ZEN ES Benchmark from french hardware Magazine

[lolfail9001]

what's the problem with Polaris?

Mostly overhype of fairly mediocre chip.

It is already better than its more expensive competitor, the 1060 6GB.

If you can get it for less, it is. Otherwise, I object, it stacks up as it did at release, mostly.

We have seen Zen being demoed with competitive performance and perf/watt against Broadwell-E . Whats happening with Zen is the opposite of whats happening with Zen. AMD made bombastic statements about Polaris only to underwhelm. With Zen AMD has been managing expectations well for the past 4 years and now when they are close to launch they are showing that they have exceeded their goals. I would say thats the exact opposite of what happened with Polaris. In fact I think the best is yet to come. I think AMD is reserving the best details for launch. Base and Max turbo clocks and max clocks. Anyway the signs are Zen is shaping up to be very competitive. We will see how it turns out at launch.

We have seen Polaris being demoed with competitive performance and perf/watt with Pascal, too.
With that said, my expectation of Zen is based around the Naples Geekbench leak and assumption they (first 2) were at 1Ghz. It does land it in Broadwell-E land, with some exceptions that CanardPC outlines in their instruction table. So far the only question that remained for me was clocks. If it clocks even as badly as Broadwell does, then it should be a fine purchase for most enthusiasts even at fairly high cost. If it clocks better? Well, greater yet!

 

[KTE]

Once Zen uarch was disclosed we knew that a 4 ALUs design with a beefed FPU could hardly have lower IPC than SB/IB wich are both 3 ALUs designs, and that 40% wasnt the real figure to expect from such a design....

So you always expected Zen to have the same/higher IPC than BDe?

Be clear in your claims.

Sent from HTC 10
(Opinions are own)

 

[zinfamous]

Mostly overhype of fairly mediocre chip.

If you can get it for less, it is. Otherwise, I object, it stacks up as it did at release, mostly.

We have seen Polaris being demoed with competitive performance and perf/watt with Pascal, too.
With that said, my expectation of Zen is based around the Naples Geekbench leak and assumption they (first 2) were at 1Ghz. It does land it in Broadwell-E land, with some exceptions that CanardPC outlines in their instruction table. So far the only question that remained for me was clocks. If it clocks even as badly as Broadwell does, then it should be a fine purchase for most enthusiasts even at fairly high cost. If it clocks better? Well, greater yet!

How does it stack up the same as it did at release? That is completely untrue, as 480 is essentially dead-even now with 1060 @ DX11 and the early advantage @DX12 has grown 2-4%. ...that's a huge swing.
http://www.hardwarecanucks.com/foru.../73945-gtx-1060-vs-rx-480-updated-review.html

Yes, price is the issue, as it isn't the same in every country. Fact is, GCN is just a better architecture for games going forward. We've been seeing it for several years now and that hasn't changed. AMD loses on raw power/clocks, but not design.

Define "mediocre." Compared to Pascal, a lazy refresh of Maxwell, how is Polaris in any way "mediocre" considering its tools and overall design for modern tools like DX12?

I guess we will have to see what Vega brings in relation to 1080, but that is the other thread.

 

[Abwx]

So you always expected Zen to have the same/higher IPC than BDe?

Be clear in your claims.

Sent from HTC 10
(Opinions are own)

From the moment we knew the uarch it was clear that they were targeting intel s latest iterations, indeed no one in his right mind would imagine that they would compete adequatly with SB/IB perfs level..

And as said CPC gave a few explanations yesterday on their site, they state that a Zen core total throughput/Hz is 1.35x the one of a Piledriver module, and we know that a PD module is largely up to a SB/IB core be it in FP or in INT, so we are talking of 20-30% better throughput/core/Hz than these two previous intel designs.

 

[minister_of_truth]

How does it stack up the same as it did at release? That is completely untrue, as 480 is essentially dead-even now with 1060 @ DX11 and the early advantage @DX12 has grown 2-4%. ...that's a huge swing.
http://www.hardwarecanucks.com/foru.../73945-gtx-1060-vs-rx-480-updated-review.html

The 480 didn't gain performance, hardwarecanucks just swapped some of the games they tested for AMD sponsored games in order to make the results appear that AMD performance had improved over time. In reality every other credible apples-to-apples comparison shows performance still on par with what we saw the 480 and 1060 at when they launched.

Define "mediocre." Compared to Pascal, a lazy refresh of Maxwell, how is Polaris in any way "mediocre" considering its tools and overall design for modern tools like DX12?

Pascal was a larger performance uplift over Maxwell than GCN4 was over GCN 1.2. Calling either a 'lazy refresh' is a sign that you have no idea what you're talking about.

 

[Thedarkchild]

Mostly overhype of fairly mediocre chip.

If you can get it for less, it is. Otherwise, I object, it stacks up as it did at release, mostly.

We have seen Polaris being demoed with competitive performance and perf/watt with Pascal, too.
With that said, my expectation of Zen is based around the Naples Geekbench leak and assumption they (first 2) were at 1Ghz. It does land it in Broadwell-E land, with some exceptions that CanardPC outlines in their instruction table. So far the only question that remained for me was clocks. If it clocks even as badly as Broadwell does, then it should be a fine purchase for most enthusiasts even at fairly high cost. If it clocks better? Well, greater yet!

Fairly mediocre chip? How? It comes from company completely strapped for cash that was on verge of bankrupcy a year ago, yet it easily rivals best mid range option from Nvidia. Its cheaper, its currently faster in 2016 titles and has considerable advantage in new APIs. It might have worse perf per watt, but who cares about 40 watts in a PC case? I care more about extra 2GB and bigger bus width. I also care more about the fact that in last few years it kinda seems like its a trend that AMD performance advantage only tends to grow as months go by. It started early with RX480...

Plus, having GloFo fab your chips will definitely result in weaker perf per watt no matter the design.

The 480 didn't gain performance, hardwarecanucks just swapped some of the games they tested for AMD sponsored games in order to make the results appear that AMD performance had improved over time. In reality every other credible apples-to-apples comparison shows performance still on par with what we saw the 480 and 1060 at when they launched.

Pascal was a larger performance uplift over Maxwell than GCN4 was over GCN 1.2. Calling either a 'lazy refresh' is a sign that you have no idea what you're talking about.

Actually it did. It was literally slower in games like Division (by 3fps), now its up by 3fps in DX11. Its even worse in DX12. It had 5fps advantage in Witcher 3, now its a draw.

Even yesterday Joker had a video on 2016 games and RX480 is actually wining it.

 

[inf64]

Time to revisit my post from August 18

I have done my own "estimation" on Summit Ridge 8C/16T performance a while ago, I just waited for more information from AMD so I could see if I was near the expected numbers based on released data regarding the uarchitecture choices/features AMD opted for. Now when they released one benchmark and some new information on Zen, I am confident I can post this up now:

I used AT's article on Excavator and applied the ~40% ST uplift on Carrizo's number while adjusting for clocks I expect Zen will launch at (3.2Ghz base and 3.7Ghz ST Turbo). I also applied a 1.2x SMT boost to well threaded benchmarks and adjusted for scaling 12% penalty that Carrizo has when running 2 threads on a module( 0.88x scaling is an average, it varies from benchmark to benchmark). The rest of the numbers used in the sheet are from AT's bench page which is accessible from AT homepage.

This is just a rough estimate of course, Zen might end up at lower clocks and/or lower performance per clock than what I used in the table above. On the other hand, Zen might end up clocking even better and performing better than 40% . We will see on Computex 2017 I guess

Looks like I underestimated Zen by some 5-10% (IPC wise) since I also used the AMD's "40% over XV" statement as axiom here. I was VERY close regarding the clocks but I failed there by a bit too . I missed base by some 7-8% and Turbo, while being still unknown, now looks to be closer to 4Ghz than 3.7Ghz

Overall, if I was to adjust the chart above with new data we have now (6 months later), I would come up with something like this(I just multiplied the scores by 15% and multiplied the run times with 0.85 to adjust for combined higher clock and higher IPC effect of the Zen core as we know it now):

Funny enough, Zen adjusted scores have strong points exactly where AMD benchmarks shined in New Horizon Reveal : Rendering(C-Ray, PovRay) and Handbrake. We know Zen has strong AES execution engine and here the projections shows it again on top(TrueCrypt). All in all this new Zen perf. projection looks to be more in line with what we saw with CanardPC's sneak peak preview than what I assumed before.

 

[DrMrLordX]

There should be plenty of motherboards equipped with beefy VRMs

Let's hope so! Gigabyte is preparing this . . . lovely board:

https://www.techpowerup.com/229076/gigabyte-ax370-gaming-k3-socket-am4-motherboard-pcb-pictured

4+3 phase VRM? Is this a cheap-arse config intended for Bristol Ridge? Eww?

Asus do better! We want 8+2 phase or better stat!

I'll give you a hint: 40% IPC increase over Excavator matches almost exactly the IPC of Sandy / Ivy Bridge. Not Broadwell-E's, which appears to be the case.

I remember going around and around about this with some people earlier in the Zen rumour cycle. See below.

You're attacking him like rabid dogs. It's not like he's going full CENSORED and refusing to admit the possibility that he was wrong even when evidence goes against it.

If anyone can bring it like . . . that guy . . . they'll win even when they're wrong! That is why he is Internet Strong Man.

Oh hell no. Don't even mention that name, who knows, he may be summoned in some way or form.

!!!!!!
!!!!
!!
omg hide

this.

people be crazy in this thread for absolutely no reason. On both sides.

I may be crazy, but I try to be nice about it.

Whats happening with Zen is the opposite of whats happening with Zen.

So you always expected Zen to have the same/higher IPC than BDe?

I won't speak for anyone but myself here, but after seeing Dresdenboy's diagrams about estimated core architecture for Summit Ridge, I realized that there was no way in hell it was going to have the same throughput as Ivy or Sandy, and that it would be possibly better than Haswell. Unless someone from AMD really screwed up the design. That plus what we were told about the changes in the cache hierarchy should have resulted in better-than-Ivy performance, though.

The only statement we had from AMD for a long time was "40% better IPC than XV" which is tricky since XV is a CMT design. Lots of people were looking at XV's horrid ST performance and making estimates on that basis. I figured that AMD would be criminally insane to make such a statement in reference to XV's ST performance. So I went back to the old-school definition of IPC, and assumed it would track linearly on a core-per-core basis as in the K10/Core2 days. So 40% higher IPC would mean 40% higher throughput, right?

It turned out that I was misinterpreting AMD's statement, and they did mean ST IPC.

My estimates were largely based around an admittedly-limited inspection of XV's MT performance in Cinebench R10 - the last Cinebench version where AMD processors seemed to hold up okay on a per-thread basis (they suffer in the later iterations). In CB R10, a 3.4 GHz Carrizo can almost hang in there with a 3.4 GHz i3 in the MT score department; that is, on a per-thread basis, XV is doin alright. I figured if Summit Ridge would give us 40% better than that, there's no way it was going to be trading blows with Sandy Bridge.

Turns out that by being wrong, I was sort of right.

It all comes back to core width, though. Summit Ridge is wide, with a lot more dedicated to fp performance than you see in Con modules.

From the moment we knew the uarch it was clear that they were targeting intel s latest iterations, indeed no one in his right mind would imagine that they would compete adequatly with SB/IB perfs level.

See above. If you only take XV's ST performance and then boost that by 40% and then add in ~%30 for SMT, you get something in the Sandy/Ivy range. Which would have sucked ass. When AMD came out and clarified their statements and confirmed that it was +40% over XV's ST performance, it looked like that might be correct! Good thing so many people were wrong!

We still don't know a lot about Summit Ridge's ST performance. Summit Ridge may just have amazing SMT implementation, or . . . something. I dunno really.

 

[Dresdenboy]

ps. Skylake is 107% ahead of XV in IPC in X265 and over 60% in VP9. They're both integer

But it's a different kind of "integer", running on the FPU, thus any FPU weaknesses apply here.

Of course not, if the performance demonstrated in the leaks and the results AMD has demonstrated are actually constant and true.
However in this case, the performance hasn't increased by 40% but around 55%.

