AMD Zen Performance Preview

Zarathustra[H] said:
It means they finally caught up to Sandy bridge IF and ONLY IF they can make it clock near 4Ghz.

All rumors about GloFo's process this will be manufactured on - however - suggest that it will be tough to get clocks much above 3Ghz on them.

I was hoping to see some hint about potential clock speeds in this slide deck to dispell those fears, but I saw nothing, and the continued harping on IPC without mentioning total performance or clock speeds makes me suggest that it is true.

In other words, IPC will me massively up from Piledriver FX chips, but clocks will be massively down, resulting in an overall 10% performance increase, when they really need a 60% overall performance increase.

Expect to see AMD trying to throw more cores at the problem again, just like they did with bulldozer, when we all know that with the exception of some narrow usage cases, what really matters is single threaded performance. :(

I had high hopes for Zen, but it's starting to look like its going to be another Bulldozer sized disappointment.
Click to expand...


I donno if it has 8/10 cores or something.. 3.0 ghz might be about right.

Unless you were one of those that complained about this 3.0 Ghz also?
***** Intel Core i7-5960X Haswell-E 8-Core 3.0 GHz LGA 2 011-v3 140W BX80648I75960X Desktop Processor
 
ThatsAgood1jay said:
Excited to read HardOCP's review of the chips. Until then ill be putting money away to build a new system.

It's all vapor until the chip is running on the bench in front of you.
Click to expand...

Agreed.

My i7-3930k is getting a bit long in the tooth. Still bloody fast at 4.8Ghz, but it is also HOT HOT HOT, and the post-lead plastic on the motherboard is degrading fast. Just the other day I broke off two RAM clips by doing barely more than looking at them, while trying to reset my RAM (luckily the ram seems to stay in place even without them, based on friction alone).

It's time for an upgrade. Some more modern features wouldn't hurt either.

Unless my system shits the bed before then, I'll be waiting until after Zen launches and we have real reviews to make any decision. I had hoped to buy AMD again for the first time in a while, but it's not looking promising to me at this point. Still, its all conjecture until we have real benchmarks on shipping silicon.
 
Axehandler said:
I donno if it has 8/10 cores or something.. 3.0 ghz might be about right.

Unless you were one of those that complained about this 3.0 Ghz also?
***** Intel Core i7-5960X Haswell-E 8-Core 3.0 GHz LGA 2 011-v3 140W BX80648I75960X Desktop Processor
Click to expand...

Firstly, this assumes you believe the Zen beating or even tying Broadwell-E benchmark, and it isn't just a benchmark anomaly, which I don't.

Secondly, I didn't buy a 5960X, and if I were shopping today, I wouldn't.

I feel like 6C/12T is the sweet spot, where - with good cooling - I can still hit the same or higher clocks as the equivalent process 4C/8T parts.

I really have no need for more cores, and won't for some time unless something major changes in software coding practices or compiler technology.

Don't get me wrong, I'll take more cores if all else is equal, but if it involves sacrificing single threaded performance (which on the same architecture essentially just comes down to clock speed), then I'll say thanks, but no thanks.

I would have absolutely no interest what so ever in a 5960X, unless I was guaranteed a golden sample that could overclock as high as a good 4C/8T Haswell chip.
 
For me personally, I learned my lesson with AMD for once and all with the 290 early on and the same driver issues year after year with that card and past video cards. From the outside ( I am no longer an AMD customer nor ever again expect to be ) it does seem AMD are issuing a lot more driver updates which is a good thing.

Here is what I wish most people would do. Understand that AMD will never again be a performance part market leader. That chance has come and gone and been gone for years. But, that's ok. 2nd place is still a good spot to be in. I just wish people would think about and manage their expectations better when it comes to AMD. There are many lessons from the past. Anyone here who thinks AMD will be beat Intel or Nvidia is just simply not paying attention. From the surface, slides, whatever the case may be, it looks all rosy but .... no .... not really.

below to average performance at best is what you should count on, no matter how many slides and benchmarks AMD puts in front of you.

And if they ever do deliver on a killer performance part. I will be first in line to congratulate them. But that's a huge fcking "IF".
 
I didn't even look at the slide deck. They're actually more useful when you print them out and make paper airplanes.

Until the processor hits the pavement, it means squat.
 
