Discussion: AMD Ryzen

[sarinaide]

Exert on Amadahls principal of multi threading

Another side-effect of the faster single core is that parts that are strictly sequential in nature (where only one thread is active) are processed faster. Amdahl’s law essentially formulates a rather paradoxal phenomenon:

The more cores you add to a CPU, the faster the parallel parts of an application are processed, so the more the performance becomes dependent on the performance in the sequential parts

In other words: the single-threaded performance becomes more important. And that is what makes the multi-core myth a myth!

As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.

We see that Intel continues to be dedicated to improve single-threaded performance. With Bulldozer, AMD decided to trade single-threaded performance for having more cores on die. This is a big part of the reason why Bulldozer is struggling to perform in so many applications, even heavily multithreaded ones.

A FX8370's SMT is barely able to keep up with an i3 because the trade off was single thread efficiency for more cores and more shared resource queing that AMD back then thought clock speed would help. It is the reason why a Phenom II destroys the FX in single thread by about 15% add Phenom II has about 10% less cache latency, 6 Phenom II cores at 3.4Ghz run the blender about 5-6 seconds faster than a FX 8370 at 4.8Ghz.

Ryzen's performance is largely derived from AMD's new core of actual micro architecture engineers deciding that SMT was never the way to go and focused more on FP performance on a per core basis, this is why Zen runs like a 5960X

Ryzen blender scaling:

1 core: 288 seconds
2 core: 144 seconds
3 core: 96s
4 core: 72s
5 core: 57.6s
6 core: 48s
7 core: 41.14s
8 core: 36s

follows amadahls scaling principle in that SMT is sometimes not even dependent upon as the threads wait for schedulers to compile instructions in parallel, single thread instructions are carried out in isolation hence why there is a fall from 66% from 1 to 2 cores all the way down to 13% scaling on SMT over 8 cores. Blender seems to be a more realistic example of how SMT gradually reduces and becomes less linear while Cinebench posts 100 over 100 percent linear equations.

 

[juanrga]

[quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.

 

[sarinaide]

[quotemsg=19232536,0,1284262][quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.
[/quotemsg]

The results will show up more on Intel CPU's, especially the i5 so those results are rather limited, in most situations while the threads are loading up the cores are working symmetrically ramming out single instructions faster. Either way Intel will show advantages with or without access. The crux of it is that the baseline performance per core is vastly improved to the point of being relevent.

Nobody is sitting here claiming AMD's IPC is going to beat Intel's latest architectures, but it is more credible than believing AMD are only going to improve to sandy level when a Phenom II is 20% slower and 5 years older than the sandy uarch.

Phenom II had considerably faster cache compared to Bulldozer and a faster memory controller despite using a 2006 uarch in its design. Zen cannot even remotely be compared to Bulldozer because Bulldozer is not even relevant to a Phenom II other than using 30% higher clock speed.

 

[sarinaide]

http://

http://

Intel have shown on enough occassions that they are willing to offset single thread to push SMT parallelism. The 5960X and 6900K are the ultimate examples of SMT laden behemoths that off set poor gaming performance with high parallelism. It is the market of today but ultimately you need the performance to match Intel at least at Haswell or thereabout to get the kind of performance. Broadwell E has more parallelism resources than Ryzen and that showed in the bench suite by CPC.

 

[jdwii]

Spoiler

[quotemsg=19232672,0,528675][quotemsg=19232536,0,1284262][quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.
[/quotemsg]

The results will show up more on Intel CPU's, especially the i5 so those results are rather limited, in most situations while the threads are loading up the cores are working symmetrically ramming out single instructions faster. Either way Intel will show advantages with or without access. The crux of it is that the baseline performance per core is vastly improved to the point of being relevent.

Nobody is sitting here claiming AMD's IPC is going to beat Intel's latest architectures, but it is more credible than believing AMD are only going to improve to sandy level when a Phenom II is 20% slower and 5 years older than the sandy uarch.

