AMD Ryzen Threadripper 3990X terrifies every CPU in Geekbench's multi-core test apart from the Threadripper 3970X and Intel Xeon W-3175X

[Redaktion]

Geekbench is the latest synthetic benchmark to experience the processing power of AMD's Ryzen Threadripper 3990X, with the 64-core monster tearing the multi-core test to shreds. However, the incredible score of 22,045 points still left it trailing its Castle Peak brethren, the Threadripper 3970X, and the Intel workstation processor Xeon W-3175X.

https://www.notebookcheck.net/AMD-Ryzen-Threadripper-3990X-terrifies-every-CPU-in-Geekbench-s-multi-core-test-apart-from-the-Threadripper-3970X-and-Intel-Xeon-W-3175X.452859.0.html

[m53]

Well that's a brand new 32 core CPU with latest zen2 architecture fabbed with TSMC 7nm process. Yet it got beaten by a CPU with 4 less core with older architecture and fabbed on a 14nm Intel process node. Honestly, how is that impressive?

[ariliquin]

Its impressive because this Intel chip is the equivalent Intel chip for the market and is $10k+ in cost and the AMD is $4K. Also the AMD performs significantly higher than 2 x Intel chips at $20K price point. So that higher performance for $16,000 less cost.

On top of all this the AMD does this at a total TDP of 280W and Intel would be 255x2  for a total of a TDP of 510W to provide performance that is 16-30% lower. Now times this by how ever many you need to stock a data centre and the difference in running costs are significant.

As for the next generation of Intel chips, sorry to say they are on the same process as the current ones so expect limited improvements. Also once they release these updated Intel chips AMD will be nearing release of their 4000 series processors with another significant gain in performance.

7nm provides real benefits which cannot be matched until manufacturing moves to 7nm or better for Intel.

[Hauven]

Hi all,
Just to mention I spotted this benchmark here and it looks very much like the one I did. It's genuine but I don't know if the result is up to the level it could be yet.

A few potential problems:
- Due to some initial hardware install nuisances I had to take off the waterblock more than once, which might mean the thermal paste isn't adequately applied at the moment. Temperatures do appear to rise a little fast while the water temperature is just a constant 24C to 25C when idle or under load.
- Windows 10 Pro crashed overnight with 0x000001ca for some reason
- Windows might not have an efficient scheduler yet
- It's fairly new and might need further AGESA changes later on to squeeze more out of the CPU

I will be trying Unraid this morning, with Geekbench 5 on the host O/S. I'm hoping it might yield a better result.

Further proof I'm telling the truth, hopefully:
- CPU-Z validation recently: valid.x86.fr/8lne7x

[Hauven]

Here's a comparison of a Linux (Unraid) benchmark, on the right side, with Windows I previously did which the news article highlights, on the left side.

browser.geekbench.com/v5/cpu/compare/1134736?baseline=1142136

You'll notice that the Linux result is significantly better than Windows. No doubt that's due to Linux having a more efficient CPU scheduler for a start.

[deksman2]

Well that's a brand new 32 core CPU with latest zen2 architecture fabbed with TSMC 7nm process. Yet it got beaten by a CPU with 4 less core with older architecture and fabbed on a 14nm Intel process node. Honestly, how is that impressive?

I strongly suggest you check your eyes and comprehension before you post, because that didn't make any sense.
First off, the Threadripped 3990x is 64 cores with 128 threads.

Second, it got beaten in what exactly?
3990X beats the Intel part in both single threaded performance and massively beats it in multithreaded performance.

On top of that,  the Intel part is 2 CPU's working together with combined 510W TDP, against 1 AMD CPU which has a TDP of mere 280W (meaning AMD consumes 44% less power than the Intel part)... and the Threadripper 3990x costs $4000, whereas the Intel is in excess of $20,000

AMD is faster in single threaded and multi-threaded tasks, consumes 44% less power, and is at least 5x cheaper.

[william blake]

bad scaling, 32>64.
news: look how fast 64 are!

[deksman2]

bad scaling, 32>64.
news: look how fast 64 are!

The 3970x has a base clock of 3.7 GhZ... meaning its all core clocks are probably at 3.9 GhZ on all 32 cores.

The 3990x has base clocks of 2.9 GhZ... which means its all core clocks are likely at 3.1 GhZ across all 64 cores.

AMD usually puts all core boost clocks at around 200 Mhz above baseline.

There's a difference of 800 MhZ in clocks between the CPU's... a difference of about 23%.

Of course scaling won't be perfect, because if AMD is targeting a specific TDP, then a CPU with double the cores will likely end up being clocked lower.

[william blake]

Of course scaling won't be perfect, because if AMD is targeting a specific TDP, then a CPU with double the cores will likely end up being clocked lower.

i am not talking about perfect. we shuold always ask software in 32>64 situations, no matter of the tdp.

