• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD Zen Performance Preview

That would not only be disappointing but down right awful.....

And find it hard to happen. You just can't go much worse then what Bulldozer did, they must have learned from their mistakes, and understood what Intel did to gain an advantage. Once something is done in a certain way, its much easier to see a vision of how things can go in the future. and with the talent AMD has should be able to cover ground.
 
I don't buy GB3 tbh, considering that it's proven falsifiable and the weird cache size in system info.
 
Well, keeping in mind that this is a low clocked server part at 1.4Ghz, lets look at the performance per clock.

just dividing the respective single core score by the clock of each part (1.44ghz and 3.5ghz respectively) it shows Zen has 20% higher IPC than Bulldozer. This is not the 40% we were promised, but also keep in mind that we don't know the turbo clock of the Diesel sample.

We know the FX8230 will turbo up to 4ghz in single core.

We have NO IDEA how far that diesel part turbo clocks. Maybe it's a engineering sample without turbo clock enabled, and it's just running at 1.4Ghz.

In that case, these figures represent a 37% increase in IPC, which is pretty close to the 40% IPC gains we were promised.

Long story short, its looking like they will make, or get pretty close to delivering on the 40% IPC gains they bragged about. The problem is, it is unclear if they will be able to reach clock speeds that will be satisfying for desktop enthusiast parts. And even if they hit 4Ghz, we are looking at a Zen that is just slightly SLOWER than a Sandy Bridge i7-2600k a part that will have launched 6 years ago when Zen finally hits.

If they can't hit the clock speed, and they are stuck at ~3Ghz speeds or slightly above, we will see a part that is only very slightly faster than their current FX chips, which will be a HUGE disappointment.
 
Correct me if I'm wrong.
Skylake has 100% more IPC than FX 8xxx series. I say this because - i5 6xxx @ 4.5 Ghz will score similar in multi threaded cinebench as 8 core fx 83xx @ 4.5 Ghz.
So a 40% increase in IPC over the FX will still lead to ~50% less IPC than skylake?
 
a3venom said:
Correct me if I'm wrong.
Skylake has 100% more IPC than FX 8xxx series. I say this because - i5 6xxx @ 4.5 Ghz will score similar in multi threaded cinebench as 8 core fx 83xx @ 4.5 Ghz.
So a 40% increase in IPC over the FX will still lead to ~50% less IPC than skylake?
Click to expand...
It was promised a 40% over Excavator.
 
a3venom said:
Correct me if I'm wrong.
Skylake has 100% more IPC than FX 8xxx series. I say this because - i5 6xxx @ 4.5 Ghz will score similar in multi threaded cinebench as 8 core fx 83xx @ 4.5 Ghz.
So a 40% increase in IPC over the FX will still lead to ~50% less IPC than skylake?
Click to expand...

This is not a perfectly fair comparison because scaling may not be perfect.

Instead I would look at single core scores, and divide by the max turbo clock frequency of each chip to get the single core score per ghz. This should give you about as good of an estimate of IPC as you can hope to get.

If you do that, we see an i5-6600k scores 1.92 with a max turbo clock of 3.9, which is ~0.492 per ghz

An FX8350 scores 1.06 with a max turbo of 4.2Ghz, which is ~0.252 per ghz. This suggests the IPC of Skylake is ~95% higher than Piledriver.

As lolfail mentions above - however - they were referring to excavator, not piledriver

So, there is only one chip in that list of benchmarks with Excavator cores, and that is the Carizzo Athlon x4 845.

It scores 1.09 at a max turbo of 3.8. which is ~0.287 per ghz.

So, Skylake has about 81.5% higher IPC than Excavator.


In other words, if Zen delivers on the +40% IPC promise, Skylake will still be 22.5% ahead in IPC, which isn't THAT bad, but it's not great either.


Now, if we assume they can turbo to 4Ghz, that gives them a score of 1.60, which - if you scroll up and down that anandtech list gives you a score equivalent to some pretty respectable not TOO old chips. It won't be a Skylake killer, but it will be perfectly reasonable in an ethusiasts desktop.

If - however - the rumors are true that the process won't allow for clocks much above 3Ghz, and we assume a 3.2Ghz clock, that gives it a score of 1.29, which is a lot less impressive, considering an AMD FX-9590 clocks in at 1.21
 
Also the key is not to be a skylake killer, but to bring performance back to something they can sell. A lot of people still run sandy bridge systems because they're still well above the performance levels really needed.
 
  • Like
Reactions: dgz
like this
Darakian said:
Also the key is not to be a skylake killer, but to bring performance back to something they can sell. A lot of people still run sandy bridge systems because they're still well above the performance levels really needed.
Click to expand...

Agreed, But, there is a difference between still running a Sandy Bridge level system, and buying one new today...


As far as I am concerned, if they come through with their Excavator+40% IPC estimate, AND they are able to clock it at 4+Ghz I'll consider an 8 core ZEN of the top variety with all the PCIe lanes.

If - however - the rumors about lower clock speeds are true, I won't even consider Zen. THe 3930k will solider on, and when I eventually get tired of this little space heater, I'll plunk down my cash on something Intel.
 
Zarathustra[H] said:
Agreed, But, there is a difference between still running a Sandy Bridge level system, and buying one new today...
Click to expand...
Very true. Price will be key.


Zarathustra[H] said:
If - however - the rumors about lower clock speeds are true, I won't even consider Zen. THe 3930k will solider on, and when I eventually get tired of this little space heater, I'll plunk down my cash on something Intel.
Click to expand...
You and I may not buy it, but if the core counts and prices work they'll make their way into servers where absolute performance doesn't matter as much. It wont be a revolution, but if amd can start selling their chips again they might survive their debt and if they can survive their debt they'll provide competitive pressure to intel and nvidia. We can only hope that they do as much.
 
Slides is one thing, reality is another.

Unless Zen starts to clock a lot higher. Zen performance wont be much better than the current Bristol Ridge top part in ST. You just end up exchanging one thing with another more or less.

excavator.png

amd_bulldozer-nextgen.jpg
 
Now I would like two Opterons (if they are priced like they used to be virtually the same as the retail versions) -> 16 cores/32 threads for less than Intel 8 core. Of course this may not happen for awhile but Opterons were killers in the past and a great buy. My big case is waiting for this :D
 
Shintai said:
Slides is one thing, reality is another.

Unless Zen starts to clock a lot higher. Zen performance wont be much better than the current Bristol Ridge top part in ST. You just end up exchanging one thing with another more or less.

excavator.png

amd_bulldozer-nextgen.jpg
Click to expand...

Yep, that is the concern.

If they hit their 40% IPC target but can't clock above 3ghz top parts won't really be much faster than current FX parts.
 
Shintai said:
XV 5% IPC over SR can also be hard to find.
83865.png
Click to expand...

I agree. Up to is vague, and AMD has a history of being a little bit optimistic with pre-launch figures, that's why all my statements are based on IF they meet their 40% target. If they don't all bets are off.

For a decent enthusiast part we need Zen to both meet the 40% IPC target AND be able to clock up to at least 4Ghz. If it can't do both, it will be a massive fail for us.

Also,

For some reason, CB15 is apparently "legacy" on Anandech, but CB11.5 isn't.

So if you look under CPU -> CPU 2015 Professional Performance Windows -> Cinebnench 11.5 - Single Threaded you can find WAY more data.


Remember to use max turbo clock when calculating IPC performance comparisons in single threaded benchmarks. Going by the base clocks will give you the wrong data.
 
Probably just best to wait for the real McCoy to come out and evaluate real performance from a variety of work loads. Personally I am more interested in the multi-threaded performance or overall performance. I never had a significant issue with the FX line with single threaded applications. Now if I played Arma 3 I would have but have not played that game so if it has 40% greater IPC and less overall performance I would just not bother. As for clock speeds I could care less if the performance is there that I could use.

Now if Zen sucks or just falls flat I may just pull out the I7 6700K in the SFF build, put back in the I7 6500, get new MB etc. and wait a couple of years from there. If I get serious again with 3d work -> Intel 8 - 10 core solutions would work.
 
Darakian said:
Very true. Price will be key.

You and I may not buy it, but if the core counts and prices work they'll make their way into servers where absolute performance doesn't matter as much. It wont be a revolution, but if amd can start selling their chips again they might survive their debt and if they can survive their debt they'll provide competitive pressure to intel and nvidia. We can only hope that they do as much.
Click to expand...

