMD can pull off a CPU that's 40% faster than Excavator clock-for-clock, that would be great so long as the chip can hit 4ghz or so. I just don't see AMD pulling it off.

 

[NaroonGTX]

Bulldozer was supposed to have 50% increased performance due to increased IPC, its about the ability to traslate that directly to performance, not just what's there.

Nonsense. The only thing I ever saw them claim anywhere close to 50% was in regards to total throughput, and that was in multi-threaded code and obviously it came from the jump from 6 to 8 cores going from Stars to Bulldozer. Nowhere at any point was it ever said by AMD that there would be a 50% IPC increase.

Process wise AMD will always be behind Intel, by the time AMD comes out with 14nm, Intel will be on 10 nm......

Intel already had issues with their 14nm node, the 10nm one has already been pushed back and no one knows how well it will be by the time they're able to do volume production on chips with it.

I'm not inpressed with Intel's recent chips but and Nehalem and prior yeah you don't come out in one year and just shut out AMD for 13 years of being even remotely competitive unless you have a good product/s

Let's not get ahead of ourselves and pretend that AMD hasn't been competitive in over a decade. Phenom 1 wasn't that great but Phenom II was decently competitive and a popular line of chips. People seem pleased with their Visheras and APU's as well. I never saw a case where someone with an Intel chip got 60fps in a game and then the AMD could only manage 25.

Also Jim Keller is surely a beast, but he's not the only one who worked on the design. There were plenty of other talented veterans who worked with him, some people from the Bobcat team, some from the Jaguar team, some that worked on Bulldozer, and others. They were given the adequate time and resources to build a brand new uarch and do it the way they wanted to. That wasn't the case with Bulldozer where they had to do what the CEO and others told them to. It's also not 100% from scratch because as Keller said in May 2014, they re-used pre-existing IP as building blocks since there was no need to throw out anything that already worked well.

40% is pretty realistic and I'm not sure why anyone would believe different. I'm sure that the engineers, the people who actually worked on all these designs, would be well aware of what Bulldozer's shortcomings were and know things they'd want to implement in a new design.

http://edge.media-server.com/m/p/fc35hhui/lan/en
about 31 minutes in

As we stated in the financial analyst day we had a target of 40% IPC improvement of Zen over our previous generation. We believe we are on track for that. Relative to process technology we have taped out multiple products to multiple fabs in FinFet, we believe that they are also on track in terms of overall ramp

They sound pretty confident with both the uarch design and the process nodes. We'll see how it turns out but I don't feel any need to be cynical or skeptical about it for now.

Zen is slated when? we don't have hard time lines for it, but 2nd half of 2016 sounds good right? 10nm is supposed to early 2017 not late. The delay was from 2016 to 2017, hence why Intel tick tock for Skylake was changed to just Tick and then the next processor is Tick again.

We do. The Q3 '16~Q4 '16 time has been known for quite a while now.

 

[razor1]

Nonsense. The only thing I ever saw them claim anywhere close to 50% was in regards to total throughput, and that was in multi-threaded code and obviously it came from the jump from 6 to 8 cores going from Stars to Bulldozer. Nowhere at any point was it ever said by AMD that there would be a 50% IPC increase.

http://www.techpowerup.com/153548/b...4-is-this-really-enough-to-counter-intel.html

First graphic, IPC and power improvements, this is what was taken as IPC improvements by some of the spinsters.

Intel already had issues with their 14nm node, the 10nm one has already been pushed back and no one knows how well it will be by the time they're able to do volume production on chips with it.

early 2017 is still the target for them, in anycase, kaby lake is what Zen will be up against in the short term. If we read between the lines of what AMD is stating, they are using multiple fab partners for Zen, this means custom libraries will not be used? Since each fab will have their own libraries the only way around this is to have custom libs for each foundry, which will make validation a pain in the ass, this gives Intel a huge advantage even on 14nm. We don't even need to talk about 10nm.

