Author Topic: Do we have some informations about the 7950x performance ?  (Read 1666 times)

2022-10-03, 10:37:35

Sebastien

  • Active Users
  • **
  • Posts: 60
    • View Profile
Hi everyone,

Since the new ryzen 7950x has been released, I wonder what should we expect about this new CPU.

Does anybody have some informations on how it will run corona render regarding of the 5950x ?

I read here and there that the 7950x would be better than the 5950x over 20% or 30%, but what about corona render ? Should it be close to the performance of a threadripper 3970x ?

Also, do you know if the DDR5 will have a significant impact with corona ?

We don't have a lot of informations yet, but it'd be nice if we could gather them on a topic for the next few weeks or month.

Thanks !

2022-10-03, 11:12:24
Reply #1

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Francesco posted this graph from somewhere, so can't source it :- ) Too soon for such complex questions on how much different DDR5 kits will influence the performance, but I would say far less than the total power envelope which is very high, performance isn't the only thing that made this chip close to Threadripper :- ).

talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2022-10-03, 13:39:02
Reply #2

Sebastien

  • Active Users
  • **
  • Posts: 60
    • View Profile
Thanks for your answer ! :)

I didn't see this graph but it looks like it's similar to what I have heard (20% more efficient).

That's quite good for sure, but I wonder if it's worth the price because you need a mother board + DDR5 (expensive) + 7950x.

And you're right, I'm aware of that's too soon to have some reliable informations, I did this topic to gather informations. If someone buys it, it'd be super nice to have some feedbacks in the next months to know if it's worth it !

2022-10-03, 13:51:26
Reply #3

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website


That's quite good for sure, but I wonder if it's worth the price because you need a mother board + DDR5 (expensive) + 7950x.


Indeed, that's a very good question. The early adopter tax is more severe for CGI artists because we need 64-128GB for memory which compounds the price even further.
Possible alternative would be Intel 13900K which works with DDR4. Can't find any Corona benchmark yet.

Threadripper unfortunately is not only generation behind, but also the massive paywall of "pro" segmentation.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2022-10-03, 14:50:39
Reply #4

Sebastien

  • Active Users
  • **
  • Posts: 60
    • View Profile
Totally ! 13900k with DDR4 should be cheaper, that's why I wonder if the 7950x worth the price. The question is, will the DDR5 really better than the DDR4 with Corona render / 3D software in general. Hard to find answers on internet.

This new generation of CPU (13900K and 7950x) isn't that bad regarding the fact that we're getting close to a threadripper 3970x for more than half the price.

2022-10-03, 17:48:49
Reply #5

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
I would wager to say it won't make much difference for Corona because it's mainly improvement in bandwidth, not latency, unless you buy really expensive high clocked kit. We've already seen that even 8-channels of ThreadripperPro (which is 4 times the bandwidth of Ryzen therefore) makes absolutely zero difference, it's only the total latency that matters.

And even 7200 MT/s CL36 is not really improvement to 3600 MT/s CL18 in terms of latency for example, it's pretty much the same. So how much it does in practice will show only in benchmarks. It does influence the Infinity Bridge between dies, but we've also seen how little that can improve with 39xx Ryzen & Threadrippers when it comes to rendering. My 3990X is faster at 2133 MT/s than 3600 MT/s because the memory controller in IO is less stressed :- ) And I've never found it to bottleneck anything else, majority of benchmarks testing latency & bandwidth focus on niche applications (like PC game Factorio, etc..,).

I would honestly not expect any miracles from that particular aspect, it's just one of those "nice to have" products that are brand new and feel future-proof. But future-proofing is failed concept best to ignore.

I would wait for reviews of 13900K and particularly compare how both these chips run when limited to similar power enveloper (like 150W). These chips are super hard to cool because they are very tiny, so it doesn't matter how overkill watercooling or anything else is used because it's just hard to transfer heat away from such small surface area.

7950X can consume up to 250W, and has 3 small dies. 3990X consumes up to 280W but has 8 medium sized dies. That's a huge difference, so these chips will always run at 90C and throttle over longer rendering period.
I would probably consider 10 minute loop of CB23 to be better performance indicator than current Corona Benchmark.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2022-10-03, 17:54:56
Reply #6

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Another cool graph to performance scaling with power. Source:

talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2022-10-04, 00:40:35
Reply #7

lupaz

  • Active Users
  • **
  • Posts: 836
    • View Profile
so these chips will always run at 90C and throttle over longer rendering period.

This kindda sucks. Not to mention the noise from the fan, I imagine...Reason why I refuse to use a GPU render engine.

2022-10-04, 08:19:34
Reply #8

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
It's just something (temperature) to be ok with, it's how they're designed. Dense shrinking nodes is the silicon tech's improvement.

So choosing reasonable power limit (150W for example) to get 95perc. of the performance, but saving 100W of boost-power will take the temperature, power consumption and noise to very reasonable level.
It's just that from stock, all new CPUs & GPUs come in overkill setup to get that last absolute benchmark performance to win the reviews.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2022-10-06, 09:44:55
Reply #9

Sebastien

  • Active Users
  • **
  • Posts: 60
    • View Profile
Many thanks for your answers Juraj ! That's really interesting, even tough it's way beyond my knowledge :)

As far as I understand your graph, it means that above 175W, the performance of the 7950x looks like even. So what you advise is to see the same graph for the 13900K to know which is better around this cap limit.

I didn't know that above a certain power consommation, the efficiency of a CPU was very similar, that's indeed a very interesting information regarding all the problems these components have to cool them down (and the cost of electricity now !).

2022-10-06, 13:08:04
Reply #10

Nejc Kilar

  • Corona Team
  • Active Users
  • ****
  • Posts: 1132
    • View Profile
    • My personal website
AFAIK the thing is they typically go considerably past the efficiency curve with these parts because they need them to perform at their peak everywhere but most importantly benchmarks. Looking at it from a marketing POV you lose 5% of performance and your CPU might not be "the best" in that segment anymore.

With chips you can kind of think of it like this: For that last 10% of performance you need 30% more power (made up numbers but you get the idea with the ratio and how it isn't 1:1).

Because of that it seems like if you cap the 7950x to a more modest wattage mode you lose an arguably negligible amount of performance but potentially gain a lot on efficiency because of the above.

They do similar things with GPUs. You cap the power limit at say 80% and you lose ~5% performance.
Nejc Kilar | chaos-corona.com
Educational Content Creator | contact us

2022-10-06, 15:19:46
Reply #11

lupaz

  • Active Users
  • **
  • Posts: 836
    • View Profile

2022-10-06, 16:26:51
Reply #12

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Taking CB23 for example, the numbers are pretty brutal :- )

At 175W ( 64 perc. of peak from 275W), you get 97,5 perc. of said 275W performance. For the loss of 36perc. points from power consumption you only lose 2,5perc. of performance.
So I would consider 175W to be good hard limit for anyone. I personally would go as low as 150W.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika