Leaked Geekbench Results: AMD Zen Performance Extrapolated

[i_build_nanosuits]

Ryan Shrout at PCperspective has worked some numbers from the recently leaked Geekbench benchmark results of an engineering sample AMD Zen processor.

http://www.pcper.com/news/Processors/Leaked-Geekbench-Results-AMD-Zen-Performance-Extrapolated

 

He compared the AMD CPU running at 1.44ghz to an ivy-bridge xeon (3.6ghz) by scalling the performance linear with clockspeed, extrapolating then the performance of what the AMD chip could potentially bring if it was clocked at 3.6ghz...results? Ryan Shrout says:

Quote

'' if these results are at all indicative of what AMD Zen will be in the next 4 months, then it will be ~70-80% of the IPC of Ivy Bridge. ''

 

 

 

I certainly hope AMD ZEN core will perform A LOT better than that...maybe this engineering sample is one of the new AM4 APU's based on excavator and NOT a ZEN core based part?

 

So, what you guys think?

 

[alexyy]

What do I think? Hype train is gonna get derailed real fast.

[Agosto]

For some reason, today AMD stocks tanked a bit. Is this because of the 1.02 billion stuff or because of this article based on almost nothing?

[AlwaysFSX]

That's really not good at all. I think that's barely better than the FX series?

[Sakkura]

This is so incredibly speculative that there's really no point.

[Aytex]

Insert Grain of salt 

[i_build_nanosuits]

Just now, Sakkura said:

This is so incredibly speculative that there's really no point.

Well...extrapolating performance numbers based on linear clockspeed increase is fairly reasonable thing to do...usually performance doesn't scale 1:1 with clockspeed but it's usually pretty close...IMHO the real question here is not wheter the methodology is correct here, but rather is that ''engineering sample AMD Zen processor'' really is based on the Zen Core...cause from the numbers IMHO this looks like an excavator core at work more than anything else.

 

[Agosto]

Just now, i_build_nanosuits said:

Well...extrapolating performance numbers based on linear clockspeed increase is fairly reasonable thing to do...usually performance doesn't scale 1:1 with clockspeed but it's usually pretty close...IMHO the real question here is not wheter the methodology is correct here, but rather is that ''engineering sample AMD Zen processor'' really is based on the Zen Core...cause from the numbers IMHO this looks like an excavator core at work more than anything else.

 

That is an A0 engineering sample with crippled clocks and ram speed. It shouldn't even be considered

 

[Trav_X]

Well, we'll just have to wait and see for ourselves on release.

[patrickjp93]

1 minute ago, Agost said:

That is an A0 engineering sample with crippled clocks and ram speed. It shouldn't even be considered

 

??? I wouldn't call 1.44 that super crippled considering the max TDP is only 180W.

 

Also, the memory bandwidth is server standard 2133 on 4 channels. It's exactly what it will be at launch.

[Herman Mcpootis]

12 minutes ago, Aytex said:

Insert Grain of salt 

we need more salt.

 

 

[Agosto]

2 minutes ago, patrickjp93 said:

??? I wouldn't call 1.44 that super crippled considering the max TDP is only 180W.

 

Also, the memory bandwidth is server standard 2133 on 4 channels. It's exactly what it will be at launch.

So what about that 7.75GB/s memory bandwidth multithreaded? That seems kinda off if you ask me

 

[patrickjp93]

Just now, Agost said:

So what about that 7.75GB/s memory bandwidth multithreaded? That seems kinda off if you ask me

 

Someone didn't read the final result.

[Cela1]

2 minutes ago, herman mcpootis said:

we need more salt.

 

MOAAARRRR SALT!

 

[Agosto]

Just now, patrickjp93 said:

Someone didn't read the final result.

I'm talking about the geekbench results http://browser.primatelabs.com/v4/cpu/105227

Moreover, where did you read 2133 MHz RAM speed?

[SpaceGhostC2C]

10 minutes ago, i_build_nanosuits said:

Well...extrapolating performance numbers based on linear clockspeed increase is fairly reasonable thing to do...usually performance doesn't scale 1:1 with clockspeed but it's usually pretty close...IMHO the real question here is not wheter the methodology is correct here, but rather is that ''engineering sample AMD Zen processor'' really is based on the Zen Core...cause from the numbers IMHO this looks like an excavator core at work more than anything else.

