Discussion: AMD Ryzen

[Ecky]

Going to toss one more quote on here:

Disregarding CPU-framerate for a moment, which is consistent sometimes. It's possible to just filter results by what michaelyuan.feng used. At Standard this leaves only 14 Entries, which are simply ranked by the benchmark score, from first to last:
i5-6600K - 6800
i7-6700K - 6600
i7-6700K - 6600
i7-6700K - 6500
i7-4820K - 6300
i7-3930K - 6200
i5-6600K - 6100
i7-4790 - 5900
Zen ES - 5300
i5-4670K - 5000
FX-8310 - 4700
FX-8370 - 4500
FX 8320 - 4000
A10-7890K-3000

There is even a 6/12 Sandy Bridge-E [3.2/3.8 GHz] in the mix scoring 6200, which IMO provides the fairest comparison in terms of cores/threads used. Adjust it for frequency (3.2 -> 2.8 GHz = minus 12.5%) and we get a score of 5425, which again is very close to Zen.

If this is representative of final silicon, AMD hit their 40% IPC improvement, but had to significantly sacrifice clockspeed.

 

[ComputerSecurityGuy]

Got any i7-5960X numbers for that?

 

[Ecky]

Check out the details in the thread I linked.

 

[ComputerSecurityGuy]

Ah. No known system configuration. Erratic CPU comparison. Poor benchmark. I'm just going to say that this benchmark is more or less useless.

 

[Ecky]

I wouldn't say useless, but we're lacking a lot of context. In many ways it lines up with expectations - Zen has significantly more than 40% IPC improvement over Piledriver, but is still slower than Haswell per clock in this benchmark, and clocked lower. Clockspeed is low, likely because it's a huge chip designed for servers which is only coming to desktop as an afterthought, using a process tech that has only been used on low power, low clockspeed parts so far. Zen doesn't have 256bit or 512bit integer pipelines (AVX2), which DX12 can now make use of.

We were expecting an 8 core 16 thread chip with somewhere between Sandy and Haswell IPC, clocked at just north of 3ghz, with a 95w TDP, and that's exactly what we got.

 

[8350rocks]

I wouldn't say useless, but we're lacking a lot of context. In many ways it lines up with expectations - Zen has significantly more than 40% IPC improvement over Piledriver, but is still slower than Haswell per clock in this benchmark, and clocked lower. Clockspeed is low, likely because it's a huge chip designed for servers which is only coming to desktop as an afterthought, using a process tech that has only been used on low power, low clockspeed parts so far. Zen doesn't have 256bit or 512bit integer pipelines (AVX2), which DX12 can now make use of.

We were expecting an 8 core 16 thread chip with somewhere between Sandy and Haswell IPC, clocked at just north of 3ghz, with a 95w TDP, and that's exactly what we got.

Keep in mind, the ES they used is likely not the final silicon.

I would expect that they will end up around 3.2-3.6 base clocks.

3.6 might be a tad bit optimistic, considering; however, original orochi flagship 8 core launched at 3.3, and I would expect we would end up in a similar ball park.

 

[ComputerSecurityGuy]

Even for an ES, they aren't awful. The i7-5960X scores with a R9 390X are in the same area for Ashes of the Singularity, and I think we can assume that they're using a Radeon card in that performance range.

 

[elbert]

Even for an ES, they aren't awful. The i7-5960X scores with a R9 390X are in the same area for Ashes of the Singularity, and I think we can assume that they're using a Radeon card in that performance range.

This isn't the ZEN cpu's to compare against the i7-5960X. The quad and octal are consumer level cpu. Understand AMD has already announce a ZEN 24 and 32 core server dual proc that leaks suggest 2.7Ghz with boosts of 2.9Ghz for the es. Now this means AMD already has a 12 and 16 core CPU for the enthusiasts.

 

[Ecky]

With any luck, this will drive up core counts for Intel CPUs, but most desktop workloads even in 2016 perform better on a fast dual core with HT than on a slower quad, making me wonder what the point of 16, 24 and 32 thread chips in desktop would be. Even in workstations, most tasks that benefit from many cores are becoming GPU accelerated.

Still, if you build it, they will come, and if AMD and Intel both offer high core count CPUs, maybe we'll see more software take advantage of that... in several years.

 

[juanrga]

How close are engineering sample chips clock speeds match the finished products? Years ago, there used to be a substantial difference but lately it seems to be pretty much the same between engineering samples and the shipping product (at least as far as Intel). Anyone know a web page that can show why newer engineering sample chips are deliberately down-clocked? I cannot find a single page that is not just someone's "common sense" opinion.
EDIT Never mind. The Skylake engineering samples had 2.2 GHz ,2.9 GHz boost while the Skylake 6700K chip runs at 4.0 GHz -4.2 GHz boost. The engineering sample clocks are NOT what the finished chips will be running at so take the Zen clock speeds as undefined until the real chips are launched (or leaked!)

Early ES have very low clocks, but late ES have clocks similar to final products. Current ES samples are clocked at 2.8GHz and they are based in A0 steeping. If AMD releases A0-based Zen, then the final clock will be 2.8GHz. If AMD wait for further silicon iterations, then final version of Zen could have 200--300MHz higher clocks.; i.e. 3.0--3.1 GHz.

 

[juanrga]

I'm going to steal a quote from a poster on Anandtech:

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/

First Summit Ridge (Zen) Benchmarks

Straight from Ashes of the Singularity's CPU benchmark:

#COMPARISON 1

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 58.9 FPS
Normal batch: 65.8 FPS
Medium batch: 62.8 FPS
Heavy batch: 50.5 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/50728c1c-7384-447b-99a9-1a7c330cce99/match-details/bfba4b4a-4b1e-4ab3-8f2f-2375321ea68b

- Core i7-980 6C/12T - 3.33 GHz (2010 Westmere)
Average: 58.6 FPS
Normal batch: 65.2 FPS
Medium batch: 59.7 FPS
Heavy batch: 52.3 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/388c81f2-8199-4446-b54b-675f5a0aa833/match-details/45e74e9d-4b0c-4557-8225-a3b8664dd4a8

- Core i7-4770 4C/8T 3.4-3.9 GHz (2013 Haswell)
Average: 66.0 FPS
Normal batch: 74.5 FPS
Medium batch: 69.5 FPS
Heavy batch: 56.6 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/2fb047a6-39a4-4039-8156-ee759c81bd89/match-details/4baad586-6271-42bd-84c5-6884cfb3341d

- Core i7-6700K 4C/8T - 4.0 GHz (2015 Skylake)
Average: 107.3 FPS
Normal batch: 125.7 FPS
Medium batch: 113.8 FPS
Heavy batch: 89.2 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/045e4b2f-dd7b-4187-aafa-cde9dfe20eb6/match-details/6096dd00-7e11-4368-b588-97413922f5c7

The Haswell score is 100% at stock and based on the latest version of the benchmark included in the search engine (1.24.20823.0). The benchmark likes cores/thread and scales with more than 4C/8T:



3.0 GHz Haswell-E beating 4.0 GHz Skylake-S.


#COMPARISON 2

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 31.5 FPS
Normal batch: 36.5 FPS
Medium batch: 33.8 FPS
Heavy batch: 26.2 FPS

- Core i5-4670K 4C/4T 3.4-3.8 GHz (2013 Haswell)
Average: 52.6 FPS
Normal batch: 56.9 FPS
Medium batch: 54.4 FPS
Heavy batch: 47.5 FPS

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/page-4#post-38414245

And as expected Zen IPC is very behind Haswell

http://www.guru3d.com/news-story/amd-zen-engineering-sample-aos-further-analysis.html

 

[juanrga]

I wouldn't say useless, but we're lacking a lot of context. In many ways it lines up with expectations - Zen has significantly more than 40% IPC improvement over Piledriver

Not in that leaked benchmark, Zen is barely ahead Piledriver. See above comparison with 8370.

That public 40% improvement over Excavator is in a specific benchmark (CB15), not valid across any benchmark.

 

[Ecky]

In that benchmark you linked, the per-clock improvement exceeds 40% at times. Unfortunately, it's nowhere near enough, as predicted. Imperfect scaling probably makes the differences smaller than they could be otherwise, but that does not obscure the fact that Zen is still well behind Intel's current CPUs.

 

[juanrga]

This isn't the ZEN cpu's to compare against the i7-5960X. The quad and octal are consumer level cpu. Understand AMD has already announce a ZEN 24 and 32 core server dual proc that leaks suggest 2.7Ghz with boosts of 2.9Ghz for the es. Now this means AMD already has a 12 and 16 core CPU for the enthusiasts.

Those 24 and 32 core server CPUs are quad die configurations. The maximum number of cores per die is eight. The highest Zen CPU for consumer is an eight core 16 thread CPU. This is the maximum for enthusiasts. Mainstream users will be served by quad-cores.