Phenom II had considerably faster cache compared to Bulldozer and a faster memory controller despite using a 2006 uarch in its design. Zen cannot even remotely be compared to Bulldozer because Bulldozer is not even relevant to a Phenom II other than using 30% higher clock speed.



[/quotemsg]

This is absolutely false i have no idea where you got that number from but it ranges far above 20% in most cases some even as far as 60% or more but on average around 30-40%.

http://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/15

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-15.html

Not 100% sure what is going on maybe that was a mistake

Also just for measure Phenom II vs bulldozer

https://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/FX-versus-Phenom-Perf-0

 

[jaymc]

[quotemsg=19226447,0,528675][quotemsg=19225128,0,1284262][quotemsg=19224700,0,528675]Lastly please show me a metric where a poor single threaded CPU can outperform a faster CPU on SMT/CMT,[/quotemsg]

Again?

On single thread the Haswell core is much faster than the IBM power 8 core. With SMT activated the IBM power 8 core matches and even slightly surpasses the Haswell chip.

I explained multiple times how SMT works. I explained in numerous occasions why a core with less IPC will get higher gains enabling SMT than a similar core with higher IPC. I even gave an illustrative example twice.[/quotemsg]


That is a server grade chip, designed for one purpose only, high levels of paralelism. Ryzen is more a enthusiast part but AMD will also pitch it to people in professional environments. Zen games well and it crunches well, enough for solid performance that would sell. I think AMD media have done a good job sandbagging this.

[/quotemsg]

They are targetting zen at the server market too.... Why wouldn't they put in good paralelism / SMT.

 

[jdwii]

http://www.pcworld.com/article/3163500/components/amd-confirms-its-ryzen-cpu-will-launch-in-early-march-followed-by-the-vega-gpu.html

Amd confirmed early march launch

 

[sarinaide]

[quotemsg=19236032,0,25866][quotemsg=19226447,0,528675][quotemsg=19225128,0,1284262][quotemsg=19224700,0,528675]Lastly please show me a metric where a poor single threaded CPU can outperform a faster CPU on SMT/CMT,[/quotemsg]

Again?

On single thread the Haswell core is much faster than the IBM power 8 core. With SMT activated the IBM power 8 core matches and even slightly surpasses the Haswell chip.

I explained multiple times how SMT works. I explained in numerous occasions why a core with less IPC will get higher gains enabling SMT than a similar core with higher IPC. I even gave an illustrative example twice.[/quotemsg]


That is a server grade chip, designed for one purpose only, high levels of paralelism. Ryzen is more a enthusiast part but AMD will also pitch it to people in professional environments. Zen games well and it crunches well, enough for solid performance that would sell. I think AMD media have done a good job sandbagging this.

[/quotemsg]

They are targetting zen at the server market too.... Why wouldn't they put in good paralelism / SMT.[/quotemsg]

Naples will have its own socket and chipset, and like Xeon, it will target server loads, a Xeon is stronger in SMT but weaker in single thread, why would AMD sacrifice performance in "HIGH END PC" for server loads when they are building one.

 

[sarinaide]

[quotemsg=19235680,0,365092]

Spoiler

[quotemsg=19232672,0,528675][quotemsg=19232536,0,1284262][quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.
[/quotemsg]

The results will show up more on Intel CPU's, especially the i5 so those results are rather limited, in most situations while the threads are loading up the cores are working symmetrically ramming out single instructions faster. Either way Intel will show advantages with or without access. The crux of it is that the baseline performance per core is vastly improved to the point of being relevent.

Nobody is sitting here claiming AMD's IPC is going to beat Intel's latest architectures, but it is more credible than believing AMD are only going to improve to sandy level when a Phenom II is 20% slower and 5 years older than the sandy uarch.