Let's not get ahead of ourselves and pretend that AMD hasn't been competitive in over a decade. Phenom 1 wasn't that great but Phenom II was decently competitive and a popular line of chips. People seem pleased with their Visheras and APU's as well. I never saw a case where someone with an Intel chip got 60fps in a game and then the AMD could only manage 25.

Also Jim Keller is surely a beast, but he's not the only one who worked on the design. There were plenty of other talented veterans who worked with him, some people from the Bobcat team, some from the Jaguar team, some that worked on Bulldozer, and others. They were given the adequate time and resources to build a brand new uarch and do it the way they wanted to. That wasn't the case with Bulldozer where they had to do what the CEO and others told them to. It's also not 100% from scratch because as Keller said in May 2014, they re-used pre-existing IP as building blocks since there was no need to throw out anything that already worked well.

I agree

40% is pretty realistic and I'm not sure why anyone would believe different. I'm sure that the engineers, the people who actually worked on all these designs, would be well aware of what Bulldozer's shortcomings were and know things they'd want to implement in a new design.

http://edge.media-server.com/m/p/fc35hhui/lan/en
about 31 minutes in


They sound pretty confident with both the uarch design and the process nodes. We'll see how it turns out but I don't feel any need to be cynical or skeptical about it for now.

Is 40% increase in IPC alone enough? No its not there needs be quite a bit more to catch up with Intel. That's easy to see.

We do. The Q3 '16~Q4 '16 time has been known for quite a while now.

Not really, its an estimation, that's all we have.

 

[NaroonGTX]

First graphic, IPC and power improvements, this is what was taken as IPC improvements by some of the spinsters.

That's not a 50% improvement in one generation, but over several of them, and wasn't really outlandish. I mean it's not hard to improve on something which already has a busted foundation to begin with.

early 2017 is still the target for them, in anycase, kaby lake is what Zen will be up against in the short term. If we read between the lines of what AMD is stating, they are using multiple fab partners for Zen, this means custom libraries will not be used? Since each fab will have their own libraries the only way around this is to have custom libs for each foundry, which will make validation a pain in the ass, this gives Intel a huge advantage even on 14nm. We don't even need to talk about 10nm.

From what I currently understand, Kaby Lake will also be 14nm, so even if their 10nm node is ready for early 2017, we won't immediately see products on it. At least not on the consumer and enthusiast platforms.

Is 40% increase in IPC alone enough? No its not there needs be quite a bit more to catch up with Intel. That's easy to see.

A 40% increase is a good foundation and will be close enough to be an incentive for people to buy AMD chips again. It's not like they will stop at Zen, as Zen+'s design was finished as well and I guess would appear in late 2017 or early 2018. I'm not expecting Zen to launch out the gate and achieve parity with Skylake or Haswell, but who knows.

Not really, its an estimation, that's all we have.

AMD themselves have listed it as the projected date.

Summit Ridge = Zen CPU and FM3 has obviously been renamed to AM4 for marketing purposes.

 

[Darakian]

Is summit ridge the zen apu then?

 

[NaroonGTX]

Whoops, got the names mixed up. Summit Ridge = Zen CPU, Raven Ridge = Zen APU, Bristol = Excavator APU. Also some recent info suggests an Excavator APU is coming to FM2+, but I'll believe it when I see it.

 

[razor1]

That's not a 50% improvement in one generation, but over several of them, and wasn't really outlandish. I mean it's not hard to improve on something which already has a busted foundation to begin with.

Err they never reached the 50% even on a busted foundation.

From what I currently understand, Kaby Lake will also be 14nm, so even if their 10nm node is ready for early 2017, we won't immediately see products on it. At least not on the consumer and enthusiast platforms.

So 10nm is irrelevant if AMD isn't able to use custom libs on 14nm, as I stated above. Even then if you are taking estimates from AMD from their slides why not take Intel's 10nm of q1 2017 for 10nm CPU's then? Hmm.

A 40% increase is a good foundation and will be close enough to be an incentive for people to buy AMD chips again. It's not like they will stop at Zen, as Zen+'s design was finished as well and I guess would appear in late 2017 or early 2018. I'm not expecting Zen to launch out the gate and achieve parity with Skylake or Haswell, but who knows.