I am really hoping Zen outperforms expectations just to give AthlonXP a reason to change his name!! The guy has been suffering long enough!!

cpusreihe.jpg
 
I've always found graphs with no scale based on "internal estimates" to be useful :rolleyes:.

Too much marketing AMD... too much marketing.
 
maxius said:
Try Broadwell-E troll
Click to expand...

No, he's right.

I don't for a second believe that test is anything but a single benchmark anomaly.

If the 40% IPC over Kaveri statement holds up, AND they can hit 4ghz thr best they are going to do is tie the single threaded performance of a Sandy Bridge core i5-2500k which will be six years old when Zen launches in volume.

What's worse is that everything suggests that hitting much above 3ghz is going to be difficult on GloFo's process.

This means while we might see 40% IPC improvement, it will be accompanied by massive reductions in clock speed, and actual single threaded performance gains over Kaveri are more like 10%, which doesn't bring them anywhere near Sandy Bridge :(
 
Axehandler said:
I donno if it has 8/10 cores or something.. 3.0 ghz might be about right.

Unless you were one of those that complained about this 3.0 Ghz also?
***** Intel Core i7-5960X Haswell-E 8-Core 3.0 GHz LGA 2 011-v3 140W BX80648I75960X Desktop Processor
Click to expand...

Yeah, I have one of those 3.0GHz chips, been running at 4.6GHz for over a year. If Zen can pull off a 50%+ OC, then it'll be a real contender.
 
I just hope that the CPU is competitive with Intel's latest, whatever you think of AMD or Intel, we all need competition to drive performance and value for money, so I personally hope that AMD have a winner here.
 
Chebsy said:
I just hope that the CPU is competitive with Intel's latest, whatever you think of AMD or Intel, we all need competition to drive performance and value for money, so I personally hope that AMD have a winner here.
Click to expand...

I try to be as optimistic as I can. I would LOVE a repeat of the price and performance war we had from ~1999-2004. Those were some of the best years to be a PC hardware geek, IMHO. It would also fill me with warm fuzzies to have an AMD CPU in my main rig again.

It's just not looking good to me. I'm trying to remain hopeful, but realism is starting to set in.

If dreams were horses, beggars would ride...
 
Zarathustra[H] said:
No, he's right.

I don't for a second believe that test is anything but a single benchmark anomaly.

If the 40% IPC over Kaveri statement holds up, AND they can hit 4ghz thr best they are going to do is tie the single threaded performance of a Sandy Bridge core i5-2500k which will be six years old when Zen launches in volume.

What's worse is that everything suggests that hitting much above 3ghz is going to be difficult on GloFo's process.

This means while we might see 40% IPC improvement, it will be accompanied by massive reductions in clock speed, and actual single threaded performance gains over Kaveri are more like 10%, which doesn't bring them anywhere near Sandy Bridge :(
Click to expand...

Just as performance cannot be derived from one benchmark nor does a processor perform similarly in all workloads. AMD's BD arch is already at or above 2500K if we are talking about encoding or well threaded streaming workloads.
 
kac77 said:
Just as performance cannot be derived from one benchmark nor does a processor perform similarly in all workloads. AMD's BD arch is already at or above 2500K if we are talking about encoding or well threaded streaming workloads.
Click to expand...

Yeah, but that's like 0.5% of typical workloads. The most important CPU benchmark of all is single threaded performance, as it speaks to how good the architecture is.

Adding more cores to make up for poor single threaded performance is just a last ditch effort that results in lesser performance in the overwhelming majority of tasks.

It is not a matter of "if they only could improve threading in code". Not all code is thread able, no matter how hard you try. Pretty much with the launch of DX12and the state of things, code is about as threaded today as it will ever get. In all but a very few workloads, having fewer zippier cores is vastly superior to more, slower ones.

Why would you voluntarily buy a CPU that excels only at a tiny minority of tasks, I stead of one that performs well at everything?
 
Zarathustra[H] said:
Yeah, but that's like 0.5% of typical workloads. The most important CPU benchmark of all is single threaded performance, as it speaks to how good the architecture is.

Adding more cores to make up for poor single threaded performance is just a last ditch effort that results in lesser performance in the overwhelming majority of tasks.
Click to expand...
I agree but that's not what we are talking about here. Going off the benchmark both processors were at the same clock and the same number of cores. If IPC was in the tank then the comparison wouldn't even come close. Even if you compiled a custom code path for the processor.