AMD has officially only stated 40% and nothing else.

I hope this won't cause another class action suit. And what should we tell the "up to 40%" believers?

BTW, from Canard PC Hardware (Googlish):

At the time of the tests, there were two types of BIOS / AGESA with quite different performances. The fastest is the one we tested. The other offers a significantly lower IPC, barely superior to that of Intel's Sandy Bridge cores. The explanation could come from one of the hardware bugs present on the first prototypes at the level of the μop cache and the SMT.

http://www.cpchardware.com/cpc-hardware-n31-precisions-elucubrations/

The bug (often compared to the TLB bug) might cause SB level IPC then. Any leaks out there might also suffer from this.

 

[.vodka]

Let's hope so! Gigabyte is preparing this . . . lovely board:

https://www.techpowerup.com/229076/gigabyte-ax370-gaming-k3-socket-am4-motherboard-pcb-pictured

4+3 phase VRM? Is this a cheap-arse config intended for Bristol Ridge? Eww?

Asus do better! We want 8+2 phase or better stat!

It looks like it has 7 chokes / phases at first, but there is an eighth choke in there.

It's kind of weird. This looks like a mid range board at best, nothing too special.

BTW, from Canard PC Hardware (Googlish):

http://www.cpchardware.com/cpc-hardware-n31-precisions-elucubrations/

The bug (often compared to the TLB bug) might cause SB level IPC then. Any leaks out there might also suffer from this.

Even bugged, unfinished Zen is better than whatever the CON family of cores could do on that regard. That's good.

 

[DrMrLordX]

It looks like it has 7 chokes / phases at first, but there is an eighth choke in there.

It's kind of weird. This looks like a mid range board at best, nothing too special.

That is odd. 4+3+1 ?

 

[.vodka]

That is odd. 4+3+1 ?

Do we have any information on what voltages or power planes Zen products need to work? This being a socket that can support APUs or CPUs, both being SoCs with a southbridge/FCH in there too..


I'd love for motherboard makers to stamp on their product boxes, clearly, with big neon sign letters:

THIS MOTHERBOARD CAN TAKE 8C16T ZEN SKY HIGH / SR7?
THIS MOTHERBOARD IS GOOD FOR 4C8T ZEN / SR5?
THIS MOTHERBOARD CAN'T OVERCLOCK A SINGLE CORE ZEN CPU, DON'T EVEN TRY / SR3?

Why not APU specific boards and CPU specific boards? If you know you're not going to touch an APU for the lifetime of your build, as cool as Raven Ridge could end up being.. that could do wonders especially for 8C16T Zen.

Not that such information can't be derived from looking at the VRM area, price or reviews... this will be needed for average joe. Just like there are 95w capable AM3+ boards and 140w capable ones. Unified socket is great, I hope AMD has their partners in control here, cooperating as much as they can to make this as good as it can be on day one... AMD's stigma.

 

[Abwx]

That is odd. 4+3+1 ?

The 1 is likely for the RAM, 3 for the GPU and 4 for the CPU, i suspect that Zen will use both the CPU and GPU power planes as that would reduce substancially the MBs cost.

 

[The Stilt]

Let's hope so! Gigabyte is preparing this . . . lovely board:

https://www.techpowerup.com/229076/gigabyte-ax370-gaming-k3-socket-am4-motherboard-pcb-pictured

4+3 phase VRM? Is this a cheap-arse config intended for Bristol Ridge? Eww?

Asus do better! We want 8+2 phase or better stat!

That appears to be a pretty standard mainstream board, not the high-end (UP-series) we've seen before.
4+3 was chosen as one-size-fits-all type of solution, which isn't really good for anything else but the 65W Bristol and Raven Ridge APUs.

We definitely do not want 8+2 phase boards, unless we have no intention of using the same board with Raven

 

[zinfamous]

The 480 didn't gain performance, hardwarecanucks just swapped some of the games they tested for AMD sponsored games in order to make the results appear that AMD performance had improved over time. In reality every other credible apples-to-apples comparison shows performance still on par with what we saw the 480 and 1060 at when they launched.