As long as server admins are still obsessed over TDP to reduce TCO, hardly anyone is going to touch AMD even with big discounts if they still lose big to Intel in perf/W, when the CPU costs are only a small fraction of the TCO.
 
Darakian said:
Very true. Price will be key.



You and I may not buy it, but if the core counts and prices work they'll make their way into servers where absolute performance doesn't matter as much. It wont be a revolution, but if amd can start selling their chips again they might survive their debt and if they can survive their debt they'll provide competitive pressure to intel and nvidia. We can only hope that they do as much.
Click to expand...


Intel will match AMD on price, AMD price cut is really not an option, it has never worked in the past, and it won't work now.
 
Just a side note - the 5x86 Cyrix chips were nothing to write home about. The 6x86 Cryrix chips were amazing in integer performance, but horrible in FPU performance. Great for business apps, bad for games.

I'm waiting to see what Zen offers before jumping ship. Right now I'm just eyeing the used Haswell stuff to upgrade my measly i5-3570k Ivy Bridge setup. I really don't feel a huge need to jump, but I'm not super impressed with my A10-7850k for my HTPC system...
 
Shintai said:
XV 5% IPC over SR can also be hard to find.
83865.png
Click to expand...
The real improvement is the efficiency per watt. I think it was a missed opportunity not launching Bristol Ridge with a 95w SKU devoting the extra 30w entirely into 200mhz more on the CPU. I have no doubt such a part will arrive sometime between now and the Zen APUs but if it misses the Christmas/New Year's sales it will be pointless.
 
Palladium@SG said:
As long as server admins are still obsessed over TDP to reduce TCO, hardly anyone is going to touch AMD even with big discounts if they still lose big to Intel in perf/W, when the CPU costs are only a small fraction of the TCO.
Click to expand...

Exactly. The perf/watt winner takes it all. The other could pretty much give their CPUs away for free and still get rejected.
 
Meanwhile over at AT forums the hypothesis that initial GB leaks had Zen running at 1Ghz find an indirect confirmation in the fact that 1.4Ghz Excavator would score similarly to first Naples sample in GB4 single thread.
 
lolfail9001 said:
Meanwhile over at AT forums the hypothesis that initial GB leaks had Zen running at 1Ghz find an indirect confirmation in the fact that 1.4Ghz Excavator would score similarly to first Naples sample in GB4 single thread.
Click to expand...

And for what exactly ? You can see as many engineering samples as you want but that does not reflect upon a retail product, you can guess how much the clock speed is going to be then you might be able to get further, until then all of those "sample" benchmarks mean very little.

You won't see major overhauls to parts between samples and retail other improvements then clock are less likely.
 
At this point clock speeds won't change much either. If they change clock speeds now anything more than 10%, motherboard manufactures and AMD themselves will have to re-validate the product (CPU) and re-validate the motherboard, for both of them to do that, it will take a solid quarter.

This is why ES's that AMD has shown so far, might actually be very close the final release, motherboard manufacturers can't validate motherboards unless they have the final specifications of the CPU's.

And its not the frequency that really matters, its the voltage changes across the chip, but a 10% increase in frequency will probably need voltage changes.
 
razor1 said:
At this point clock speeds won't change much either. If they change clock speeds now anything more than 10%, motherboard manufactures and AMD themselves will have to re-validate the product (CPU) and re-validate the motherboard, for both of them to do that, it will take a solid quarter.

This is why ES's that AMD has shown so far, might actually be very close the final release, motherboard manufacturers can't validate motherboards unless they have the final specifications of the CPU's.

And its not the frequency that really matters, its the voltage changes across the chip, but a 10% increase in frequency will probably need voltage changes.
Click to expand...

I agree with this statement, but the part I disagree with is not what you actually write, but the implied part.

We have seen leaks of 3ghz and 1.4ghz server parts thus far. While I agree last minute changes are unlikely, I don't think we have the data to conclude that these parts that have leaked are their ONLY parts.

They may very well have higher clocked parts that haven't leaked yet that either have finished or are close to finishing validation.

Now, there is that pesky rumor about GloFo not being able to get their process to scale well above 3Ghz too, so things are pointing to a slow ass Zen clocked at 3.2Ghz and performing scantly better than current FX chips, but there is still a narrow slice of hope that this isn't the case.
 
Last edited:
razor1 said:
Intel will match AMD on price, AMD price cut is really not an option, it has never worked in the past, and it won't work now.
Click to expand...

Intel will match AMD on price? What sort of other worldly place do you exist? :D
 
ManofGod said:
Intel will match AMD on price? What sort of other worldly place do you exist? :D
Click to expand...


It will come down to price vs. performance, so doesn't really matter what AMD does, Intel will place their chips where they should be, if AMD was smart they would place their CPU's with Intel's current prices and not force a price war. Then they get higher margins with the same amount of sales. If it goes into a price war, Intel will win out.
 
Zarathustra[H] said:
I agree with this statement, but the part I disagree with is not what you actually write, but the implied part.

We have seen leaks of 3ghz and 1.4ghz server parts thus far. While I agree last minute changes are unlikely, I don't think we have the data to conclude that these parts that have leaked are their ONLY parts.

They may very well have higher clocked parts that haven't leaked yet that either have finished or are close to finishing validation.

Now, there is that pesky rumor about GloFo not being able to get their process to scale well above 3Ghz too, so things are pointing to a slow ass Zen clocked at 3.2Ghz and performing scantly better than current FX chips, but there is still a narrow Alice of hope that this isn't the case.
Click to expand...


True they might not be the only parts. Well I don't think GF is having any issues with their process, nV seems to have switched over to Samsung 14nm for their 1050, so we should be able to get a better picture of the node once that chip is tested.
 
Pieter3dnow said:
And for what exactly ? You can see as many engineering samples as you want but that does not reflect upon a retail product, you can guess how much the clock speed is going to be then you might be able to get further, until then all of those "sample" benchmarks mean very little.

You won't see major overhauls to parts between samples and retail other improvements then clock are less likely.
Click to expand...
ES actually reflect on retail product, contrary to your claims.

The point here (and the sole reason this hypothesis may exist) is that this ES was running downclocked to mere 1GHz on all cores in first benches. Excavator GB4 benches sort-of confirm this assumption. Since we know that 8c will be at least 3Ghz. Multiply it by 3 and suddenly you are only 25% behind 6400 in single thread :p
 
razor1 said:
That would not only be disappointing but down right awful.....

And find it hard to happen. You just can't go much worse then what Bulldozer did, they must have learned from their mistakes, and understood what Intel did to gain an advantage. Once something is done in a certain way, its much easier to see a vision of how things can go in the future. and with the talent AMD has should be able to cover ground.
Click to expand...

It's both easy and difficult to understand what Intel did to gain an advantage.

The difficult part is that there isn't any one thing, any one design aspect that Intel went with that would turn the tables.

The easy part is this. Intel has 95,000 employees, and tons of cash on hand, and spend almost $12 billion on R&D every year. AMD has about 8000 employees, and a dwindling amount of cash and spends only about a a billion USB annually on R&D.

Now, not all of Intel's employees and R&D spending are on CPU's, but the same can also be said for AMD. AMD spends less on R&D for the entire company (CPU's, GPU's Memory, etc.) than Nvidia spends on just GPU's.

Intel has tons of people to throw at meticulously optimizing each CPU design, and doing it rapidly with their tick-tock plan, as well as owning their own fabs that they have tons of people working on keeping the smallest process sizes in the world. AMD is now fabless, has to use whatever process is available to them from vendors, and has WAY fewer people involved in their CPU design process.

This is not just a slight David vs Goliath scenario. It is over the top.

AMD can't just be expected to "figure out what Intel did and replicate it" It's a million little design tweaks both on the product and in the fab process, and they have neither the money or the people to be able to do that.

I'm hopeful that Zen will turn out to be a great chip for enthusiasts, but I am not delusional regarding what the chances of this actually happening are.
 
Indeed. Intel is not at a pentium 4 state these days where they were ready to jump ship (at least on the high end) to a new uarch. Intel have committed to x86 and have been executing year in and year out with improvement after improvement despite the technical difficulties which only get larger. Thankfully AMD can learn from intel in making zen and save on R&D. Intel does publish some good docs on what they do in each processor. This won't push AMD ahead of intel, but it makes it easier to catch up or get close. AMD for there part does have some wind in their sails with rising sales, stock prices, and (I think) consumer perception.
 
Zarathustra[H] said:
It's both easy and difficult to understand what Intel did to gain an advantage.