Zarathustra[H] said:
It is not a matter of "if they only could improve threading in code". Not all code is thread able, no matter how hard you try. Pretty much with the launch of DX12and the state of things, code is about as threaded today as it will ever get. In all but a very few workloads, having fewer zippier cores is vastly superior to more, slower ones.
Click to expand...
Not arguing that. Just saying that making a blanket statement that AMD needs a 40% boost across the board isn't entirely accurate. They need a minimum of a 40% boost at the same clockspeed whenever single core performance is more desirable.


Zarathustra[H] said:
Why would you voluntarily buy a CPU that excels only at a tiny minority of tasks, I stead of one that performs well at everything?
Click to expand...

Well general performance has been good enough for at least 5 - 8 years. No one who has even a ULV processor like an Atom will have problems doing basic tasks. So we are already well beyond your scenario. Everything now is niche markets. HEDT, ULV, etc etc. We are well beyond needing general performance increases for the common person. Now efficiency is another matter. But overall if you are buying that 8-core processor you better have some special reason to have it other than bragging rights because Word or 95% of games don't need the extra cores.
 
Last edited:
sirmonkey1985 said:
true, but even if it gets within 10% of sandy bridge at least it's competition so i'd be happy. we'll have to wait and see though.
Click to expand...

I really hope it is 10% within Haswell rather than Sandy Bridge, if it is within Sandy Bridge, then they show up to a party that everyone left and just eating the scraps. But if performance is within Haswell, they might be late to the party but still can get the booze and score the chick!
 
Zarathustra[H] said:
No, he's right.
I don't for a second believe that test is anything but a single benchmark anomaly.
If the 40% IPC over Kaveri statement holds up, AND they can hit 4ghz thr best they are going to do is tie the single threaded performance of a Sandy Bridge core i5-2500k which will be six years old when Zen launches in volume.
What's worse is that everything suggests that hitting much above 3ghz is going to be difficult on GloFo's process.
This means while we might see 40% IPC improvement, it will be accompanied by massive reductions in clock speed, and actual single threaded performance gains over Kaveri are more like 10%, which doesn't bring them anywhere near Sandy Bridge :(
Click to expand...

The clock speed is the key , if anything it is important to see some other benchmarks from a retail Zen. But we are still months away from launch so were not gong to see anything.
 
freeloader1969 said:
If any of you think Zen will only reach SB IPC levels, then you've lost all your marbles.
Click to expand...


Other way around. If you think it will beat SB in IPC you have lost your marbles. Kaveri is more than 40% behind SB in IPC. So, adding 40% IPC doesn't bring you to SB levels. It's that simple...

Sandy Bridge has approximately 54% higher IPC than Kaveri.

So, if you add 40% to Kaveri (like AMD claims) you are still 10% or so behind SB.

And AMD's claims tend to be exaggerated anyway, so the real IPC gain is probably less than 40%...

The blender render test AMD picked was obviously picked because it skews heavily in AMD's favor. If Zen - on average - matches or exceeds Broadwell in IPC I will eat my hat.
 
Zarathustra[H] said:
Other way around. If you think it will beat SB in IPC you have lost your marbles. Kaveri is more than 40% behind SB in IPC. So, adding 40% IPC doesn't bring you to SB levels. It's that simple...

Sandy Bridge has approximately 54% higher IPC than Kaveri.

So, if you add 40% to Kaveri (like AMD claims) you are still 10% or so behind SB.

And AMD's claims tend to be exaggerated anyway, so the real IPC gain is probably less than 40%...

The blender render test AMD picked was obviously picked because it skews heavily in AMD's favor. If Zen - on average - matches or exceeds Broadwell in IPC I will eat my hat.
Click to expand...
And here I thought it was Haswell/Ivy Bridge they were shooting for which would only put them 2-3 generations behind Intel, depending on if you are comparing to Broadwell or Skylake.

An 8 core cpu in the 3.5ghz range with every core as fast clock per clock as a Haswell but selling cheaper than Intel's current 8 core CPU would have been a Grand Slam for AMD even if it didn't clock too much higher.
 
I think the real problem with zen is going to be glofo. If you look at recent things they've made, they don't clock well. The rx480 has a clock that I have a feeling amd wished was higher. And another great example is apples a9 (6s Soc). The ones on tsmc run cooler and use less power than the Samsung/glofo socs.
 
ManofGod said:
Wow, did AMD kick your dog and kill your cat? :D Did you lose a bunch of money on their stocks or do you just feel animosity towards them just because? :D
Click to expand...
lol.

The question I have is why aren't more people up in arms about this? If Intel or nVidia's marketing programs were as full of blatant lies, obfuscation and deception as AMDs have been for the past several years you would hear the outcry from the South Pole. But AMD somehow gets a pass? Not from me they don't.
 
I don't get all the butt hurt for an unreleased product. like somehow AMD forces people to purchase.
We know what it will be. AMD isn't going for the top %10 market, it is going for the %90 market. It will push the $ for 16 thread CPU's way the hell down, which will be great for enthusiasts no matter who you buy from.
 
Cobalt35 said:
lol.

The question I have is why aren't more people up in arms about this? If Intel or nVidia's marketing programs were as full of blatant lies, obfuscation and deception as AMDs have been for the past several years you would hear the outcry from the South Pole. But AMD somehow gets a pass? Not from me they don't.
Click to expand...

you've been here since 2008? what forum sections are you visiting? lol.. AMD's been on the shit list for the last 5-6 years, there's literally nothing they can do right or wrong where they don't get bashed into the ground by people here.
 
wait is everyone expecting 8 core zen and 4 core zen to be clocked at 3ghz? If there are 4 core zen chips I expect them to be running at 4.0 or so. Because no way we have 8 core chips and 4 core chips running at 3ghz. I understand GF might not be able to do too much about 3ghz on first revision but to expect 4 core chip stuck at 3.0 is little too low. I expect 4 core chips to be higher clocked than 8 core as it is always the case.
 
As usual here. The calm minds wait out the storm then discuss.
Kinda sad really. [H] = hyper sensitive fan boi? I really hope not.
But when %95 of posts are based on imaginary projections, and that is considered standard forum etiquette where bizarre statements are let to slide. I have to ask. WTF boys?
Are ya all sideways on stupidity? The market will show what is the best deal.
Are any of you really believing Zen will be a poorly priced item? If you do, your honestly taking the piss!
AMD will never have the crown again. AMD is not aiming at having the crown again. Understand???
The single biggest concern of AMD is that Intel will drop its hidden next gen at half/cost. And they can do it.
If it wasn't for share holders and anti monopoly practice's, Intel would have killed AMD long ago. AND WOULD!!
But why should they? Our evidence is that when there is no market pressure Intel sits on its ass and does marketing slide advancements. Mostly focused on non market pressured categories such as power draw.
And they sell, sell, sell.
It's good to be the King ;)
 
All bark, no bite. I'll believe it when they hand out samples, and people start running them vs Intel platforms with nearly the same specs.

Till then... it's all Marketing PR in hopes of making there share holders wet themselves.

Reminds me of the slides of "Piledriver" when it then completely failed to get past an i3.
 
Last edited:
NKD said:
wait is everyone expecting 8 core zen and 4 core zen to be clocked at 3ghz? If there are 4 core zen chips I expect them to be running at 4.0 or so. Because no way we have 8 core chips and 4 core chips running at 3ghz. I understand GF might not be able to do too much about 3ghz on first revision but to expect 4 core chip stuck at 3.0 is little too low. I expect 4 core chips to be higher clocked than 8 core as it is always the case.
Click to expand...

That depends what the limit is. And Fmax sounds like one of them.
 
NKD said:
wait is everyone expecting 8 core zen and 4 core zen to be clocked at 3ghz? If there are 4 core zen chips I expect them to be running at 4.0 or so. Because no way we have 8 core chips and 4 core chips running at 3ghz. I understand GF might not be able to do too much about 3ghz on first revision but to expect 4 core chip stuck at 3.0 is little too low. I expect 4 core chips to be higher clocked than 8 core as it is always the case.
Click to expand...
trick0502 said:
I think the real problem with zen is going to be glofo. If you look at recent things they've made, they don't clock well. The rx480 has a clock that I have a feeling amd wished was higher. And another great example is apples a9 (6s Soc). The ones on tsmc run cooler and use less power than the Samsung/glofo socs.
Click to expand...

There is something to this we all know how things are with Global Foundries , not a single time that I can remember where GF has saved the day for AMD (regardless in terms that AMD were expecting miracles from them , just attainable goals (Bulldozer to Piledriver clocks were suggested that Piledriver clocks were launch target for Bulldozer)). Then you have to wonder about this right, AMD knew this when Bulldozer happened and now it is in for a repeat? That would be really sad if they managed again to fail at attainable goals with clock speed.



