Der8auer breaks dual-socket record!

[mynameisjuan]

So this has just turned into a pissing contest. Bottom line is just compare single die scores and call it a day. 

[leadeater]

Just now, Swatson said:

I didn't miss the point, I took issue with your claim. I pointed out in the OP that the DUAL socket epyc almost reached the quad socket score.

Yes correct, but it's meaningless because anyone with any amount of sense would know that that quad socket system does have higher performance. All it shows is that the benchmark is unable to test the higher end systems.

 

The test is not a valid comparison of actual performance of the tested systems, you've reached the limitation of the benchmark so it's only valid for that which makes it a poor benchmark and of little value as a decision aid to figure out the relative performance of the systems. 

 

It's not a slight at this achievement of the dual socket EPYC system, it's an extremely good score but comparing it against a quad socket system like this is to try and show it's near equivalent when it's not and the supplier of the quad socket test system explains why is a bit misleading.

 

1 8180 can score around 4300

2 8180 can score around 8000

 

I think you can see where this is going, roughly 5000 missing points.

[S w a t s o n]

35 minutes ago, leadeater said:

1 8180 can score around 4300

2 8180 can score around 8000

 

I think you can see where this is going, roughly 5000 missing points.

A stock 8180 is not doing 4300 nor are 2 doing 8000, the numbers matters because it's not 5k missing at stock speeds. Also harder to overclock 4 sockets vs 1 or 2

 

Everything I've seen has a stock 8180 2 socket config doing ~6500

[S w a t s o n]

The reason you can't extrapolate the max score an overclocked 8180 could reach to the missing scores on a 4 socket system is because no one is running 4 sockets at those speeds, the world record was on a stock 8180 quad socket system.

[Blademaster91]

47 minutes ago, mynameisjuan said:

So this has just turned into a pissing contest. Bottom line is just compare single die scores and call it a day. 

Yeah it would be the only fair way to be comparatively testing this,especially when it was mentioned how broken Cinebench is.

[leadeater]

20 minutes ago, Swatson said:

A stock 8180 is not doing 4300 nor are 2 doing 8000, the numbers matters because it's not 5k missing at stock speeds. Also harder to overclock 4 sockets vs 1 or 2

 

Everything I've seen has a stock 8180 2 socket config doing ~6500

I've seen a lot of varying scores from different sources, the 6500 I more commonly see for the 8176 not the 8180 though. I've also seen higher than 6500 for the 8176 too.

 

hwbot is hard to fully trust even when it stays stock on the submitted score but that's where I got the 8000 from.

https://hwbot.org/benchmark/cinebench_-_r15/rankings?start=0#start=0#interval=20

https://hwbot.org/submission/3630842_rauwomos_cinebench___r15_2x_xeon_platinum_8180m_8000_cb

 

CPU Monkey lists the single score I mentioned, again same issue with not knowing the test conditions.

https://www.cpu-monkey.com/en/cpu-intel_xeon_platinum_8180-820

 

Trusted results are hard to come by, looking past that there is no way a 8180 is worth 2875 points (1/4 score). Edit: And CB scales much better than this when the run length is sufficiently long enough.

 

Edit 2:

Quote

The benchmarks saw an average all-core CPU clock of 2.7GHz on all 64C/128T of AMD's dual EPYC 7601 system, while the dual Intel Xeon Platinum 8180 had 3.4GHz across all of its dual-CPU goodness at 56C/112T. In the Cinebench R15 performance, we're looking at AMD's EPYC 7601 scoring around 6879 points, while Intel's new Xeon Platinum 8180 pushes 8301.

https://www.tweaktown.com/news/58180/amd-epyc-64c-128t-dual-cpu-against-intel-56c-112t-cpus/index.html

[S w a t s o n]

26 minutes ago, leadeater said:

Edit 2:

https://www.tweaktown.com/news/58180/amd-epyc-64c-128t-dual-cpu-against-intel-56c-112t-cpus/index.html

8180 doesn't do 3.4ghz across all cores stock, anyways it's clear that there is high variance.

Here's a bone stock system with no turbo lock or multi core enhancement enabled.

 

 

[leadeater]

4 hours ago, Swatson said:

8180 doesn't do 3.4ghz across all cores stock, anyways it's clear that there is high variance.

Here's a bone stock system with no turbo lock or multi core enhancement enabled.

8180 all core stock non AVX is 3.2Ghz, the 3.4Ghz from the tweaktown test is probably not quite correct due to how short the CB run is and Intel's turbo behavior. The video you linked shows a system that is significantly under the all core turbo spec from Intel (only 2.73Ghz), likely the server bios is not set to maximum performance power setting. Most servers come default to "Optimized" or something of the like which tries to save a bit of power, by altering the boost tables.

 

Even the STH system doesn't hit the all core spec either but at least that system is doing 2.89Ghz and actually hitting 100% on all cores unlike this video which seems to average 75%-77% on all cores.

[S w a t s o n]

1 minute ago, leadeater said:

8180 all core stock non AVX is 3.2Ghz, the 3.4Ghz from the tweaktown test is probably not quite correct due to how short the CB run is and Intel's turbo behavior. The video you linked shows a system that is significantly under the all core turbo spec from Intel (only 2.73Ghz), likely the server bios is not set to maximum performance power setting. Most servers come default to "Optimized" or something of the like which tries to save a bit of power, by altering the boost tables.

 

Even the STH system doesn't hit the all core spec either but at least that system is doing 2.89Ghz and actually hitting 100% on all cores unlike this video which seems to average 75%-77% on all cores.

The system hit 3.03GHz on all cores but didnt sustain it probably due to the shortness of the test and such.

[leadeater]

Just now, Swatson said:

The system hit 3.03GHz on all cores but didnt sustain it probably due to the shortness of the test and such.

Actually yea, paused it at a bad time right as it was ramping down.

[S w a t s o n]

Just now, leadeater said:

Actually yea, paused it at a bad time right as it was ramping down.

Also hit 100%

[leadeater]

3 minutes ago, Swatson said:

Also hit 100%

It's still 170Mhz down across 56 cores than it should be, probably not 1500 points worth but it should be getting over 7000 just. 8176 are getting 6500 scores with lower all core clocks specs, 2.8Ghz vs 3.2Ghz.