Monday, August 6th 2018

AMD Ryzen Threadripper 2990WX Cinebench Numbers Out

AMD France blurted out the Cinebench R15 score of the upcoming Ryzen Threadripper 2990WX 32-core/64-thread HEDT processor. The web-design team of AMD's French website inadvertently posted Cinebench R15 numbers of the 2990WX, along with their own tested numbers of Intel's current flagship, the Core i9-7980XE. Cinebench is AMD's favorite multi-threaded benchmark, and it should come as no surprise that its new 32-core/64-thread 2990WX absolutely smashes the 18-core/36-thread i9-7980XE.

The Ryzen Threadripper 2990WX has an nT (multi-threaded) score of 5,099 points, compared to 3,355 points scored by the i9-7980XE. The comparison saw memory (4x 8 GB DDR4-3200), graphics (NVIDIA GTX 1080), and storage (Samsung 850 Pro) constant between the two machines. The Intel machine featured a GIGABYTE X299 Aorus Gaming 9 motherboard, while the AMD machine used an unnamed socket TR4 motherboard. CPU cooling was not mentioned. AMD was, of course, quick to redact the web-page, but the Internet never forgets.
Source: Guru3D
Add your own comment

24 Comments on AMD Ryzen Threadripper 2990WX Cinebench Numbers Out

#1
Prima.Vera
66% more performance for 56% in Cores count. And price probably is cheaper too. A 100% win-win-win situation for AMD. Nice.
Posted on Reply
#2
noname00
Prima.Vera said:
66% more performance for 56% in Cores count. And price probably is cheaper too. A 100% win-win-win situation for AMD. Nice.
I don't understand what you are comparing.

(5099/3355) * 100 ~= 152 ==> 52% performance increase
(32/18) * 100 ~= 178 ==> 78% moar cores.
Posted on Reply
#3
Fierce Guppy
Making things very simple, just dividing the score by the core count shows the Intel chip to be the better performer per core.

AMD -- 5099/32=159 per core
Intel -- 3355/18=186 per core

However, that doesn't mean much until the price is factored in.
Posted on Reply
#4
bubbly1724
52% performance for 52% increase in TDP? Sounds about right.
Posted on Reply
#5
DeathtoGnomes
Prima.Vera said:
66% more performance for 56% in Cores count. And price probably is cheaper too. A 100% win-win-win situation for AMD. Nice.
its been noted that the price for 2990's will be $1835 at launch.
Posted on Reply
#6
R0H1T
Fierce Guppy said:
Making things very simple, just dividing the score by the core count shows the Intel chip to be the better performer per core.

AMD -- 5099/32=159 per core
Intel -- 3355/18=186 per core

However, that doesn't mean much until the price is factored in.
There's also the clockspeeds and TDP to consider, which I'm sure will be higher for Intel, per CB point.
Posted on Reply
#7
Vya Domus
Fierce Guppy said:
Making things very simple, just dividing the score by the core count shows the Intel chip to be the better performer per core.
Performance doesn't scale linearly.
Posted on Reply
#9
Caring1
noname00 said:
I don't understand what you are comparing.

(5099/3355) * 100 ~= 152 ==> 52% performance increase
(32/18) * 100 ~= 178 ==> 78% moar cores.
So why confuse your little mind with math then?
First figure out what you are looking at before doing the sums.

Vya Domus said:
Performance doesn't scale linearly.
The Tr has a 100% increase in cores for only a 69.96% increase in CB score.
Posted on Reply
#10
R0H1T
Caring1 said:
So why confuse your little mind with math then?
First figure out what you are looking at before doing the sums.


The Tr has a 100% increase in cores for only a 69.96% increase in CB score.
Cinebench also doesn't scale linearly with so many cores, it does with clock speeds though.
Posted on Reply
#11
RejZoR
Also be aware that AMD's implementation of SMT is actually more efficient than Intel's. Which probably scales even further with so many cores doing it.
Posted on Reply
#12
noname00
Caring1 said:
So why confuse your little mind with math then?
First figure out what you are looking at before doing the sums.
The first comments has some percentages, I just wanted to point out that those percentages are not related to the performance and core count difference of the two CPUs in the article. I know what I'm looking at, Prima.Vera is the one that does not know what he's looking at. And, apparently, neither do you.
Posted on Reply
#13
efikkan
PerfectWave said:
Does intel have a 32 core cpu?
No, up to 28 cores, and that's not going to change any time soon.

