AMD Ryzen 9 3950X Cinebench R15 Performance Spied

[btarunr]

Market availability of the 16-core Ryzen 9 3950X may be far away, given its September 2019 launch, but engineering samples (ESes) of the chip seem to be already in circulation. "uzzi38" on Twitter posted this spy-shot of a 3950X ES making short work of Cinebench R15. CPU-Z recognizes the chip by its codename "Matisse," and puts out the correct CPU core and thread count, but doesn't give a name-string. It also recognizes the MSI MEG X570 GODLIKE motherboard this test is run on.

The purported Ryzen 9 3950X ES, overclocked to 5.42 GHz, scores a gargantuan 5,501 points in the multi-threaded benchmark. To put this number into perspective, at stock frequencies, a Ryzen Threadripper 2950X (same core-count, double the memory bus width), scores 3,645 points. The 3950X benefits from not just its massive overclock that's over 1 GHz higher than the stock TR-2950X, but also higher IPC, and a more consolidated memory interface. This feat goes to show that AMD's upcoming Ryzen chips love to overclock, and deliver a significantly higher single-thread performance over the previous generation.



View at TechPowerUp Main Site

 

[UrbanCamper]

That is pretty quick. There must have been some extreme robust cooling on the chip though. 1.775v is pretty high. If this is true that is. Nice find if it is.

 

[EarthDog]

LN2 for those clocks all c/t.

 

[Aerpoweron]

just for reference:

2200ish on the 9900K at 5.2 GHz if you just double that we would be around 4400ish for 16 cores / 32 threats

I am not sure how relieable these results are. And since my 9900K is out of commission at the moment i can't verify for myself.

 

[Vayra86]

Really, @btarunr ?

This feat goes to show that AMD's upcoming Ryzen chips love to overclock, and deliver a significantly higher single-thread performance over the previous generation.

We are looking at some extreme OC result (not described in article) at 1.7v, and what exactly is the single thread performance boost we see here? Its a MT score. Clickbait is one thing, but now you're just flat out lying.

And its a pretty sad sight. Ryzen doesn't need this, and neither does TPU. Stahp

 

[s3thra]

This chip may turn out to be an absolute monster! Wow!

 

[HTC]

Clearly LN2, judging from both the clocks and the voltage used in CPU-Z.

Compared to the original TR 1950X @ 5.2 GHz (same cores / threads: see pic below), that's over 33% increase for roughly 240 MHz higher clocks. I'd say that's very impressive.

Also, and dunno if anyone else noticed, the RAM is @ 4000 MHz CL 18, meaning the IF is supposed to be running @ half speed due to being higher than 3733.

 

[Aerpoweron]

I would like to see AVX / AVX2 benchmarks compared to the latest intel generation. AMD had / has a lot of catching up to do there.

 

[ratirt]

Maybe the voltage is being read wrongly? I hope it has been read wrongly otherwise there isn't anything great about it. I think most of people don't care about extreme oc under LN2.
Anyway the score is impressive (even if it is under LN2) and the IPC has increased exponentially.

 

[TheLostSwede]

Maybe the voltage is being read wrongly? I hope it has been read wrongly otherwise there isn't anything great about it. I think most of people don't care about extreme oc under LN2.
Anyway the score is impressive (even if it is under LN2) and the IPC has increased exponentially.

Indeed, it's pointless for 99.9999999999999999999999% of people.

I would like to see AVX / AVX2 benchmarks compared to the latest intel generation. AMD had / has a lot of catching up to do there.

And how many people use AVX anything in every day usage scenarios? I know Intel is pushing AVX512 heavily, but it doesn't seem widely used, except in some niche applications.
I'd say the most widely used application is OpenSSL an derivatives of it, at least from what I can tell.

 

[ratirt]

And how many people use AVX anything in every day usage scenarios? I know Intel is pushing AVX512 heavily, but it doesn't seem widely used, except in some niche applications.
I'd say the most widely used application is OpenSSL an derivatives of it, at least from what I can tell.

I think Intel knows it is losing ground to AMD now and the only thing why Intel's pushing AVX512 is cause that's the only one thing it can stay on top of a benchmark chart with AVX over AMD.
So Intel can still say "we are better in AVX" than AMD even if the usage of this is marginal.

 

[Aerpoweron]

As far as i have seen the Zen2 CPUs won't support AVX 512. Even CPU-Z says for the 3950X as well as for the 1950X that AVX2 is supported. A clock for clock comparison of the 2 CPUs would be interesting to see for Cinebench R15 and R20. Does Cinebench R20 support AVX? The changelog on TPU in the download section it only says that it support modern CPU feature from AMD and Intel to run twice as fast as R15 did.
I have to agree with TheLostSwede, AVX is rarely used. I think too that AMD does not focus on it too much is a good decision. Since it adds a huge load of complexity, and can heat the CPU very well.

I hope the person who posted the 3950X screenshot will post results from other programs so we can see how it does there.

 

[Basard]

Seems like they will "love to overclock" just about as much as previous Ryzen chips--if not less. Almost 1.8v!

 

[Tomgang]

Good score. But hornestly i am more interesting to see what the cpu can do stock and overclock on normal cooling, meaning air/aio/custom water.