Squall_Rinoa89 said:
All bark, no bite. I'll believe it when they hand out samples, and people start running them vs Intel platforms with nearly the same specs.
Till then... it's all Marketing PR in hopes of making there share holders wet themselves.Reminds me of the slides of "Piledriver" when it then completely failed to get past an i3.
Click to expand...
The outline is a demo with an engineering sample not a retail product. AMD will shoot themselves in the foot when their retail product can't reach decent clock speeds. Thus this marketing is not so much marketing when it can all still go pear shaped ...

No one (consumers) will buy Zen on the premise that it can match Intel on 3ghz this is _terrible_ marketing at best
 
Last edited:
sirmonkey1985 said:
you've been here since 2008? what forum sections are you visiting? lol.. AMD's been on the shit list for the last 5-6 years, there's literally nothing they can do right or wrong where they don't get bashed into the ground by people here.
Click to expand...


Well with AMD it's been a case of 'Disappoint me once, shame on you. Disappoint me twice, shame on me. Disappoint me a third time.. F*** off!"
 
Zarathustra[H] said:
It means they finally caught up to Sandy bridge IF and ONLY IF they can make it clock near 4Ghz.

All rumors about GloFo's process this will be manufactured on - however - suggest that it will be tough to get clocks much above 3Ghz on them.

I was hoping to see some hint about potential clock speeds in this slide deck to dispell those fears, but I saw nothing, and the continued harping on IPC without mentioning total performance or clock speeds makes me suggest that it is true.

In other words, IPC will me massively up from Piledriver FX chips, but clocks will be massively down, resulting in an overall 10% performance increase, when they really need a 60% overall performance increase.

Expect to see AMD trying to throw more cores at the problem again, just like they did with bulldozer, when we all know that with the exception of some narrow usage cases, what really matters is single threaded performance. :(

I had high hopes for Zen, but it's starting to look like its going to be another Bulldozer sized disappointment.
Click to expand...

Yeah, pretty much my conclusion here too. "3ghz" is the only number being thrown around, not "4", let alone 5.

The silver-lining is that maybe it'll panic Intel just enough to overreact and make Kabylake an overclocker's dream, as happened with Sandybridge.
 
Last edited:
spine said:
Yeah, pretty much my conclusion here too. "3ghz" is the only number being thrown around, not "4", let alone 5.

The silver-lining is that maybe it'll panic Intel just enough to overreact and make Kabylake an overclocker's dream, as happened with Sandybridge.
Click to expand...

You dont just change electrical characteristics night over just because someone released something. Kaby Lake and Cannon Lake for that matter will perform as planned long ago, no matter how Zen performs.

You can only change 2 things, price and release date.
 
The last time I checked, most of us are still GPU constrained. Heck, there are people still using 2500K CPUs with the very latest video cards. And SB is nearly five years old now. All AMD needs to do is price ZEN competitively and it will do well.
 
Creig said:
The last time I checked, most of us are still GPU constrained. Heck, there are people still using 2500K CPUs with the very latest video cards. And SB is nearly five years old now. All AMD needs to do is price ZEN competitively and it will do well.
Click to expand...

Case in point:

CPU_01.png


The 1080 GTX can barely max the game at 1080p, but the FX-4230? No issues whatsoever. CPU performance certainly helps, but games have been GPU limited for ages now. Upgrading CPUs for gaming purposes is a thing of the past; they're fast enough to get all the necessary work done, and performance is almost entirely dominated by the GPUs ability to pump out frames.
 
It has to be faster then my 4770 at 4.2 and use less power for me to even think of upgrading. CPUs have been so stale lately. I used to get a new CPU every two years, but I've had this 4770 for what seems like ten years!
 
It has to be faster then my 4770 at 4.2 and use less power for me to even think of upgrading.
Click to expand...

I expect the 8 core will not be faster than your 4770 at 4.2 for applications that use less than 6 or 7 threads (because of the lower clock speed and similar if not lower IPC). The 4C / 8T may be a little faster than the 4770 at 4.2 provided AMD can achieve 4+ GHz clocks on that part.
 
Last edited:
You must log in or register to reply here.
Back
Top