 

I think it is much more reasonable to do as a user in this forum did, namely, downclocking a comparison CPU to the same frequency. As AMD did when releasing their official IPC statement. Not perfect, but much better than extrapolation. Since you don't have a Zen, you want to leave the Zen numbers at what they are, and then play around with a CPU you do have, and can control and adjust at will, to make a meaningful comparison.

[patrickjp93]

Just now, Agost said:

I'm talking about the geekbench results http://browser.primatelabs.com/v4/cpu/105227

Moreover, where did you read 2133 MHz RAM speed?

 

Ivy Bridge DDR3 support could be 1600 or 1866 in server applications.

2133/1600 = 1.333125

899 * 1.333125 = 1198.479375

 

Given imperfect scaling because of page swapping, multiply by 4x and you get a 10% efficiency shave and there's your bandwidth score.

[i_build_nanosuits]

1 minute ago, SpaceGhostC2C said:

I think it is much more reasonable to do as a user in this forum did, namely, downclocking a comparison CPU to the same frequency. As AMD did when releasing their official IPC statement. Not perfect, but much better than extrapolation. Since you don't have a Zen, you want to leave the Zen numbers at what they are, and then play around with a CPU you do have, and can control and adjust at will, to make a meaningful comparison.

that would be interesting to see...i might get on it tonight when i'm back from work...i can provide results for haswell at 1.44ghz and compare it to this...but in all honesty...Ryan gave AMD a chance there by doing the 1:1 scale with clockspeed...in real world, performance doesn't scale linear with clockspeed...it's usually lower than 1:1...and the more you increase clocks, the further from 1:1 you get.

[LucidMew]

Well, uhhh, this isn't good news, especially since I thought that Zen was supposed to be Low Clock Speeds (3.2 to 3.8GHz), but high IPC. Let's hope for some other leaked benchmarks that paint a better picture for AMD

[patrickjp93]

1 minute ago, LucidMew said:

Well, uhhh, this isn't good news, especially since I thought that Zen was supposed to be Low Clock Speeds (3.2 to 3.8GHz), but high IPC. Let's hope for some other leaked benchmarks that paint a better picture for AMD

You consider 3.2 LOW for a 32-core chip?! I know power scales quadratically with clock speed but good God man!

[TheRandomness]

21 minutes ago, Aytex said:

Insert Grain of salt 

If the grain was the size of the moon, then yes. 

[LucidMew]

1 minute ago, patrickjp93 said:

You consider 3.2 LOW for a 32-core chip?! I know power scales quadratically with clock speed but good God man!

I mean the high end mainstream chip that's supposed to compete with 4790K to 6700K and 7700K. The unified platform of Zen means that there's the same IPC on servers as there is on desktop. and IPC for IPC, this leaked zen loses against Ivy

[Agosto]

3 minutes ago, patrickjp93 said:

 

Ivy Bridge DDR3 support could be 1600 or 1866 in server applications.

2133/1600 = 1.333125

899 * 1.333125 = 1198.479375

 

Given imperfect scaling because of page swapping, multiply by 4x and you get a 10% efficiency shave and there's your bandwidth score.

Again, I'm talking about direct geekbench results. Even a 980X has more bandwidth than that, I'd say these are quite strange results and I wouldn't consider them to approximate Zen performance

[patrickjp93]

Just now, LucidMew said:

I mean the high end mainstream chip that's supposed to compete with 4790K to 6700K and 7700K. The unified platform of Zen means that there's the same IPC on servers as there is on desktop. and IPC for IPC, this leaked zen loses against Ivy

You expect 3.8GHz on 8 cores in a 95W TDP with Intel's level of performance?! Even IBM's Power 8 can't do that, and that's a RISC architecture!

[LucidMew]

Did I just find another one, literally by searching for http://browser.primatelabs.com/v4/cpu/search?utf8=✓&q= AMD+Eng+Sample

http://browser.primatelabs.com/v4/cpu/199148