Phenom II had considerably faster cache compared to Bulldozer and a faster memory controller despite using a 2006 uarch in its design. Zen cannot even remotely be compared to Bulldozer because Bulldozer is not even relevant to a Phenom II other than using 30% higher clock speed.



[/quotemsg]

This is absolutely false i have no idea where you got that number from but it ranges far above 20% in most cases some even as far as 60% or more but on average around 30-40%.

http://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/15

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-15.html

Not 100% sure what is going on maybe that was a mistake

Also just for measure Phenom II vs bulldozer

https://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/FX-versus-Phenom-Perf-0

[/quotemsg]


I think the Phenom II x 4 965 scores 90 in Cinebench R15, at 3Ghz that is worked out at 80, the i5 2500K is 103 at 3Ghz so it is only 25% thereabout faster, Bulldozer got slower than Phenom, at similar clocks a FX8370 is only 76 despite 4 extra threads.

Obviously a benchmark where a FX runs at 4.4Ghz vs a Thuban or Denebs low 3.3Ghz tells you just how weak the FX IPC is, that it needs that much clock boost to just about beat a Phenom

 

[jdwii]

Spoiler

[quotemsg=19236741,0,528675][quotemsg=19235680,0,365092]

Spoiler

[quotemsg=19232672,0,528675][quotemsg=19232536,0,1284262][quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.
[/quotemsg]

The results will show up more on Intel CPU's, especially the i5 so those results are rather limited, in most situations while the threads are loading up the cores are working symmetrically ramming out single instructions faster. Either way Intel will show advantages with or without access. The crux of it is that the baseline performance per core is vastly improved to the point of being relevent.

Nobody is sitting here claiming AMD's IPC is going to beat Intel's latest architectures, but it is more credible than believing AMD are only going to improve to sandy level when a Phenom II is 20% slower and 5 years older than the sandy uarch.

Phenom II had considerably faster cache compared to Bulldozer and a faster memory controller despite using a 2006 uarch in its design. Zen cannot even remotely be compared to Bulldozer because Bulldozer is not even relevant to a Phenom II other than using 30% higher clock speed.



[/quotemsg]

This is absolutely false i have no idea where you got that number from but it ranges far above 20% in most cases some even as far as 60% or more but on average around 30-40%.

http://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/15

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-15.html

Not 100% sure what is going on maybe that was a mistake

Also just for measure Phenom II vs bulldozer

https://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/FX-versus-Phenom-Perf-0

[/quotemsg]


I think the Phenom II x 4 965 scores 90 in Cinebench R15, at 3Ghz that is worked out at 80, the i5 2500K is 103 at 3Ghz so it is only 25% thereabout faster, Bulldozer got slower than Phenom, at similar clocks a FX8370 is only 76 despite 4 extra threads.

Obviously a benchmark where a FX runs at 4.4Ghz vs a Thuban or Denebs low 3.3Ghz tells you just how weak the FX IPC is, that it needs that much clock boost to just about beat a Phenom
[/quotemsg]

In toms results you can see a 970 gets 4.1 and a 2500K gets 5.4 that would relate to a 31.7% difference and the 970 has a 6% clock speed advantage. Both have no SMT so i find it to be relevant to test them that way. Some other tests can show 60% and others 20%. I used to own a 1100T was an amazing CPU i know what Phenom could do but i also know sandy-bridge is more then 20% faster in IPC on average. If you look at every benchmark in the test suite that both Tomshardware and Anandtech you will see a 2500K vs 970 IPC despite the 970 having a 6% clock speed advantage.

Heck as a example in far cry 2(WOW is also a good test as it uses basically one core back then so is basically all the Game tests in Anandtech) one can see a 2500K beating a 970 by 50%. I can safely say 40% over phenom II would be sandy-bridge performance on average.

Cinebench is only one test, and i think it mainly measures FP unit performance no? I seem to keep finding that what about the performance of integer based operations? I could be wrong however.

