Thursday, June 13th 2019

AMD's Upcoming $750 Ryzen 9 3950X (16C, 32T) Shown Beating Intel's $2,000 i9-9980XE (18C, 36T)

When we said AMD was readying a presentation on their Ryzen 9 3950X CPUs to awe crowds at E3, we weren't thinking of something of this magnitude. But apparently, it's true: a Geekbench test result has shown AMD's $750, 16 core, 32 thread Ryzen 9 9 3950X beating Intel's 18 core, 36 thread $2,000 i9-9980XE monster. Now, you may be thinking: ok, it beat it because of AMD's announced 4.7 GHz boost, and did so only on single threaded performance, obviously... but you would be wrong.

The Geekbench scores show AMD's Ryzen 9 3950X delivering 5,868 points in single, and 61,072 points in multicore workloads. Intel's i9-9980XE, on the other hand, scores just 5,391 single core, and 46,876 multicore points (on average and at stock clocks of 3,000 MHz base and 3,400 MHz boost). This is an incredible performance difference (particularly in the multicore score), and was apparently done with an engineering sample for AMD's upcoming chip that didn't even run at its announced 4.3 GHz base and 4.7 GHz boost clocks, but at 3.3 GHz and 4.3 GHz respectively. AMD's 105 W TDP, 16 core chip beats Intel's 185 W TDP, 18 core one... Where has the world come? Take the usual dosage of NaCl, and let's keep things in perspective - even if AMD's Ryzen 9 3950X equals, and doesn't beat, Intel's i9-9980XE, it's still a huge win for the red company. Almost as big a win as that huge stone on Lisa's hand.
Sources: Tom's Hardware, Intel i9-9980XE GeekBench score example, GeekBench Ryzen 9 3950X test result
Add your own comment

70 Comments on AMD's Upcoming $750 Ryzen 9 3950X (16C, 32T) Shown Beating Intel's $2,000 i9-9980XE (18C, 36T)

#2
diatribe
I hesitant to believe this just yet. The 18 core Intel should be a least negligible faster than the AMD 16 core counterpart.
Posted on Reply
#3
Konceptz
*awaiting Intel fanboy defense*
Posted on Reply
#4
Agent_D
Healthy doses of skepticism, but exciting if true; AMD I've been waiting for you to come back.
Posted on Reply
#5
Vya Domus
Geekbench does not scale well with cores. That being said, the 3950X should technically be faster per core by a little bit.
Posted on Reply
#6
HwGeek
diatribe, post: 4064448, member: 93915"
I hesitant to believe this just yet. The 18 core Intel should be a least negligible faster than the AMD 16 core counterpart.
Why?
9980XE all core boost = 3.8Ghz Non AVX or 3.3Ghz AVX, so with the IPC advantage for the 3950X- it should only run at ~3.8Ghz all core on Non AVX - Magic right? :).
18C is only 12.5% more cores then 16C - same as the IPC advantage of 3950X + AMD's SMT give better performance uplift.
I do believe that Intel will bring back 20C,22C options to X299- same like it was on X99.
Posted on Reply
#7
mouacyk
Konceptz, post: 4064450, member: 59570"
*awaiting Intel fanboy defense*
come get your popcorn
Posted on Reply
#8
windwhirl
AMD's multiyear bets are paying off
Posted on Reply
#9
trparky
I'm not saying that I'm a fanboy but I don't consider GeekBench scores to be very valid. With that said, I'm really not surprised that AMD is kicking Intel's ass; they have after all been sleeping at the switch for so long and more than likely all the exploit fixes have eaten into Intel's performance lead.
Posted on Reply
#10
Wavetrex
HwGeek, post: 4064456, member: 185585"
I do believe that Intel will bring back 20C,22C options to X299- same like it was on X99.
They can't.

HCC tops up at those 18 cores.
For anything higher they need XCC



Those dies are absolutely ginormous, and do not fit on the LGA 2066, they need LGA 3647, meaning ... new boards.
Yep, that's the monster Xeon W-3175X and those "sell your kidney" ultra-expensive LGA 3647 boards...

They aren't going to compete with AMD's tiny chiplets with that ...
Posted on Reply
#11
JB_Gamer
@Raevenlord - I believe there's an error in the article, it says "...announced 4.3 GHz base a...", it should be 3.5 GHz.
Posted on Reply
#12
HTC
This benchmark isn't so good for this kind of comparison, IMO: not taxing enough for this many cores / threads.

Perhaps a better comparison would be something like CB R20 (R15 may be too fast as well to properly tax these CPUs), with MCE and XFR / PBO enabled, matched RAM and a very good CPU cooler, and then let them perform with a window of HWinfo opened with the relevant sensor tab, in order to find out, in real time, max clocks / sustained clocks / power usage during the bench run.