 

[cdrkf]

I'm going to steal a quote from a poster on Anandtech:

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/

First Summit Ridge (Zen) Benchmarks

Straight from Ashes of the Singularity's CPU benchmark:

#COMPARISON 1

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 58.9 FPS
Normal batch: 65.8 FPS
Medium batch: 62.8 FPS
Heavy batch: 50.5 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/50728c1c-7384-447b-99a9-1a7c330cce99/match-details/bfba4b4a-4b1e-4ab3-8f2f-2375321ea68b

- Core i7-980 6C/12T - 3.33 GHz (2010 Westmere)
Average: 58.6 FPS
Normal batch: 65.2 FPS
Medium batch: 59.7 FPS
Heavy batch: 52.3 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/388c81f2-8199-4446-b54b-675f5a0aa833/match-details/45e74e9d-4b0c-4557-8225-a3b8664dd4a8

- Core i7-4770 4C/8T 3.4-3.9 GHz (2013 Haswell)
Average: 66.0 FPS
Normal batch: 74.5 FPS
Medium batch: 69.5 FPS
Heavy batch: 56.6 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/2fb047a6-39a4-4039-8156-ee759c81bd89/match-details/4baad586-6271-42bd-84c5-6884cfb3341d

- Core i7-6700K 4C/8T - 4.0 GHz (2015 Skylake)
Average: 107.3 FPS
Normal batch: 125.7 FPS
Medium batch: 113.8 FPS
Heavy batch: 89.2 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/045e4b2f-dd7b-4187-aafa-cde9dfe20eb6/match-details/6096dd00-7e11-4368-b588-97413922f5c7

The Haswell score is 100% at stock and based on the latest version of the benchmark included in the search engine (1.24.20823.0). The benchmark likes cores/thread and scales with more than 4C/8T:



3.0 GHz Haswell-E beating 4.0 GHz Skylake-S.


#COMPARISON 2

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 31.5 FPS
Normal batch: 36.5 FPS
Medium batch: 33.8 FPS
Heavy batch: 26.2 FPS

- Core i5-4670K 4C/4T 3.4-3.8 GHz (2013 Haswell)
Average: 52.6 FPS
Normal batch: 56.9 FPS
Medium batch: 54.4 FPS
Heavy batch: 47.5 FPS

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/page-4#post-38414245

And as expected Zen IPC is very behind Haswell

http://www.guru3d.com/news-story/amd-zen-engineering-sample-aos-further-analysis.html

My issue with that comparison: We are comparing quad channel vs dual channel memory... which is interesting when you see the Haswell part scaling almost perfectly twice zen. Also looking at the gap between FX -> Zen looks like memory bandwidth. There's something very odd about how AOTS scales, I wouldn't take this as indicative of IPC necessarily- The Zen cpu (and FX) *don't gain anything from reduced load* which points to a platform bottleneck other than the cpu imo. I'd need to see cpu utilisation numbers to back that up, but I'd bet the Zen part isn't maxing out any of it's cores.

Of course a platform limitation like that is still an issue for AMD.... however I want to see single thread vs single thread before we consider the 40% IPC gains.

 

[juanrga]

With any luck, this will drive up core counts for Intel CPUs, but most desktop workloads even in 2016 perform better on a fast dual core with HT than on a slower quad, making me wonder what the point of 16, 24 and 32 thread chips in desktop would be. Even in workstations, most tasks that benefit from many cores are becoming GPU accelerated.

Still, if you build it, they will come, and if AMD and Intel both offer high core count CPUs, maybe we'll see more software take advantage of that... in several years.

Intel has been increasing the core count of CPUs systematically during years:

45nm: 4 core
32nm: 6 core
2nm: 8 core
14nm: 10 core
10nm: 12 core?

The reason why mainstream continues being quad-core is because most of software doesn't benefit from moar cores and will not do in next years. Intel and AMD could release 16-core for mainstream desktop tomorrow and 99% of software will continue to run faster on a higher clocked quad-core. I also hope the hype about GPU acceleration dies soon. GPU only works in a pair of special applications and fails in the rest.

 

[gamerk316]

With any luck, this will drive up core counts for Intel CPUs, but most desktop workloads even in 2016 perform better on a fast dual core with HT than on a slower quad, making me wonder what the point of 16, 24 and 32 thread chips in desktop would be. Even in workstations, most tasks that benefit from many cores are becoming GPU accelerated.