 

[sarinaide]

[quotemsg=19236795,0,365092]

Spoiler

[quotemsg=19236741,0,528675][quotemsg=19235680,0,365092]

Spoiler

[quotemsg=19232672,0,528675][quotemsg=19232536,0,1284262][quotemsg=19232064,0,528675]As I have stated on many occassions multi threading is very dependent on single threaded performance, bad single threaded performance equates to bad multithread performance.
[/quotemsg]

Not if the multithreading mode can access to execution units are not accessible for a single thread. Proof given above.
[/quotemsg]

The results will show up more on Intel CPU's, especially the i5 so those results are rather limited, in most situations while the threads are loading up the cores are working symmetrically ramming out single instructions faster. Either way Intel will show advantages with or without access. The crux of it is that the baseline performance per core is vastly improved to the point of being relevent.

Nobody is sitting here claiming AMD's IPC is going to beat Intel's latest architectures, but it is more credible than believing AMD are only going to improve to sandy level when a Phenom II is 20% slower and 5 years older than the sandy uarch.

Phenom II had considerably faster cache compared to Bulldozer and a faster memory controller despite using a 2006 uarch in its design. Zen cannot even remotely be compared to Bulldozer because Bulldozer is not even relevant to a Phenom II other than using 30% higher clock speed.



[/quotemsg]

This is absolutely false i have no idea where you got that number from but it ranges far above 20% in most cases some even as far as 60% or more but on average around 30-40%.

http://www.anandtech.com/show/4083/the-sandy-bridge-review-intel-core-i7-2600k-i5-2500k-core-i3-2100-tested/15

http://www.tomshardware.com/reviews/sandy-bridge-core-i7-2600k-core-i5-2500k,2833-15.html

Not 100% sure what is going on maybe that was a mistake

Also just for measure Phenom II vs bulldozer

https://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/FX-versus-Phenom-Perf-0

[/quotemsg]


I think the Phenom II x 4 965 scores 90 in Cinebench R15, at 3Ghz that is worked out at 80, the i5 2500K is 103 at 3Ghz so it is only 25% thereabout faster, Bulldozer got slower than Phenom, at similar clocks a FX8370 is only 76 despite 4 extra threads.

Obviously a benchmark where a FX runs at 4.4Ghz vs a Thuban or Denebs low 3.3Ghz tells you just how weak the FX IPC is, that it needs that much clock boost to just about beat a Phenom
[/quotemsg]

In toms results you can see a 970 gets 4.1 and a 2500K gets 5.4 that would relate to a 31.7% difference and the 970 has a 6% clock speed advantage. Both have no SMT so i find it to be relevant to test them that way. Some other tests can show 60% and others 20%. I used to own a 1100T was an amazing CPU i know what Phenom could do but i also know sandy-bridge is more then 20% faster in IPC on average. If you look at every benchmark in the test suite that both Tomshardware and Anandtech you will see a 2500K vs 970 IPC despite the 970 having a 6% clock speed advantage.

Heck as a example in far cry 2(WOW is also a good test as it uses basically one core back then so is basically all the Game tests in Anandtech) one can see a 2500K beating a 970 by 50%. I can safely say 40% over phenom II would be sandy-bridge performance on average.

Cinebench is only one test [/quotemsg]

K10 was old by the time Sandy arrived, but in principal it shows a few things;

1) despite half the cache adn half the cache speed it was able to soundly thump a bulldozer at higher clocks, by as much as 20% IPC gains at 3ghz.

2) Sandybridge's bigest leap was the memory controller, AMD's IMC is said to be 5.5x better than the FX which is marginly better than Phenom II so that is round about Haswell type memory controller performance. IMC is probably the biggest influence on light thread performance, it is why a i5 4460 is 20% faster than the 2400 I replaced the HTPC from. Haswell does things that leaves Sandybridge dead and archaic, all benchmarks show that Intel made massive SMT gains from Haswell and on top of decent IPC gains Haswell is the minimum acceptable performance for AMD to hit.