[Lorne Kwechansky]

This benchmark result doesn't really many any sense, the Epyc 7702p is a similar CPU to the Threadripper 3990X except it has much lower clock speeds...and it scores close to 40,000 on Geekbench 5.

I'd expect a 3990X to score much higher considering the huge clock difference. The 7702p only posted a single CPU score of 1062 compared to the 3990X score of 1208...that means the 3990X is at least 12% faster than the Epyc equivalent. I'm sure it's even more than 12% faster when all cores are lit up.

So here's my guess. This multi-core score of 22,045 was only using half of the 64 cores. That's why it's almost identical to scores for the the 32 core 3970X.

The real 3990X using all 64 cores should be posting scores well over 44,000 since it's ~12% faster than the 7702p...which means a score of ~22,000 makes perfect sense if only half the CPU was running.

[m53]

Well that's a brand new 32 core CPU with latest zen2 architecture fabbed with TSMC 7nm process. Yet it got beaten by a CPU with 4 less core with older architecture and fabbed on a 14nm Intel process node. Honestly, how is that impressive?

I strongly suggest you check your eyes and comprehension before you post, because that didn't make any sense.
First off, the Threadripped 3990x is 64 cores with 128 threads.

Second, it got beaten in what exactly?
3990X beats the Intel part in both single threaded performance and massively beats it in multithreaded performance.

On top of that,  the Intel part is 2 CPU's working together with combined 510W TDP, against 1 AMD CPU which has a TDP of mere 280W (meaning AMD consumes 44% less power than the Intel part)... and the Threadripper 3990x costs $4000, whereas the Intel is in excess of $20,000

AMD is faster in single threaded and multi-threaded tasks, consumes 44% less power, and is at least 5x cheaper.

(1) Well, Here is what the article says,
"In Geekbench's processor benchmark charts, this mighty score is only topped by the 32-core Ryzen Threadripper 3970X on 22,916 points and the 28-core Intel Xeon W-3175X on 23,521 points."
So the 32 core brand new 3970X scored 22916 in multithread benchmark which is lower than 28 core W-3175X which scored 23521. So yeah I was correct to mention that the 28 core beats the 32 core.

(2) This is what the article mentioned about the multicore performance of the 64 core 3990X

"It's the multi-core score of 22,045 points that is worth pointing out; a great result that should be expected from a powerful chip with 64 cores to play with."

As you can see that is lower than both the 32 core 3970X and 28 core W-3175X which didn't make sense to me. I expect the score to be better for the 64 core CPU. So I didn't comment on the CPU.

(3) You clearly haven't read the article yet you have blamed me for the same and that was not in a nice way. Would it be to much to expect a "sorry" from you?

[m53]

Its impressive because this Intel chip is the equivalent Intel chip for the market and is $10k+ in cost and the AMD is $4K. Also the AMD performs significantly higher than 2 x Intel chips at $20K price point. So that higher performance for $16,000 less cost.

On top of all this the AMD does this at a total TDP of 280W and Intel would be 255x2  for a total of a TDP of 510W to provide performance that is 16-30% lower. Now times this by how ever many you need to stock a data centre and the difference in running costs are significant.

As for the next generation of Intel chips, sorry to say they are on the same process as the current ones so expect limited improvements. Also once they release these updated Intel chips AMD will be nearing release of their 4000 series processors with another significant gain in performance.

7nm provides real benefits which cannot be matched until manufacturing moves to 7nm or better for Intel.

My comment was regarding this portion of the article which has the 32 and 28 core benchmarks,
"In Geekbench's processor benchmark charts, this mighty score is only topped by the 32-core Ryzen Threadripper 3970X on 22,916 points and the 28-core Intel Xeon W-3175X on 23,521 points."
So the 32 core brand new 3970X scored 22916 in multithread benchmark which is lower than 28 core W-3175X which scored 23521. So yeah I was correct to mention that the 28 core beats the 32 core.

Regarding the 64 core 3990X this article provided a lower score than both the 28 and 32 core CPUs mentioned above,
"It's the multi-core score of 22,045 points that is worth pointing out; a great result that should be expected from a powerful chip with 64 cores to play with."
It didn't make sense to me. I expect the score to be better for the 64 core CPU. So I didn't comment on the 64 core CPU.

Now regarding vendor provided benchmark and the price comparison I would suggest you to take that with some grains of salt. In CES 2019 AMD demoed that the engineering sample of the 8 core Ryzen CPU beats the 8 core 9900K in single threaded performance and gaming. But after it was released it didn't come out to be true. Yes they had 12 core and later 16 core parts but that's not what they demo'ed and those parts also have lower single threaded performance.

Regarding the price comparison, AMD is comparing server class Xeon with a enthusiast PC. A server class 64 core Epyc costs twice as much than the 64 core threadripper. Don't you think that was an apple vs orange comparison?