The difficult part is that there isn't any one thing, any one design aspect that Intel went with that would turn the tables.

The easy part is this. Intel has 95,000 employees, and tons of cash on hand, and spend almost $12 billion on R&D every year. AMD has about 8000 employees, and a dwindling amount of cash and spends only about a a billion USB annually on R&D.

Now, not all of Intel's employees and R&D spending are on CPU's, but the same can also be said for AMD. AMD spends less on R&D for the entire company (CPU's, GPU's Memory, etc.) than Nvidia spends on just GPU's.

Intel has tons of people to throw at meticulously optimizing each CPU design, and doing it rapidly with their tick-tock plan, as well as owning their own fabs that they have tons of people working on keeping the smallest process sizes in the world. AMD is now fabless, has to use whatever process is available to them from vendors, and has WAY fewer people involved in their CPU design process.

This is not just a slight David vs Goliath scenario. It is over the top.

AMD can't just be expected to "figure out what Intel did and replicate it" It's a million little design tweaks both on the product and in the fab process, and they have neither the money or the people to be able to do that.

I'm hopeful that Zen will turn out to be a great chip for enthusiasts, but I am not delusional regarding what the chances of this actually happening are.
Click to expand...


I agree, I don't expect Zen to match up with Intel at the top end, I do expect them to do much better in the mid range, if they can do that, Zen would be a good stepping stone to get better if not, well that's it.
 
razor1 said:
I agree, I don't expect Zen to match up with Intel at the top end, I do expect them to do much better in the mid range, if they can do that, Zen would be a good stepping stone to get better if not, well that's it.
Click to expand...

Gosh I hate to bang the same drum.... but we'd be seeing more by this point if this was anywhere near a win for AMD.

I have been on AMD for 16 years- and even I have to admit we'd have to see something substantial by now.
 
Spirit_Retro said:
Gosh I hate to bang the same drum.... but we'd be seeing more by this point if this was anywhere near a win for AMD.

I have been on AMD for 16 years- and even I have to admit we'd have to see something substantial by now.
Click to expand...


You gotta understand, Intel was able to make a decent architecture with core 2 duo and consistently updated it gen for gen for close to 8 generations now I think. ever since core line was interoduced, AMD has kinda been stuck at what ever they had. Its not like Intel hasn't been innovating either, yeah pure performance wise they only have been going 10-15% per generation, but power consumption wise they have gotten quite a bit better ever generation.

So if Zen's base can hold up and keep AMD going, they can do improvements to core just like Intel did but at a faster pace, and catch up to Intel. To expect AMD to just match up with Intel is great expectations but realistically unless Intel makes a folly I don't see AMD doing that.
 
It still is a long time before release. I mean the APUs are 1 year away. The 8C / 16T is 5 months or so away. Perhaps the problem is GF is not able to clock their 14nm high enough now. Although I would have expected APUs first if that was the case..
 
Spirit_Retro said:
Gosh I hate to bang the same drum.... but we'd be seeing more by this point if this was anywhere near a win for AMD.

I have been on AMD for 16 years- and even I have to admit we'd have to see something substantial by now.
Click to expand...

Not really so, AMD was waving their dick around like crazy years before the bulldozer launch, and that was a complete turd.

Silence is unusual for AMD. So is making a good product. So hopefully there is a correlation.
 
drescherjm said:
It still is a long time before release. I mean the APUs are 1 year away. The 8C / 16T is 5 months or so away. Perhaps the problem is GF is not able to clock their 14nm high enough now. Although I would have expected APUs first if that was the case..
Click to expand...


Looks to me Samsung 14nm has no issue with clock speeds with the 1050 (rumors put it with the 6 pin attached to it it can get to 2ghz + still need confirmation on it because they are only rumors but lets see) so I don't think GF's process which is the same as Samsung's would be a problem for Zen.
 
razor1 said:
Looks to me Samsung 14nm has no issue with clock speeds with the 1050 (rumors put it with the 6 pin attached to it it can get to 2ghz + still need confirmation on it because they are only rumors but lets see) so I don't think GF's process which is the same as Samsung's would be a problem for Zen.
Click to expand...

Well, it IS GlobalF**ups we're talking about. What node have they had that was good?
 
You must log in or register to reply here.
Back
Top