AMD 32core CPUs benchmarked!!

[PCGuy_5960]

1 minute ago, patrickjp93 said:

SMT is confirmed disabled. It's the 32-core.

So these are 64 PHYSICAL Cores??!?!?!

[patrickjp93]

Just now, PCGuy_5960 said:

So these are 64 PHYSICAL Cores??!?!?!

It's 2 CPUs, so 32 cores a piece.

[don_svetlio]

Just now, PCGuy_5960 said:

So these are 64 PHYSICAL Cores??!?!?!

these are 2 PHYSICAL CPUS!!!?!?%!@#?~

[patrickjp93]

17 minutes ago, don_svetlio said:

It's more likely to be 32 cores and 64 threads at 1.4 GHz without any sort of firmware optimization.

It's 2 processors, 64 total cores, no SMT. Read the damn entry pls. And firmware?! Really?! Jesus people CPU socket performance is not determined by firmware or microcode.

[PCGuy_5960]

Just now, patrickjp93 said:

It's 2 CPUs, so 32 cores a piece.

 

Just now, don_svetlio said:

these are 2 PHYSICAL CPUS!!!?!?%!@#?~

I know... It says 2 Processors but the total cores are 64 then, that was my point...

[Herman Mcpootis]

Just now, PCGuy_5960 said:

So these are 64 PHYSICAL Cores??!?!?!

32 cores for each processor

[PCGuy_5960]

Just now, herman mcpootis said:

32 cores for each processor

as I said in this comment

Just now, PCGuy_5960 said:

I know... It says 2 Processors but the total cores are 64 then, that was my point...

 

[Herman Mcpootis]

Just now, PCGuy_5960 said:

as I said in this comment

 

i posted it before your comment appeared, so 

[Sauron]

As other said, for a cpu that's running at 1.4GHz it's pretty impressive. Also, geekbench sucks.

[PCGuy_5960]

Just now, Sauron said:

As other said, for a cpu that's running at 1.4GHz it's pretty impressive. Also, geekbench sucks.

It was the only leaked benchmark...

[PCGuy_5960]

Just now, PCGuy_5960 said:

It was the only leaked benchmark...

And I said kind of disappointing, because these are 64 cores... Not threads....They should be able to score nearly 30000 points no problem

[patrickjp93]

11 minutes ago, Sauron said:

As other said, for a cpu that's running at 1.4GHz it's pretty impressive. Also, geekbench sucks.

At least the results are valid between members of the same ISA. It's not like the lunacy comparison between ISAs when ARM gets vector optimizations and x86 doesn't.

[GirlFromYonder]

Meh. I dont know. Maybe wait for more benchmarks. Maybe this is special type with is extermely power efficent

 

Used where efficiency is needed over performance ( storage stuff) 

[PCGuy_5960]

Just now, DeezNoNos said:

Meh. I dont know. Maybe wait for more benchmarks. Maybe this is special type with is extermely power efficent

 

Used where efficiency is needed over performance ( storage stuff) 

Maybe these CPUs are comparable to Intel's Xeon L lineup, which as you said prioritise power efficiency over performance (not that they don't perform good though)....

If that's the case, these CPUs crush their competition

[Sauron]

18 minutes ago, patrickjp93 said:

At least the results are valid between members of the same ISA. It's not like the lunacy comparison between ISAs when ARM gets vector optimizations and x86 doesn't.

Also, turns out that these are very old samples, pretty much from 1 year ago. You can tell from the stepping, we're supposed to be on A2 now but the benchmarks are on A0.

[PCGuy_5960]

Just now, Sauron said:

Also, turns out that these are very old samples, pretty much from 1 year ago. You can tell from the stepping, we're supposed to be on A2 now but the benchmarks are on A0.

That explains a lot!!! Newer samples will definitely perform better!!

[unsubscribed]

Nice CPU. Clearly not for gamers and more than likely not for content creators. This is typical server CPU - bunch of cores, low TDP. Higher frequency would mean more (much more) power consumed, more heat and issues. Running 4-8 cores at 4GHz is hard enough, never mind whole 32. Heat dissipation is a big and very limiting factor in any electronics.

The only thing what that means to us as consumers is that hopefully we will move more and more towards multicore environment and lower the frequency a bit to stay nice and cool while offloading workload to a bunch of threads.

[rcarlos243]

1 hour ago, FloRolf said:

^this

 

try your 5820 at 1.4ghz (or any other cpu really) and it'll not do any better.

 

1 hour ago, Dackzy said:

dude that CPU is running at 1.44GHz... That is pretty damn good, try and underclock your CPU to 1.44GHz on all cores and no turbo

 

I tried Geekbench4 with my Intel Core i7 6700K underclocked at 1.4Ghz with Turbo Boost Disabled

Now I overclocked my Intel Core i7 6700k clocked at 4.7Ghz with Turbo Boost Disabled

[Notional]

1 hour ago, PCGuy_5960 said:

Idk... Is there any other reason to run the CPUs at just 1.4GHz??

Because that is the weirdest part here...

Engineer samples never run at finalized speeds, but always slower. These are early samples where the chip manufacturing hasn't entered mass production yet. By having a very low clock rate, they can use samples that would otherwise either be downgraded or scrapped. After all early chips like these are very difficult to make.

Furthermore the point of an engineering sample is not benchmarks per se, but rather testing IPC, architecture technologies, microcode, etc. You also use them to test and develop motherboards/UEFI's. So the clock rate simply isn't relevant in testing. This is not an AMD thing either. Exact same modus operandi from Intel.

 

That being said, we don't know what has been disabled here via microcode. We know SMT has been, which means these cores runs up to 30% worse in IPC, as they idle. Also Geekbench sucks. Only proper bench on Linux, which is why they use it I guess.

[SpaceGhostC2C]

1 hour ago, PCGuy_5960 said:

Idk... Is there any other reason to run the CPUs at just 1.4GHz??

Because that is the weirdest part here...

Of course there is. Have a look at current Xeons and Opterons with trillions cores, and teh frequencies they run at. Somethink like the Opteron 6370P or the Xeon E5-2650Lv4.

 

AMD is going to use the new architecture everywhere from laptop to servers, there's no reason to assume every test is about consumer CPUs -especially when they involve 32 corers  

[Vode]

20 minutes ago, rcarlos243 said:

 

 

I tried Geekbench4 with my Intel Core i7 6700K underclocked at 1.4Ghz with Turbo Boost Disabled

Now I overclocked my Intel Core i7 6700k clocked at 4.7Ghz with Turbo Boost Disabled

You should have tried it with hyperthreadong disabled. The Zen ES had it disabled as well.

[Agosto]

20 minutes ago, rcarlos243 said:

I tried Geekbench4 with my Intel Core i7 6700K underclocked at 1.4Ghz with Turbo Boost Disabled

Did you leave the uncore at stock clocks?

[rcarlos243]

13 minutes ago, Vode said:

You should have tried it with hyperthreadong disabled. The Zen ES had it disabled as well.

I am comparing the single core performance so Hyperthreading is irrelevant.

11 minutes ago, Agost said:

Did you leave the uncore at stock clocks?

It is left at Auto which is the default.

[Agosto]

1 minute ago, rcarlos243 said:

I am comparing the single core performance so Hyperthreading is irrelevant.

It is left at Auto which is the default.

I think you should lower it as well... stock frequency is 4GHz

[patrickjp93]

21 minutes ago, Notional said:

Engineer samples never run at finalized speeds, but always slower. These are early samples where the chip manufacturing hasn't entered mass production yet. By having a very low clock rate, they can use samples that would otherwise either be downgraded or scrapped. After all early chips like these are very difficult to make.

Furthermore the point of an engineering sample is not benchmarks per se, but rather testing IPC, architecture technologies, microcode, etc. You also use them to test and develop motherboards/UEFI's. So the clock rate simply isn't relevant in testing. This is not an AMD thing either. Exact same modus operandi from Intel.

 

That being said, we don't know what has been disabled here via microcode. We know SMT has been, which means these cores runs up to 30% worse in IPC, as they idle. Also Geekbench sucks. Only proper bench on Linux, which is why they use it I guess.

Linpack, SAP, and plenty more exist. In fact Google has an entire benchmarking framework available open source.