Any benchmark that scales well with cores and isn't "lightning fast" should work as well.
Posted on Reply
#13
JB_Gamer
HTC, post: 4064508, member: 51238"
This benchmark isn't so good for this kind of comparison, IMO: not taxing enough for this many cores / threads.

Perhaps a better comparison would be something like CB R20 (R15 may be too fast as well to properly tax these CPUs), with MCE and XFR / PBO enabled, matched RAM and a very good CPU cooler, and then let them perform with a window of HWinfo opened with the relevant sensor tab, in order to find out, in real time, max clocks / sustained clocks / power usage during the bench run.

Any benchmark that scales well with cores and isn't "lightning fast" should work as well.
Maybe the result will be far worse for the Intel cpu as the Amd cpu was not running at its announced best boost?
Posted on Reply
#14
efikkan
Geekbench and Cinebench, the kings of benchmarks… :rolleyes:
Posted on Reply
#15
Metroid
HTC, post: 4064508, member: 51238"
This benchmark isn't so good for this kind of comparison, IMO: not taxing enough for this many cores / threads.
I guess some people want the hype to continue hehe

Lisa, Monday e3 10 June, as amazed as it was, published very dignified charts and in those charts 9900k won x 3900x in some games. So even Lisa, AMD CEO is cautious about the hype.
Posted on Reply
#16
HwGeek
Wavetrex, post: 4064495, member: 182738"
They can't.

HCC tops up at those 18 cores.
For anything higher they need XCC



Those dies are absolutely ginormous, and do not fit on the LGA 2066, they need LGA 3647, meaning ... new boards.
Yep, that's the monster Xeon W-3175X and those "sell your kidney" ultra-expensive LGA 3647 boards...

They aren't going to compete with AMD's tiny chiplets with that ...
Then they are truly F**ed :) .
Looks like they are going for the large socket, they offer 8~28C and actually the prices looking good compared to current EPYC. :
This board looks nice at $499:
https://www.newegg.com/p/N82E16813183686?Description=3647%20&cm_re=3647-_-13-183-686-_-Product
Posted on Reply
#17
HTC
JB_Gamer, post: 4064514, member: 173137"
Maybe the result will be far worse for the Intel cpu as the Amd cpu was not running at its announced best boost?
Question: What were the conditions of the test, from both camps?

- cooler(s) used?
- RAM used and it's parameters?
- MCE and XFR / PBO enabled?
- mitigations used?

I'm proposing best possible scenario under non-manual OC, meaning very good CPU cooling, RAM @ matching parameters, MCE and XFR / PBO enabled and mitigations disabled.
Posted on Reply
#18
holyprof
As already mentioned on this thread, geekbench isn't exactly a measure of real world performance. But the AMD fan (not fanboy) in me wants this to be at least partly true, so I can re-edit my first PC's composition - AMD processor, Nvidia (nVidia at the time) graphics.

Not sure if I'm getting the expectations right, but drooling at the 3700 or 3600X to replace my current i5-4690K (workload is 50% gaming, 20% CAD work, 30% image and video editing, so I need a CPU that performs well in all areas).
Posted on Reply
#19
dicktracy
Is it really that hard to use an actual real world application to demonstrate actual CPU performance? I rather they use Blender as a standard test to measure CPU performance (which is used by Intel and AMD) than Cinebench and Geekbench and whatever synthetic suites out there. It’s as relevant as testing games in 480p low settings with dual Titan RTXs. I mean Intel might as well do that to combat against AMD’s obsession with Cinebench and you’ll probably see a double standard from the fanboys claiming how Intel is bad AMD is good.
Posted on Reply
#20
JB_Gamer
HTC, post: 4064529, member: 51238"
Question: What were the conditions of the test, from both camps?

- cooler(s) used?
- RAM used and it's parameters?
- MCE and XFR / PBO enabled?
- mitigations used?

I'm proposing best possible scenario under non-manual OC, meaning very good CPU cooling, RAM @ matching parameters, MCE and XFR / PBO enabled and mitigations disabled.
Yes I agree, all those considerstions being taken.
Posted on Reply
#21
bszalman
Maybe:
Intel per core score = 46876 / 18 = 2604.2

%Different clock = ((4300 - 3400) / 3400) * 100 = 26.5%

AMD score = (16*per core + 26.5% value) + 15% IPC claims
= 41667.2 + 11041.8 + 7906.4
= 60615.4
Posted on Reply
#22
phill
I'm just waiting and really looking forward to the reviews. All these 'leaked' benchmarks are great but I prefer the hard evidence instead :)
Posted on Reply
#23
natr0n
We just need more multi-threaded programs/apps/games to really enjoy all these cores.
Posted on Reply
#24
ArchStupid
You thinking that constitutes any sort of evidence and warrants defense makes you an AMD fanboy and just as bad.
Posted on Reply
#25
matar
Great news as this 3950x is my new CPU
Posted on Reply
Add your own comment