RejZoR said:
Also be aware that AMD's implementation of SMT is actually more efficient than Intel's. Which probably scales even further with so many cores doing it.
Even if so, adding more cores will not increase SMT efficiency per core, and only decrease the efficiency of scheduling for the OS.
Posted on Reply
#15
Vya Domus
R0H1T said:
Cinebench also doesn't scale linearly with so many cores, it does with clock speeds though.
Nothing ever scales linearly , not just Cinebench.
Posted on Reply
#16
R0H1T
Vya Domus said:
Nothing ever scales linearly , not just Cinebench.
Does that include going from one core to two or two to four cores, in other synthetic benches as well?
What about memory benches or GPU tests? The blanket statement might be accurate but what's the reason for it?
Posted on Reply
#17
slehmann
R0H1T said:
Does that include going from one core to two or two to four cores, in other synthetic benches as well?
What about memory benches or GPU tests? The blanket statement might be accurate but what's the reason for it?
Thats true for all operations as there is always some management overhead for the actual task to compute.
And with increasing units, who do the compute work (cores, shader, whatever), the management overhead rises.

In general tasks are never only compute. You always need to feed the units with data (=IO) and everything outside a CPU is really really really slow compared to the internal caches (1-3rd).
Posted on Reply
#18
Vya Domus
R0H1T said:
Does that include going from one core to two or two to four cores, in other synthetic benches as well?
What about memory benches or GPU tests? The blanket statement might be accurate but what's the reason for it?
It has to do with the very concept of parallel computing. No program can be scaled up to run on any number of processing nodes with 100% efficiency and no overhead. First of all no matter how highly parallelized said program may be , there are always chunks of it which can only be computed in a serial fashion and therefore performance will never scale 100%. Not to mention that the hardware itself is subject to all sorts of upper bounds in terms of processing power , just because you double the core count that may not necessarily mean you doubled up all resources in that specific computing system.


R0H1T said:
The blanket statement
You may think it's a blanket statement because you are not well versed on the matter, but research has already been done a long time ago in order to figure out these things.

https://en.wikipedia.org/wiki/Amdahl's_law

https://en.wikipedia.org/wiki/Gustafson's_law
Posted on Reply
#19
Effting
What people don't get is that this CPU is performance limited by TDP and by memory channel. Simply put, if all dies could gol full speed at 110W TDP (same as 2700X) and evey die had full acess to its two memory channel (wich is not the case, only 4 channels are availible fot the entire chip) the performace would have been scaled more linearly. But it just wont happen because of the aforementioned constrains. But even then its the most powerfull chip ever made. Not to mention it's price tag, it's backwards compatibility, cooler that comes with it...
Posted on Reply
#20
Xaled
Getting 52% more raw performance for 53$ less is what matters for me.
Posted on Reply
#21
GhostRyder
Sounds like a good buy to me if you need the cores or can make use of them.
Posted on Reply
#23
Bones
noname00 said:
I don't understand what you are comparing.

(5099/3355) * 100 ~= 152 ==> 52% performance increase
(32/18) * 100 ~= 178 ==> 78% moar cores.
noname00 said:
The first comments has some percentages, I just wanted to point out that those percentages are not related to the performance and core count difference of the two CPUs in the article. I know what I'm looking at, Prima.Vera is the one that does not know what he's looking at. And, apparently, neither do you.
So.... You say you don't understand it yet you know what you are looking at.
Good job of showing what you don't know and lashing out too.

Simply put, if it gives more performance with a given bench, application, gaming in general or whatever else for less then it's worth it - Well worth it in fact.

I'd grab one with the same or even a little less performance because the price is right, doesn't make sense to spend more for just 2-5% worth of difference.
Speaking of price here's how much the 7980XE is going for ATM, this being the price of it ATM from Newegg: https://www.newegg.com/Product/Product.aspx?Item=N82E16819117836
With the stated price at launch of the 2990WX (If correct - $1835) then it's a no-brainer.

Also note the extra threads will offset whatever advantage the 7980XE may have in certain tasks too at the very least. Since you only get 2 threads per core with either one the 2990WX clearly has the advantage in multi-threaded applications.
Posted on Reply
#24
stazorm
AMD making progress is a good thing.
I might get that 1950X on sale.
Posted on Reply
Add your own comment