Pascal was a larger performance uplift over Maxwell than GCN4 was over GCN 1.2. Calling either a 'lazy refresh' is a sign that you have no idea what you're talking about.

oh yes, just look Orwell's "Minister of Truth," dispensing "information."

 

[DrMrLordX]

Do we have any information on what voltages or power planes Zen products need to work? This being a socket that can support APUs or CPUs, both being SoCs with a southbridge/FCH in there too..

Only thing I can recall is that 4+3 configs are supposed to be common-ish for Bristol Ridge:

4 for CPU

1 for RAM
1 for Northbridge
1 for GPU (separate power plane from NB on AM4, which is a change from earlier sockets)

I'd love for motherboard makers to stamp on their product boxes, clearly, with big neon sign letters:

Only if they promise to tell the truth.

Why not APU specific boards and CPU specific boards? If you know you're not going to touch an APU for the lifetime of your build, as cool as Raven Ridge could end up being.. that could do wonders especially for 8C16T Zen.

It would reduce board costs, but it wouldn't really help Summit Ridge per se. No reason why we should have 8+2 rather than 8+3 or whatever.

The 1 is likely for the RAM, 3 for the GPU and 4 for the CPU, i suspect that Zen will use both the CPU and GPU power planes as that would reduce substancially the MBs cost.

I'm not sure about that one off to the side, but the 3 VRM section is probably as above (RAM, NB, GPU). So for Summit Ridge you'd wind up with an effective 4+2 config here, unless you can figure out how to repurpose the extra GPU VRM to make it work for the CPU instead? 5+2?

That appears to be a pretty standard mainstream board, not the high-end (UP-series) we've seen before.
4+3 was chosen as one-size-fits-all type of solution, which isn't really good for anything else but the 65W Bristol and Raven Ridge APUs.

We definitely do not want 8+2 phase boards, unless we have no intention of using the same board with Raven

Okay fine 8+3, I did say "or better".

 

[SpaceBeer]

4+3 was chosen as one-size-fits-all type of solution, which isn't really good for anything else but the 65W Bristol and Raven Ridge APUs.

My 4+2 phase MB (ASUS A88XM-Plus) can handle 95W TDP A10-7850K pretty well I actually disabled CPB, left CnQ and set multiplier at 40, so I have all core max at 4GHz. It's actually stable at 4.3 GHz, but I don't need that.

 

[Thunder 57]

Mostly overhype of fairly mediocre chip.

If you can get it for less, it is. Otherwise, I object, it stacks up as it did at release, mostly.

We have seen Polaris being demoed with competitive performance and perf/watt with Pascal, too.
With that said, my expectation of Zen is based around the Naples Geekbench leak and assumption they (first 2) were at 1Ghz. It does land it in Broadwell-E land, with some exceptions that CanardPC outlines in their instruction table. So far the only question that remained for me was clocks. If it clocks even as badly as Broadwell does, then it should be a fine purchase for most enthusiasts even at fairly high cost. If it clocks better? Well, greater yet!

Mediocre? It's close to a 290/390 while using much less power. It also wins more often in DX12 and Vulkan games. The 1060 wins overall in DX11, but the RX480 is more than fast enough to play those.

 

[bjt2]

Once Zen uarch was disclosed we knew that a 4 ALUs design with a beefed FPU could hardly have lower IPC than SB/IB wich are both 3 ALUs designs, and that 40% wasnt the real figure to expect from such a design....

Actually skylake can do 4 int uops/cycle. It is limited to 3 uops/cycle (but 256 bit) in FPU. The real problem is that the 4 ALU and 3 FPU share 4 ports. So in int workloads, INTEL and Zen have the same potential, but in heavy FP workloads not. Why? INTEL can do 4 among INT and FP and 4 memory per cycle. AMD can do 4 int, 4 fp and 2 memory. In not heavy memory load (even 128 bit, since INTEL shares ports with int, including branch and compare) AMD shines, principally in SMT, because in ST mode even the relatively low ports of INTEL shine. This is the main reason of higher gain in SMT for Zen. Zen SMT implementation is not better than INTEL's, because it has critical resources statically partitioned (retire, uop queue and store queue), that INTEL don't have, but since Zen has way more ports for actual computations, the net result is better SMT scaling on Zen. I wonder what SMT scaling could be in Zen+ if these limitations will be removed...

 

[bjt2]

From the moment we knew the uarch it was clear that they were targeting intel s latest iterations, indeed no one in his right mind would imagine that they would compete adequatly with SB/IB perfs level..

And as said CPC gave a few explanations yesterday on their site, they state that a Zen core total throughput/Hz is 1.35x the one of a Piledriver module, and we know that a PD module is largely up to a SB/IB core be it in FP or in INT, so we are talking of 20-30% better throughput/core/Hz than these two previous intel designs.

Given the probable zen architecture depicted by Dresdenboy on his blog (almost confirmed, with some little difference by hot chips event), i already predicted higher IPC than sandy bridge.
Hell, even skylake does not match zen in some feature:
Zen: 8 retire/cycle. SKL: 8. Previous INTEL archs: 4-6.
Zen: 2 branches/cycle. SKL: 2. Previous: 1.
Zen: 4 int+4 fp+2 mem, total 10. SKL: 4 int/FP, 4 mem, total 8.. Previous: similar to SKL but with some disadvantages
Zen: 2k uop cache. INTEL: 1.5k uop cache
Zen: 64KB L1I. INTEL: 32KB L1I
Zen: 512KB L2. INTEL: 256KB L2
Zen: L3 on star topology. INTEL: L3 on ring bus topology (slower)

These are the first thing that comes to mind in which Zen is better than INTEL.
The main is in INT+FP throughput: 8 zen and 4 INTEL. And INTEL can do max 3 FP, max 4 INT and the total for INT and FP is max 4. Zen can do 4 INT and 4 FP WITHOUT LIMITS. This is the main reason for which Zen SMT is better. It's not than AMD is better at designing SMT. Simply Zen has more resources...

 

[William Gaatjes]

Zen will do fine.
Not that hard to improve over the construction cores, 4x times the fpu for the same amount of cores (with smt disabled) alone will help a lot.
If you take the blender example, the improvement is obvious.
Besides that, all the architectural enhancements will help on top of that.

I do would like to see a raw throughput INT/FPU/AVX benchmark with only one construction core running and compared to one zen core and compared to one I7-broadwell core.
Would be fun. But likely not going to happen. And not representative for real world use, but interesting for a architectural breakdown.

 

[AtenRa]

The 480 didn't gain performance, hardwarecanucks just swapped some of the games they tested for AMD sponsored games in order to make the results appear that AMD performance had improved over time. In reality every other credible apples-to-apples comparison shows performance still on par with what we saw the 480 and 1060 at when they launched.

Really ??

http://videocardz.com/review/msi-geforce-gtx-1070-sea-hawk-x-review/15

Except Project Cars, the RX 480 is in the lead vs GTX 1060 6GB.

 

[Insert_Nickname]

Yes, price is the issue, as it isn't the same in every country. Fact is, GCN is just a better architecture for games going forward. We've been seeing it for several years now and that hasn't changed. AMD loses on raw power/clocks, but not design.

Well, I'm walking into somewhat of a minefield here, but the AMD/GCN cards do have a small design "fault". It's not relevant for gaming, but they are quite power hungry during video decoding, because AMD chooses to run the memory at full speed. Again this is not an issue as such if you use your PC for gaming, but there are a few situations where it is relevant.

 

[Abwx]

At the end those GPU discussions that regularly derail AMD s CPU/APU threads are just tiring, the guy that started the derailing has disappeared but seems that no one noticed about it...

Besides the thread is about benchmarks of an ES sample, so i dont even understand how a GPU could end being discussed here, granted if we were talking of an APU but that s a pure CPU that is the subject.

 

[Erenhardt]

My 4+2 phase MB (ASUS A88XM-Plus) can handle 95W TDP A10-7850K pretty well I actually disabled CPB, left CnQ and set multiplier at 40, so I have all core max at 4GHz. It's actually stable at 4.3 GHz, but I don't need that.

95 Watts at 1.4V is not even close to 95 Watts at 1.1V (don't know the voltage range for Zen)