LN2 results are useless for every day use. For normal cooling results, we will have to wait to 7 juli.

 

[Mephis]

Sweet, no reason for any websites to do reviews. Or at least do an overclocking section in reviews. Guys, it is official. Ryzen 3000 loves to overclock. /s

 

[infrared]

You can't gauge the overclocking potential of the retail chips based on an engineering sample.. They don't usually overclock well so imo the voltage and cooling are irrelevant and to be expected, all you can get from this is the impressive score at a fairly modest frequency. As always we need to wait until people have their hands on retail chips to see what they are actually capable of.

 

[HTC]

Good score. But hornestly i am more interesting to see what the cpu can do stock and overclock on normal cooling, meaning air/aio/custom water.

LN2 results are useless for every day use. For normal cooling results, we will have to wait to 7 juli.

I don't expect the 3950X to have high all core speed on conventional cooling: @ least when compared to Intel's.

Reason for that is described in the following video (skip to 6:50):

The above video may also actually explain Intel's problems with 10nm: smaller process leads to a reduction in clocks barring any "extraordinary things", such as chiplet design, instead of monolithic. As such, 10nm Intel's CPUs are having clocks low enough to not be considered upgrades @ all, when compared to current offerings, despite being more energy efficient. Add to that, yield issues and the problem just gets monumental.

 

[Camm]

Those are monstrous, and very close to the Cinebench benchmarks I posted on a dual Epyc 64 core server.

 

[Vlada011]

Intel is on knees and wait to be finished with 3950X.
It's important that Linus sponsored by Intel learn people how to invest 550$ in obsolete CPU and 500$ in obsolete motherboards with PCI-E 3.0 Interface and stay on Gen 3.0 SSD in future years.
Linus make video to build RIG for self, I can bet he will use AMD machine in reality, not PCI-E 3.0.
It's not Graphic card improvement, it's important that everything high end in future will launched with Gen 4.0 support and Intel no platform to support that and show up with even more overclocked i9-9900K models because his factory sleep long time and didn't develope next gen processor to completely replace Intel Core.
People who are serious to stay with Intel no better choice then waiting. That's best what Intel can do from his customers, to get mercy and time to build better something because investing in Z390 is abnormal from moment when Gen 4.0 SSDs reach market and AMD X570.

2019 will not be finished and we will have SSD NVMe x2 faster then Samsung 970 PRO.
But on AMD platform. And I would not be surprised if NAVI after help of Microsoft and new drivers work even 5% better on PCI-E 4.0 then on PCI-E 3.0.

 

[champsilva]

Clearly LN2, judging from both the clocks and the voltage used in CPU-Z.

Compared to the original TR 1950X @ 5.2 GHz (same cores / threads: see pic below), that's over 33% increase for roughly 240 MHz higher clocks. I'd say that's very impressive.

View attachment 125986

Also, and dunno if anyone else noticed, the RAM is @ 4000 MHz CL 18, meaning the IF is supposed to be running @ half speed due to being higher than 3733.

4000mhz vs 2133mhz memory?

Same frequency and same core count, RAM is 3600mhz vs 4000mhz this will make a difference.

15.61%

 

[EarthDog]

RAM is 3600mhz vs 4000mhz this will make a difference.

Not much in Cine tho...

Intel is on knees and wait to be finished with 3950X.
It's important that Linus sponsored by Intel learn people how to invest 550$ in obsolete CPU and 500$ in obsolete motherboards with PCI-E 3.0 Interface and stay on Gen 3.0 SSD in future years.
Linus make video to build RIG for self, I can bet he will use AMD machine in reality, not PCI-E 3.0.
It's not Graphic card improvement, it's important that everything high end in future will launched with Gen 4.0 support and Intel no platform to support that and show up with even more overclocked i9-9900K models because his factory sleep long time and didn't develope next gen processor to completely replace Intel Core.
People who are serious to stay with Intel no better choice then waiting. That's best what Intel can do from his customers, to get mercy and time to build better something because investing in Z390 is abnormal from moment when Gen 4.0 SSDs reach market and AMD X570.

2019 will not be finished and we will have SSD NVMe x2 faster then Samsung 970 PRO.
But on AMD platform. And I would not be surprised if NAVI after help of Microsoft and new drivers work even 5% better on PCI-E 4.0 then on PCI-E 3.0.

Your ignorance precedes you...

 

[HTC]

4000mhz vs 2133mhz memory?

Same frequency and same core count, RAM is 3600mhz vs 4000mhz this will make a difference.

15.61%

RAM @ 3600 but with CL14: quite tighter than @ 4000 with CL18, although, with the new way IF works and IF would be working @ half speed due to the divider.

Dunno how that affects scores, if @ all.

 

[Gungar]

Loves to overclock? its 5.4 ghz @ 1.8 volt!

 

[lexluthermiester]

1.775v is pretty high.

I noticed that too! Holy crap that voltage!

 

[Vayra86]

Loves to overclock? its 5.4 ghz @ 1.8 volt!

Exactly. Not seeing the OC potential that much over previous Zen. I think this will be another round of XFR doing just a bit better unless you dig really deep and touch memory as well.