Still, if you build it, they will come, and if AMD and Intel both offer high core count CPUs, maybe we'll see more software take advantage of that... in several years.

For reasons I've explained MANY times in the past: Most software doesn't scale, and the software that does tends to scale to infinity, and are being processed by GPUs instead. There's a reason why performance for non-benchmark software peaks somewhere between an i3 and i5, and why per-core performance matters more then core count.

Honestly, Zen would be better off if they cut the number of cores in half and gained 500 MHz. The 16% Clockrate boost is worth FAR more then a doubling of core count.

 

[juanrga]

In that benchmark you linked, the per-clock improvement exceeds 40% at times. Unfortunately, it's nowhere near enough, as predicted. Imperfect scaling probably makes the differences smaller than they could be otherwise, but that does not obscure the fact that Zen is still well behind Intel's current CPUs.

The IPC gain claimed by AMD is for a single thread. It is one thread on Zen vs one thread on Excavator. The AoS benchmark is multitthreaded. Therefore the performance gains over Piledriver clock-for-clock are a combination of IPC gains per core/thread plus SMT gains (~20% for Zen) plus the CMT module penalty (~20% for Piledriver).

Therefore if Zen is 50% faster in multitthread and ~20% is from the module penalty on Piledriver. This means that Zen core is ~25% faster than Piledriver core. I think SMT brings very little gains in this benchmark (I doubt it scales well to 16T), therefore the IPC gain must be ~20% over Piledriver.

 

[elbert]

I'm going to steal a quote from a poster on Anandtech:

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/

First Summit Ridge (Zen) Benchmarks

Straight from Ashes of the Singularity's CPU benchmark:

#COMPARISON 1

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 58.9 FPS
Normal batch: 65.8 FPS
Medium batch: 62.8 FPS
Heavy batch: 50.5 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/50728c1c-7384-447b-99a9-1a7c330cce99/match-details/bfba4b4a-4b1e-4ab3-8f2f-2375321ea68b

- Core i7-980 6C/12T - 3.33 GHz (2010 Westmere)
Average: 58.6 FPS
Normal batch: 65.2 FPS
Medium batch: 59.7 FPS
Heavy batch: 52.3 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/388c81f2-8199-4446-b54b-675f5a0aa833/match-details/45e74e9d-4b0c-4557-8225-a3b8664dd4a8

- Core i7-4770 4C/8T 3.4-3.9 GHz (2013 Haswell)
Average: 66.0 FPS
Normal batch: 74.5 FPS
Medium batch: 69.5 FPS
Heavy batch: 56.6 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/2fb047a6-39a4-4039-8156-ee759c81bd89/match-details/4baad586-6271-42bd-84c5-6884cfb3341d

- Core i7-6700K 4C/8T - 4.0 GHz (2015 Skylake)
Average: 107.3 FPS
Normal batch: 125.7 FPS
Medium batch: 113.8 FPS
Heavy batch: 89.2 FPS

http://www.ashesofthesingularity.com/metaverse#/personas/045e4b2f-dd7b-4187-aafa-cde9dfe20eb6/match-details/6096dd00-7e11-4368-b588-97413922f5c7

The Haswell score is 100% at stock and based on the latest version of the benchmark included in the search engine (1.24.20823.0). The benchmark likes cores/thread and scales with more than 4C/8T:



3.0 GHz Haswell-E beating 4.0 GHz Skylake-S.


#COMPARISON 2

- Summit Ridge 8C/16T ES - 2.8-3.2 GHz (2016/2017 Zen)
Average: 31.5 FPS
Normal batch: 36.5 FPS
Medium batch: 33.8 FPS
Heavy batch: 26.2 FPS

- Core i5-4670K 4C/4T 3.4-3.8 GHz (2013 Haswell)
Average: 52.6 FPS
Normal batch: 56.9 FPS
Medium batch: 54.4 FPS
Heavy batch: 47.5 FPS

https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/page-4#post-38414245

And as expected Zen IPC is very behind Haswell

http://www.guru3d.com/news-story/amd-zen-engineering-sample-aos-further-analysis.html

Unless the ZEN 8 cores are 250~300. Again the 16 core ZEN is what should be compared to the i7 8 and 10 core. The 12 core ZEN against the i7 6 cores. These are the enthusiast CPU's.
Update: I was speaking of the zen 4 and 8 cores being Kaby lake compared. The 12 core to the i7 6 to possible 8 core broadwell-e/skylake-e. The 16 core comparing to the broadwell-e/skylake-e 8, 10, and possible the 12 core. Once the 4 and 8 core ZEN are launched they should be 4Ghz parts able to compete with any of the consumer main stream CPU's including 4 core kaby lake.
Note: In the test above the Intel 5960x was run at a 400Mhz advantage for a good deal of the test. The 3.2Ghz boost clock doesn't stay in effect for long to keep the CPU below 95watts in the zen part. Given the advantage of the 4 channel the Intel should have been run at 2.8Ghz only.

 

[juanrga]

With any luck, this will drive up core counts for Intel CPUs, but most desktop workloads even in 2016 perform better on a fast dual core with HT than on a slower quad, making me wonder what the point of 16, 24 and 32 thread chips in desktop would be. Even in workstations, most tasks that benefit from many cores are becoming GPU accelerated.

Still, if you build it, they will come, and if AMD and Intel both offer high core count CPUs, maybe we'll see more software take advantage of that... in several years.

For reasons I've explained MANY times in the past: Most software doesn't scale, and the software that does tends to scale to infinity, and are being processed by GPUs instead. There's a reason why performance for non-benchmark software peaks somewhere between an i3 and i5, and why per-core performance matters more then core count.

Honestly, Zen would be better off if they cut the number of cores in half and gained 500 MHz. The 16% Clockrate boost is worth FAR more then a doubling of core count.

But Zen is specifically designed for servers where cores count a lot of. That is why their base die is 8-core instead 2-core. They will be reusing failed server dies for desktop, that is all. AMD doesnt have the money to do a die for servers and another die for consumer desktop, as Intel does. Moreover, AMD choose a mobile oriented processor node for similar reasons. Most servers are clocked down. Enthusiast desktop users will be perplexed because their Zen CPU probably will not hit 4.5GHz under watter cooling, but AMD care little about those. The target is servers.

 

[juanrga]

Unless the ZEN 8 cores are 250~300. Again the 16 core ZEN is what should be compared to the i7 8 and 10 core. The 12 core ZEN against the i7 6 cores. These are the enthusiast CPU's.

As explained to you a pair of posts above there is no 12-core and 16-core Zen for desktop. The top desktop Zen is 8-core CPU.

 

[elbert]

This isn't the ZEN cpu's to compare against the i7-5960X. The quad and octal are consumer level cpu. Understand AMD has already announce a ZEN 24 and 32 core server dual proc that leaks suggest 2.7Ghz with boosts of 2.9Ghz for the es. Now this means AMD already has a 12 and 16 core CPU for the enthusiasts.

Those 24 and 32 core server CPUs are quad die configurations. The maximum number of cores per die is eight. The highest Zen CPU for consumer is an eight core 16 thread CPU. This is the maximum for enthusiasts. Mainstream users will be served by quad-cores.

Actually the max number is 4 if you notice cache stacks in 4 core increments. Not that it matters but ive yet to see either company slap 4 seperate silicon together as you suggest.

 

[elbert]

Unless the ZEN 8 cores are 250~300. Again the 16 core ZEN is what should be compared to the i7 8 and 10 core. The 12 core ZEN against the i7 6 cores. These are the enthusiast CPU's.

As explained to you a pair of posts above there is no 12-core and 16-core Zen for desktop. The top desktop Zen is 8-core CPU.

You do understand that only talking about summit ridge.

 

[juanrga]

This isn't the ZEN cpu's to compare against the i7-5960X. The quad and octal are consumer level cpu. Understand AMD has already announce a ZEN 24 and 32 core server dual proc that leaks suggest 2.7Ghz with boosts of 2.9Ghz for the es. Now this means AMD already has a 12 and 16 core CPU for the enthusiasts.

Those 24 and 32 core server CPUs are quad die configurations. The maximum number of cores per die is eight. The highest Zen CPU for consumer is an eight core 16 thread CPU. This is the maximum for enthusiasts. Mainstream users will be served by quad-cores.

Actually the max number is 4 if you notice cache stacks in 4 core increments. Not that it matters but ive yet to see either company slap 4 seperate silicon together as you suggest.

I predicted years ago that AMD would use quad-die SoCs of 8-core each.

And it was confirmed recently.

 

[Ecky]

I was under the impression that AM4 was going to be dual channel only, and it wouldn't make much sense to me to have a 32 thread CPU fed by dual channel RAM.