40% gets Zen to around Sandy bridge level, Sandy bridge to what Intel has now is another 30% or maybe around 40%, so no its not good enough. Of course 40% IPC with other things on top of that they might get up to 60% out of the gate its not unrealistic, which is where they need to be to some what compete with Intel's current offerings. By the time Zen+ comes out, they have to have a lot more, its not like Intel has just been sitting around and twirling their thumbs, it might look like that to us, but they will never let AMD have another netburst vs, AMD 64, from their end, any other words, they won't make another netburst type failure, AMD has to make Intel look bad, and not Intel falling on their face.

AMD themselves have listed it as the projected date.

Summit Ridge = Zen CPU and FM3 has obviously been renamed to AM4 for marketing purposes.

yeah its a projected date, same thing with Intel's too.

And if you want to go that way Summit Ridge is mid Q4 of 2016, which doesn't give Zen much time at all if Intel is planning for Q1 2017, and yeah they might do it if they feel Zen is getting to close. Its not like the pentium days where they didn't have the capability of doing a fast launch after releasing something because with their 10nm they would have already started validating their CPU's earlier it was planned for, which Intel didn't plan for the Athlon 64's which is what hurt them.

Looks like Zen's high end will be Q4 2016, and mainstream is Q1 2017, so yeah that gives time for Intel even with the 10nm delay (aslong as there is no more than q1 2017 delay) to respond if they feel any pressure.

 

[NaroonGTX]

Err they never reached the 50% even on a busted foundation.

I never said they reached it, I said it wasn't outlandish. They originally planned to extend the 15h family into 2018 with revisions, but then GloFo kept fucking up the nodes and AMD opted to can big-die versions of SR and XV successors to Piledriver server parts, which meant no more updates to the AM3+ platform as a result. Also the 10~15% performance increase wasn't about IPC, but perf/watt on that graph. Just shitty journalism as usual which tried to claim that AMD was saying they'd go get a 50% IPC increase over four iterations of the design.

So 10nm is irrelevant if AMD isn't able to use custom libs on 14nm, as I stated above. Even then if you are taking estimates from AMD from their slides why not take Intel's 10nm of q1 2017 for 10nm CPU's then? Hmm.

What 10nm CPU's do you honestly expect to see so quickly? It's not like they'd be anything amazing anyway, regardless of who would make something that would get fabbed on said process. Intel hasn't done anything compelling in terms of CPU's in ages now.

40% gets Zen to around Sandy bridge level, Sandy bridge to what Intel has now is another 30% or maybe around 40%, so no its not good enough. Of course 40% IPC with other things on top of that they might get up to 60% out of the gate its not unrealistic, which is where they need to be to some what compete with Intel's current offerings. By the time Zen+ comes out, they have to have a lot more, its not like Intel has just been sitting around and twirling their thumbs, it might look like that to us, but they will never let AMD have another netburst vs, AMD 64, from their end, any other words, they won't make another netburst type failure, AMD has to make Intel look bad, and not Intel falling on their face.

Do you know how Excavator performs? Do you have data that no one else does? XV is decently faster than SR and 40% over that will be about Sandy/Ivy levels perhaps, which is good enough for the vast majority of people except those who jack off to synthetic benches all day.

As for Intel "sitting around", no one said they would be. These uarchs take many years to go from design phase to volume productions, it's not like they could do anything to retaliate against AMD if Zen is a winner besides doing a marketing push and releasing Emergency Edition variants of pre-existing chips.

As for A64 vs Netburst, let's just not even get into this sensationalist garbage already. Anyone with realistic expectations is not expecting Zen to shit all over Intel's current crop. That's absurd and it's not even necessary. AMD just needs a design that is energy efficient and with high enough IPC that anyone who isn't an ardent fanboy would consider buying it.

And if you want to go that way Summit Ridge is mid Q4 of 2016, which doesn't give Zen much time at all if Intel is planning for Q1 2017, and yeah they might do it if they feel Zen is getting to close. Its not like the pentium days where they didn't have the capability of doing a fast launch after releasing something because with their 10nm they would have already started validating their CPU's earlier it was planned for, which Intel didn't plan for the Athlon 64's which is what hurt them.

It didn't matter if Intel had planned for A64, they would've gotten demolished no matter what because A64 was just flat-out superior to P4 in every way imaginable. I don't know what mythical 10nm halo product you think Intel is secretly holding out specifically for retaliation against Zen is supposed to come from, but the doomsday assertion is laughable. I don't think anyone will be reeling from an amazing 5% IPC increase over their pre-existing Skylake line.

 

[razor1]

What 10nm CPU's do you honestly expect to see so quickly? It's not like they'd be anything amazing anyway, regardless of who would make something that would get fabbed on said process. Intel hasn't done anything compelling in terms of CPU's in ages now.

Intel has more than one design team, actually has more than 4 design teams, all of which are working on varying products on different next gen CPU's. How many design teams does AMD have? I can give you a fairly quick answer to this .

Intel didn't need to push the envelope because there was no competition but their own. Why would they push anything outside of just enough to get people to upgrade?

Do you know how Excavator performs? Do you have data that no one else does? XV is decently faster than SR and 40% over that will be about Sandy/Ivy levels perhaps, which is good enough for the vast majority of people except those who jack off to synthetic benches all day.

"good enough" what is that, marketing terms? What can AMD use to market against Intel, if its just good enough, that's not enough. When people look at all those marketing performance charts from both venders, good enough isn't whats on their mind.

Just an example, for good enough was my penryn dual xeon system with 16gb ram, until recently 6 months back I was still using it and didn't need to upgrade, the only reason was because graphics cards didn't support the old bios anymore and then Windows 10 had some issues with legacy hardware, forced me to upgrade. Guess what most people will not keep a system for 5 years and still think yeah I still have 2 more years of CPU life on this thing before I need to upgrade.

As for Intel "sitting around", no one said they would be. These uarchs take many years to go from design phase to volume productions, it's not like they could do anything to retaliate against AMD if Zen is a winner besides doing a marketing push and releasing Emergency Edition variants of pre-existing chips.

There is no such thing as an emergency edition, but they can fast track a project, if they were thinking of getting a CPU that was on 10nm out in 2016, but was delayed because of the process and quickly substituted it with a 14nm variant, that 14nm variant was already in the works. It was fast tracked to get into mass production, just like that, all Intel needs is to make sure their 10nm process is ready for mass production, their CPU is already ready for production and validation.

As for A64 vs Netburst, let's just not even get into this sensationalist garbage already. Anyone with realistic expectations is not expecting Zen to shit all over Intel's current crop. That's absurd and it's not even necessary. AMD just needs a design that is energy efficient and with high enough IPC that anyone who isn't an ardent fanboy would consider buying it.

Zen will have to compete with Intel, there is no way around it, if it doesn't AMD dies, they have too much at stake to be behind Intel, in performance and performance per watt. They have to reach around 10% of what Intel has right now. I don't expect Zen to beat Intel, but if they don't get close to Intel's current crop, they are died, they don't have the capital to keep on going for another 2 years after Zen's release if Zen isn't competitive, that is a major problem for AMD right now. The only way around this is if they can regain marketshare in the GPU market to even out the losses they are getting from the CPU side and right now that will be tricky too since they lost so much.

It didn't matter if Intel had planned for A64, they would've gotten demolished no matter what because A64 was just flat-out superior to P4 in every way imaginable. I don't know what mythical 10nm halo product you think Intel is secretly holding out specifically for retaliation against Zen is supposed to come from, but the doomsday assertion is laughable. I don't think anyone will be reeling from an amazing 5% IPC increase over their pre-existing Skylake line.

Pentium M was on par A64 or a little bit better, and it was out already, but it was too late because they had no platform for Pentium M to be used as a desktop.

Intel isn't just about consumer end either, Cannonlake is going to be a boon for corporations that want to do compute programming on Intel platforms without buying Knights landing, and this is something that will actually come with Xeon skylake first, before it goes into mainstream CPU's.

 

[NKD]

Intel has more than one design team, actually has more than 4 design teams, all of which are working on varying products on different next gen CPU's. How many design teams does AMD have? I can give you a fairly quick answer to this .

Intel didn't need to push the envelope because there was no competition but their own. Why would they push anything outside of just enough to get people to upgrade?

"good enough" what is that, marketing terms? What can AMD use to market against Intel, if its just good enough, that's not enough. When people look at all those marketing performance charts from both venders, good enough isn't whats on their mind.

Just an example, for good enough was my penryn dual xeon system with 16gb ram, until recently 6 months back I was still using it and didn't need to upgrade, the only reason was because graphics cards didn't support the old bios anymore and then Windows 10 had some issues with legacy hardware, forced me to upgrade. Guess what most people will not keep a system for 5 years and still think yeah I still have 2 more years of CPU life on this thing before I need to upgrade.

There is no such thing as an emergency edition, but they can fast track a project, if they were thinking of getting a CPU that was on 10nm out in 2016, but was delayed because of the process and quickly substituted it with a 14nm variant, that 14nm variant was already in the works. It was fast tracked to get into mass production, just like that, all Intel needs is to make sure their 10nm process is ready for mass production, their CPU is already ready for production and validation.

Zen will have to compete with Intel, there is no way around it, if it doesn't AMD dies, they have too much at stake to be behind Intel, in performance and performance per watt. They have to reach around 10% of what Intel has right now. I don't expect Zen to beat Intel, but if they don't get close to Intel's current crop, they are died, they don't have the capital to keep on going for another 2 years after Zen's release if Zen isn't competitive, that is a major problem for AMD right now. The only way around this is if they can regain marketshare in the GPU market to even out the losses they are getting from the CPU side and right now that will be tricky too since they lost so much.


Pentium M was on par A64 or a little bit better, and it was out already, but it was too late because they had no platform for Pentium M to be used as a desktop.

Intel isn't just about consumer end either, Cannonlake is going to be a boon for corporations that want to do compute programming on Intel platforms without buying Knights landing, and this is something that will actually come with Xeon skylake first, before it goes into mainstream CPU's.

I dont even have amd and havn't had it for 3 years now. But you are really talking like a pure fanboy and most of what you are saying has no value in real world. If zen is anywhere even close to sandy bridge that is more than enough performance and I am sure it will be lil faster. That will be most competitive product AMD has launched in years. Intel hasn't been doing shit, they have been sitting on their asses with nothing major or mind blowing and may be zen will wake them up. AMD needs a product that is competitive they don't need to blow intel out of water. Once they have that they can focus on further enhancement with zen 2 and beyond. Thats all they need. If they can deliver that they are back in the game. All this do or die scenario if they don't deliver a knock out punch is non sense. They need a processor that brings them back in the game at a price that is best bang for the buck. I am still rocking an i5 3570k and I haven't seen the need to upgrade for a few years now. If Zen is priced right I might give it a shot and build a rig with it just to support a good product from AMD that I haven't seen for a long time.

 

[XoR]

If todays 4 core APU had 40% better IPC than it is now, even considering worse muticore scaling due to CMT it would easily beat Core i3 in everything and rival and and beat (especially after OC) lower end i5's. That would be enough to make those chips sell well.

Improvement in performance between Intel generations in recent years is rather pathetic so the same thing would still stand in 2017 or even 2018. It doesn't really matter when 10nm Intel CPU are coming out. They most probably won't be that much better for it to really matter.

 

[Pieter3dnow]

When your hardware is not competitive a band aid doesn't alleviate the actually problem, because that band aid actually helps the competition too.

I'm not inpressed with Intel's recent chips but and Nehalem and prior yeah you don't come out in one year and just shut out AMD for 13 years of being even remotely competitive unless you have a good product/s

business is war, if you want an AMD processor for your system you can still get them at discounted prices, are those discounted prices worth it, nope, market says that.

The 10 nm process comes along with Intel's next gen CPU's after Skylake (2017), which is when Zen is coming out. It will include some very nice features for consumer brands like AVX 512 which AMD probably will not have, in essence it will create a eco system for compute for Intel CPU's without getting a knights landing (phi).

Intel also has much more control over their process and libraries vs AMD now because AMD is no longer in control over its fabrication process, this is another areas where AMD can't compete in and this areas is vast, everything from cost to customization of their silicon. Actually no one can compete with Intel when it comes to CPU fabrication, everyone is 2 +years behind. Intel's R&D into their fabs is unmatched by anyone. To ignore that is quite......

The cpu is only to push 10+ teraflops of gpu power to make it anything more then that is a mistake, to call more cores a band aid is a fundamental lack of understanding what drives the gaming business. When progress has been stagnant for all these years and there is no other way to push graphics besides having more cores how can this be even considered a band aid. There is no other way to keep pushing then cores, ipc stopped graphics from evolving...

When you are fighting someone with their hands tied behind their back while someone is holding them down at the same time is not really a war or fighting...

Maybe you did not understand what I was telling you in the first place if you compare Intel to AMD then you are out of your mind. Intel will never purposely kill of AMD because then Intel would become a monopoly. Intel R&D is so superior yet they can't compete on the android market given that they rolled over Intel with ARM.

 

[ebduncan]

I swear people act like Intel can pull tricks out of its hat if AMD releases a competitive product. NO, if Zen launches and beats Intel in performance then Intel will not magically have a new chip on the market the next day. Cpu's are designed over a time frame of years, so they take time to get to market. Intel isn't sitting on any advancements they can just throw into new cpus. They are on their planned schedule for release.

as far as Zen performance is concerned. Zen will destroy Sandy Bridge and Ivy Bridge in terms of performance. Not enough is known about the clock speeds the node can bring into fruition. I know one thing wherever Jim Keller goes they have class leading products, so I'm sure AMD will bring something decent to the table this go around.

 

[razor1]

I swear people act like Intel can pull tricks out of its hat if AMD releases a competitive product. NO, if Zen launches and beats Intel in performance then Intel will not magically have a new chip on the market the next day. Cpu's are designed over a time frame of years, so they take time to get to market. Intel isn't sitting on any advancements they can just throw into new cpus. They are on their planned schedule for release.

as far as Zen performance is concerned. Zen will destroy Sandy Bridge and Ivy Bridge in terms of performance. Not enough is known about the clock speeds the node can bring into fruition. I know one thing wherever Jim Keller goes they have class leading products, so I'm sure AMD will bring something decent to the table this go around.

Please the apple cpu's suck (this really had nothing to do with Keller btw, it was Apple's dictating to Keller what they needed, Keller is good, but I don't think he is as good as Intel's top multiple minds. Core M told us that. Intel has the resources to have multi teams, many more than AMD, AMD next gen CPU dev teams (actually comprised of two teams one for the GPU and one for CPU), only 1, Intel has 4 that I know of, and there are 2 other teams that are working on next portions of CPU's but not the core of the CPU itself, guess what, that is many back up plans for Intel, and none for AMD. This is why we were able to see, Intel change direction with Broad well, and bring it out early, as with Kaby lake, they have multiple products that can be fast tracked if they feel any pressure. When some one has more resources, they can do this. AMD didn't do this when they had the Athlon 64, they should have, they didn't have the foresight. This is why when they make a CPU they are stuck for 5 years with what they have. If Intel makes a mistake they have plans they can put into motion that can alleviate the issue rather quickly within one year, and that's what they did with Core M. They used that as their foundation for Core 2 duo and iterated on that when Pentium 4 wasn't enough.

Btw prior to Core line of CPU's from Intel, I only bought AMD and before that NexGen (this is where Keller and many of the guys that made the Athlon and A64's came from) which was bought out by AMD, ever since I had a 486 till Core line, I had no Intel's in my systems.

The cpu is only to push 10+ teraflops of gpu power to make it anything more then that is a mistake, to call more cores a band aid is a fundamental lack of understanding what drives the gaming business. When progress has been stagnant for all these years and there is no other way to push graphics besides having more cores how can this be even considered a band aid. There is no other way to keep pushing then cores, ipc stopped graphics from evolving...

When you are fighting someone with their hands tied behind their back while someone is holding them down at the same time is not really a war or fighting...

Maybe you did not understand what I was telling you in the first place if you compare Intel to AMD then you are out of your mind. Intel will never purposely kill of AMD because then Intel would become a monopoly. Intel R&D is so superior yet they can't compete on the android market given that they rolled over Intel with ARM.

its a band aid to the actual problem, its like cancer, you want to treat the symptom or do you want cure it? Its a lack of understanding or blind faith when you don't want to have AMD to compete with Intel on a per processor level, but say hell use this API, it will help your noncompetitive CPU, guess what though, the other side of the coin it help Intel's CPU too! So its good for everyone not just AMD's CPU, OK?

When you are fighting someone with their hands tied behind their back while someone is holding them down at the same time is not really a war or fighting...

Sun Tzu, know your enemy, know thy self, you win the battle.

Doesn't matter if someone is crippled, if go in to win, and that is it. If some one that is crippled comes to your door and says tomorrow I'm taking your job, what will you do? Oh its ok, he is a cripple, give it to him? Yeah I see how far you will get.

Maybe you did not understand what I was telling you in the first place if you compare Intel to AMD then you are out of your mind. Intel will never purposely kill of AMD because then Intel would become a monopoly. Intel R&D is so superior yet they can't compete on the android market given that they rolled over Intel with ARM.

Intel can kill AMD when it comes straight to market pressure, there is nothing wrong with that, if AMD isn't competitive, AMD is killing itself, Intel has to do nothing.

WTF are you talking about, I think you are confused with Intel's wishes for low power devices, their CPU's are gettting there, but its a different design philosophy to create ultra low power chips. Intel's mind set wasn't there until recently. ARM technology, that was there mind set from the beginning so yeah ARM has this advantage, but Intel is catching up.

 

[razor1]

I dont even have amd and havn't had it for 3 years now. But you are really talking like a pure fanboy and most of what you are saying has no value in real world. If zen is anywhere even close to sandy bridge that is more than enough performance and I am sure it will be lil faster. That will be most competitive product AMD has launched in years. Intel hasn't been doing shit, they have been sitting on their asses with nothing major or mind blowing and may be zen will wake them up. AMD needs a product that is competitive they don't need to blow intel out of water. Once they have that they can focus on further enhancement with zen 2 and beyond. Thats all they need. If they can deliver that they are back in the game. All this do or die scenario if they don't deliver a knock out punch is non sense. They need a processor that brings them back in the game at a price that is best bang for the buck. I am still rocking an i5 3570k and I haven't seen the need to upgrade for a few years now. If Zen is priced right I might give it a shot and build a rig with it just to support a good product from AMD that I haven't seen for a long time.

AMD is losing money right now, not just because of lower marketshare, they have to sell bigger chips at lower prices. They are killing their margins. So no, this is why they need to be competitive on all fronts with Intel. Because Intel is dictating the pricing, not AMD. Being a fanboy is understanding how to be competitive in a marketplace is what? With AMD's marketshare loss they can't sell chips at lower margins, specially taking an 8 core chip to go against a mid range 4 core Intel. Die size you are looking at a 300mm chip vs a 180mm chip but selling the 300mm chip at a lower price, without enough volume sales AMD can't sustain such a situation for long, if this was 10 years ago, and they had 1.5 billion cash or cash equivalents they could, they don't have that anymore.

They need to be competitive, they can't give Intel more than 10% performance or power usage, because then they will get back in selling more expensive chips for less.

 

[Zarathustra[H]]

Is 40% increase in IPC alone enough? No its not there needs be quite a bit more to catch up with Intel. That's easy to see.