If AMD delivers sandybridge performance, the market will abandon ship and people will buy an i3 7350K which runs rings around a 2600K and 3960X in everything but extreme threaded applications. In gaming sandybridge performance is terrible in 2017. If AMD delivers 2010 performance in 2017 they probably deserve to go out of business.

The i5 has a turbo boost that x4's don't have, that is why the i5 2500K has a single thread score of 126 but a baseline of 102, it is because of turbo

 

[jdwii]

Still in many games tested one can see sandy-bridge at even a 50-60% boost compared to Phenom II. Overall the two CPUs were no where near close in performance.

If i had to say anything about Amd at least they caught up to phenom II if not beaten it with excavator in older software while easily beating it with newer software that can take advantage of the added instruction sets Bulldozer brought to the family.

Ha ha i don't want to bring this thread to this but man i wish Amd would have kept a Phenom like design instead of jumping to Bulldozer, Phenom III 8 core with the newer instruction sets would have been far better IMO.

I have lots of respect for the Phenom architecture however i have very little to none for bulldozer. Was just terrible from every aspect the very fact that they shared even L1 cache can make anyone know, Latency would be a major issue and high latency in a processor just isn't a good idea.

I personally think marketing took over Amd's engineering department with bulldozer. I can never believe the engineering department thought 5Ghz processors was ever gonna be a thing in the mainstream market, even more so after the Pentium 4 failed.

Anyways no comment on march's launch? I was hoping for February but soon we will see the performance of Ryzen.

 

[sarinaide]

[quotemsg=19236870,0,365092]Still in many games tested one can see sandy-bridge at even a 50-60% boost compared to Phenom II. Overall the two CPUs were no where near close in performance.

If i had to say anything about Amd at least they caught up to phenom II if not beaten it with excavator in older software while easily beating it with newer software that can take advantage of the added instruction sets Bulldozer brought to the family.

Ha ha i don't want to bring this thread to this but man i wish Amd would have kept a Phenom like design instead of jumping to Bulldozer, Phenom III 8 core with the newer instruction sets would have been far better IMO.

I have lots of respect for the Phenom architecture however i have very little to none for bulldozer. Was just terrible from every aspect the very fact that they shared even L1 cache can make anyone know, Latency would be a major issue and high latency in a processor just isn't a good idea.

I personally think marketing took over Amd's engineering department with bulldozer. I can never believe the engineering department thought 5Ghz processors was ever gonna be a thing in the mainstream market, even more so after the Pentium 4 failed.

Anyways no comment on march's launch? I was hoping for February but soon we will see the performance of Ryzen. [/quotemsg]

the first thing Lisa Su did when she took over was to hire engineers for positions of engineering that were overun by people with no degree in any kind of engineering.

 

[gamerk316]

[quotemsg=19236915,0,528675]the first thing Lisa Su did when she took over was to hire engineers for positions of engineering that were overun by people with no degree in any kind of engineering.
[/quotemsg]

Speaking as an Engineer: It sucks when you're managed by the Finance department.

 

[sarinaide]

[quotemsg=19237313,0,133194][quotemsg=19236915,0,528675]the first thing Lisa Su did when she took over was to hire engineers for positions of engineering that were overun by people with no degree in any kind of engineering.
[/quotemsg]

Speaking as an Engineer: It sucks when you're managed by the Finance department.[/quotemsg]

I believe AMD put a lot of resources into Ryzen and Vega development, most of the allocation was done once Lisa took over, she is a qualified engineer so that alone was a major step forward.

 

[juanrga]

[quotemsg=19236870,0,365092]
Anyways no comment on march's launch? I was hoping for February but soon we will see the performance of Ryzen. [/quotemsg]

I only can say that it agrees with what I was expecting. I bold the relevant part

http://www.tomshardware.co.uk/forum/id-2986517/discussion-amd-zen/page-35.html#19189678

[quotemsg=juarnga]
1. Ryzen is ready. The last delay is due to bug on mobos: a bug on the BIOS.

2. RyZen official presentation in late February in a special conference by AMD.

3. Ryzen available in first half of March. Only 8C/16T model will be available at launch.

4. The 4C/8T model available one month latter.

5. "Prices are not as cheap as you might think".
[/quotemsg]

 

[Clariska]

TBH i miss the old amd's 2004-2008.My psu eventually blew so while i wait on new psu for my fx.I setup my 4600+ in my main pc's case with a zalman 5x cooler and a custom made steel fan that runs on house power(don't try that at home as that fan can remove fingers/hand -_-)Clocked it from 2.4 to 2.7ghz(my gigabyte mother board doesn't support volt change :x)was hoping to get 3.2ghz out of this baby.Running with an msi 9800gt 1gb.Still plays stuff like borderlands 2 on medium details.To think where did amd go wrong altho i know why .

Now regarding zen.I do think the price gonna be good here in South Africa.Doubt it will come it at 12k like Intel does.My predicament are that it will cost around 3k cheaper than intel here(same with ati/nvidia.ati's usually around 3k cheaper for same performance nvidia)I do not really follow benchmarks as most benches are made for intels architecture.I prefer to do real world performance testing using frame capture software.Any ways i just hope amd delivers as i refuse to go intel with there prices .

 

[jdwii]

[quotemsg=19243447,0,893229]TBH i miss the old amd's 2004-2008.My psu eventually blew so while i wait on new psu for my fx.I setup my 4600+ in my main pc's case with a zalman 5x cooler and a custom made steel fan that runs on house power(don't try that at home as that fan can remove fingers/hand -_-)Clocked it from 2.4 to 2.7ghz(my gigabyte mother board doesn't support volt change :x)was hoping to get 3.2ghz out of this baby.Running with an msi 9800gt 1gb.Still plays stuff like borderlands 2 on medium details.To think where did amd go wrong altho i know why .

Now regarding zen.I do think the price gonna be good here in South Africa.Doubt it will come it at 12k like Intel does.My predicament are that it will cost around 3k cheaper than intel here(same with ati/nvidia.ati's usually around 3k cheaper for same performance nvidia)I do not really follow benchmarks as most benches are made for intels architecture.I prefer to do real world performance testing using frame capture software.Any ways i just hope amd delivers as i refuse to go intel with there prices .[/quotemsg]

Ah i used to own a 4800+ they were very good processors at the time i actually think they were better then the Intel dual cores back then.I expect to see a SR3 be very competitive in terms of price to a I3- low-end I5. In USA that would be 150-180$.

 

[jaymc]

check it out:

http://wccftech.com/amd-ashes-ryzen-4-0-ghz-benchmarks/

 

[gamerk316]

[quotemsg=19245598,0,25866]check it out:

http://wccftech.com/amd-ashes-ryzen-4-0-ghz-benchmarks/[/quotemsg]


I can do some rough math, but I don't have a 1:1 exact setting match against an Intel CPU with the Titan X on Crazy defaults. The closest I can find is:

http://www.ashesofthesingularity.com/metaverse#/personas/9ea335b6-9477-494f-8aac-190e1ba785e4/match-details/c14a6de6-6451-49c0-a141-51cbfeafec8a

Which has a lower Terrain Shadowing Sample setting [12 Million versus 16 Million], and is on a much newer version of the benchmark. I'll do the math, but it's almost certainly going to bias Intel due to the settings differences. If anyone has an Intel CPU/Titan X or can find a run with similar settings, just post and I'll re-do the results. For now, this should get a *ballpark* estimate. I'll use the average results/CPU Framerate for now.

Performance = IPC * Clock * Number of Cores

5930k: 84.4 = IPC * 3.5 * 12
IPC = 84.4 / 3.5 / 12
IPC = ~2

Ryzen: 70.5 = IPC * 4 * 16
IPC = 70.5 / 4 / 16
IPC = ~1.10

So according to these results [which I stress: BIAS INTEL due to setting difference, Ryzen has 90% lower IPC on this specific benchmark, likely due to core scaling stalling out after a few cores. Even if you factor in the bias in settings, this...isn't great.

What worries me: Remember the benches from late last year? Where I computed sub-Sandy Bridge IPC? Same thing here: Sub Sandy-Bridge IPC again. Yes, the results have bias due to setting differences, and yes, ES sample, but I doubt the differences account to 90%.

If someone has a run with more accurate settings, post and do the math [it's not hard]. Or I'll do it. I'd be VERY interested in a Titan X run, with an Intel CPU, at the same settings. I'm really starting to think we may have aimed high on IPC.

EDIT

I don't want to say something is screwy here, but AMD is using version 1.50 of the benchmark. Most of the results are showing AoS is well past version 2 [2.2 to be precise]. Am I missing something, or is AMD using an ancient version of the benchmark? And the fact they tested at a higher setting then the default for Crazy is odd to say the least. Oh, and I just saw that Half Resolution Terrain was set to off in the Ryzen results, likely dragging performance down a bit, but 90%.

Putting aside the benchmark/version difference: The Ryzen results for IPC are outright CRAP. AMD better hope this isn't indicative of Ryzen performance, otherwise it's DOA.

Calling all Titan X owners: Run the same version of AoS at the same settings. Please. Because something is seriously off here. And i want to know what.

EDIT 2

"Half" resolution terrain, not "High" resolution Terrain. That plus the sample difference are likely dragging AMD's results down. Not sure what to make of this here, as I'm being told by some people 1.50 is a newer version then 2.2. Can anyone please elaborate here?

 

[sarinaide]

I dont like the leak,. It seems fishy and the AMDfanboy user name is derogatory. It is probably a photoshopped image based off a older setup

An i7 4790 non k scores 70FPS and both have 3.6-4 clock rates. I don't think half the posted scores are stock as the only non k score is a 4790 while the 4970k gets 100fps

Im calling it a fake but if to assume it was right then it would be a bad situation for the PC market, that would never sell other than low target income markets and would likely set in motion the end of AMD,. Branch out it private SOC and server markets and make phi level cards. Radeon may survive as its market is strong.

 

[8350rocks]

6900K @ 4.0 GHz + Titan X Pascal at various resolutions:

https://www.pugetsystems.com/pic_disp.php?id=40804&width=800&height=800

It only gives average, but shows 65 FPS average on Crazy settings @ 4K

That means that Ryzen @ 3.6-4.0 is not far off at all if this is accurate.

 

[sarinaide]

[quotemsg=19246415,0,1280575]6900K @ 4.0 GHz + Titan X Pascal at various resolutions:

https://www.pugetsystems.com/pic_disp.php?id=40804&width=800&height=800

It only gives average, but shows 65 FPS average on Crazy settings @ 4K

That means that Ryzen @ 3.6-4.0 is not far off at all if this is accurate.[/quotemsg]

Talk about conflicting scores.

There is to much scewering of results because AOTS doesnt scale beyond 4 threads and loves clock rate.

Just going on Gamerks calculation a 8C/8T Zen: 70 (rough round off) / 3.5 / 8 = ~2.5 Adding threads is always going to drop the score rapidly.

 

[sarinaide]

I am officially calling the leak BS and I have some graphical evidence of it. To be fair it wasn't really hard.

Spoiler

I have red ringed the problems with the bench but please pay particular interest to the CPU ID.

AMD's official CES2017 ES running, notice the CPU ID in device manager.

The two give aways:

1) AMD Ryzen is not used, Engineering Sample is used in the string.

2) N implies no iGPU, the AOTS states Y

 

[jdwii]

sarinaide good job finding that