If we take what has been stated to date as fact (40% IPC bump over Carizzo, not a high clock design, targeting ~ 4Ghz max) we wind up with a CPU that per core performs faster than a high end Ivy Bridge, but slower than a high end Haswell.

If this is true, it will certainly close the gap a lot, but itbwobt have caught up with Intel.

I also don't think it has to. Get it close enough and people will buy it.

 

[variant]

If we take what has been stated to date as fact (40% IPC bump over Carizzo, not a high clock design, targeting ~ 4Ghz max) we wind up with a CPU that per core performs faster than a high end Ivy Bridge, but slower than a high end Haswell.

If this is true, it will certainly close the gap a lot, but itbwobt have caught up with Intel.

I also don't think it has to. Get it close enough and people will buy it.

The 40% number being thrown around is independent of the process being used. Even if the the Zen was manufactured at the 28nm of the Carrizo, it would still have a 40% IPC increase. There will still be benefits from it being manufactured at 14nm or 16nm. From what I also understand, the concept that it will be on par with Haswell is not based off the IPC of Excavator, instead is based off the IPC of Steamroller. Excavator is suppose to have an IPC of 4-15% higher than Steamroller.

 

[razor1]

If we take what has been stated to date as fact (40% IPC bump over Carizzo, not a high clock design, targeting ~ 4Ghz max) we wind up with a CPU that per core performs faster than a high end Ivy Bridge, but slower than a high end Haswell.

If this is true, it will certainly close the gap a lot, but itbwobt have caught up with Intel.

I also don't think it has to. Get it close enough and people will buy it.

yeah it has to be close enough to Intel in both performance and performance per watt, I would say around that 10%, and then prices to match, 15% less within its comparative brackets, this will solve two of the three problems AMD has in the CPU market by increasing margins for all their CPU and provide a healthy bottom line for AMD, and this will stimulate the increase in marketshare (this is will take many quarters possible even years to get back to where they were, but its a start), which is AMD's last issue on the CPU side of things.

The 40% number being thrown around is independent of the process being used. Even if the the Zen was manufactured at the 28nm of the Carrizo, it would still have a 40% IPC increase. There will still be benefits from it being manufactured at 14nm or 16nm. From what I also understand, the concept that it will be on par with Haswell is not based off the IPC of Excavator, instead is based off the IPC of Steamroller. Excavator is suppose to have an IPC of 4-15% higher than Steamroller.

No the process has a lot to do with it too, making chips today, is always done by looking at the process and what it can offer to what new architectural changes can give in performance and power consumption. Although design is more important in chip design now then before when power and performance is concerned, and this is why Moore's law started failing, process itself doesn't drive performance gains but it still has a great deal to do with what architectural changes can be done and how.

 

[variant]

No the process has a lot to do with it too, making chips today, is always done by looking at the process and what it can offer to what new architectural changes can give in performance and power consumption. Although design is more important in chip design now then before when power and performance is concerned, and this is why Moore's law started failing, process itself doesn't drive performance gains but it still has a great deal to do with what architectural changes can be done and how.

The process has a lot to do with the performance of the actual chip, but the 40% improvement for the Zen architecture is regardless of the manufacturing process. It's purely architecture design improvement. The improvement they will get from the process is in addition to the 40%.

 

[razor1]

The process has a lot to do with the performance of the actual chip which it's produced, but the actual 40% improvement number is regardless of the manufacturing process. It's purely architecture design improvement. The improvement they will get from the process is in addition to the 40%.

I think its talking about both, because you may have to use more transistors to get that increased IPC.

 

[variant]

I think its talking about both, because you may have to use more transistors to get that increased IPC.

You can increase the IPC by adding more transistors as the silicon shrinks, but that isn't what the 40% improvement figure is based around. The 40% number is solely design improvements independent of the process. If Zen was a 28nm CPU like the Carrizo, it would have a 40% IPC improvement.

The Excavator alone is getting a 4-15% IPC improvement over Steamroller without any reduction.