Review AMD Ryzen 9 3950X Review: 16 Cores Muscles Into the Mainstream

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
876
394
19,360
"The 3950X comes with an AMD-defined 105W TDP just like the 3900X, but the four extra active cores require a more robust cooling solution. "

It looks like an extra 50mhz, maybe 100 with extremely good silicon in single core boost might be the reason (since Ryzen tends to boost variable based on cooling a little), not sure that's enough ... again, a high end noctua or even the 3900x stock cooler is adequate at stock.

If you chart out multi-core boosts, you'll see that the chip sustains higher nT clocks for a longer period of time with better cooling. Again, at stock settings. AMD's boost algorithms respond directly to cooling.
 
At least the 3950x does not require a BIOS update unlike the I9 9900ks.

So far, multi-core performance is about what I expected. I am a good bit surprised just how good gaming and single core performance.

However i am most surprised at power consumption. I get binning does a lot, but identical power consumption to a CPU with 4 cores less and lower clocks is impressive given the same architecture
 
  • Like
Reactions: RodroX and dorsai

Mr.Vegas

Honorable
Mar 13, 2019
68
13
10,535
This review did the opposite for me, it was counter productive, it actually pushed me to wait for the i9-10908XE.
In almost all benchmarks the 4.4Ghz last gen 18 core won over PBO 3950x, in the final page where it shows combined Geomean stat, it also wins by a lot, the new 18 core will do higher overclock so we can only go higher, sadly even with custom loop AMD wont overclock much.


So it goes like this, I almost ordered x570 board [ASUS X570 Crosshair VIII Hero (Wi-Fi) or MSI PRESTIGE X570 CREATION], even the lack of PCIe and the fact that i have 8 HDDs and capture card and optane PCIe card, that ill need to remove M.2 heat sinks and install M.2 to PCIe adapters to utilize my Optane.
x570 has 2 PCIe slots and one Untouchable that halfs the graphics slot to x8, so I can install one 2080ti + x4 Capture card, but what about my Optane?
I also have 4TB U.2 SSD that can be installed with adapter into M.2 or PCIe slot, but i wont be able to use it since most mobos have x2 m.2, and only asus boards have 8 SATA which i need for 2 for my SSDs and 6 for my HDDs.
So i found a solutions to all my problems, ill have to give up only NVMe drive [fastest but 2Tb] or use the M.2 slot for the 4TB U.2 SSD, i can live with it and my optane will be droped into m.2 slot using m.2 to PCIe adapter i already got them and tested, they work.
BUT after i finished reading the review i noticed that the 18 Intel with only 4.4Ghz overclock is beating 3950x in almost everything, and Intel has more PCIe lanes [solves all my issues] and if the truth is that the new 18 core can do 5Ghz on AIO [I plan on custom loop] anyway], and I already checked x299 mobos are cheaper then x570 [when compared by same manufacturer and mobo line], for me personally the Intel is kinda more interesting.
Of course ill wait for reviews to make final decision and I might even end up with TR3 24 core [It will cost me much more but because its new platform it has future, i wont need to replace my mobo until PCie gen5 comes since new TR3 has everything already]

P.S. I wish AMD had 8 more PCIe lanes on the chipset at least, even Gen 3 lanes, just more.
If next Nvidia GPU has PCIe Gen 4.0 then it means I will be able to use the second PCIe slot that shares bandwidth with the first one, since x8 Gen 4.0 = x16 Gen 3.0 and that im sure will be enough for 3080ti
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
876
394
19,360
We always do far more testing than we present, just due to time constraints. For instance, I ran the entire test regimen with the 3950X at stock settings with both the Corsair H115i cooler and the custom loop. Cinebench nT scores 9360 with the custom loop at stock settings, compared to 9111 with the open loop. Deltas with sT CB aren't as high, but you're looking at 543 vs. 538. Differences are more pronounced in bursty workloads, too. 13280 for open loop vs 12826 in VRMark, for instance.
 
If you chart out multi-core boosts, you'll see that the chip sustains higher nT clocks for a longer period of time with better cooling. Again, at stock settings. AMD's boost algorithms respond directly to cooling.
Right, but that's true of every 3rd Gen Ryzen CPU... and it's true of every modern GPU and has been for some time.

The statement in your article is that the extra 4 active cores on the 3950X require more robust cooling, yet the data you present shows the 16 active cores drawing the same power as the lower binned 12 core part. I can't see how both statements can be true.

I appreciate the additional data you just posted, however wouldn't you expect to get similar jumps in performance when running the 3900X under the beefier custom loop? My understanding is from all the data I've seen so far that the better binned 3950X runs on lower voltages across its frequency and boost range which (and I acknowledge I'm generalising here) tends to completely offset the power draw from the additional four cores. Am I mistaken here?

Isn't it more accurate to state that the 3950X, like all 3rd gen Ryzen CPUs, will boost higher and for longer with better cooling. However with 3950X, AMD have been able to offset the added power draw of the additional cores through binning and lower voltages?
 
  • Like
Reactions: RodroX and dorsai
That adds up and uses more power, one review I read showed the 3950X OCed to 4.3GHz was hitting 363W, just shy of the 9980XE, in power draw under load.

https://hothardware.com/reviews/amd-ryzen-9-3950x-zen-2-review?page=6

"Tested at the Outlet"

It's worth pointing out that they are measuring full system power consumption there, not just the processor itself. That includes all other components in the system, along with the efficiency of the power supply, which introduces a lot of variables into the equation.

Even at stock clocks though, that review shows the 3950X often outperforming the 9980XE, while drawing 130 watts less from the wall. Sure, pushing the processor to its overclocking limits might drive total system power draw up to those levels, but the same could be said for the 9980XE. In fact, that site's 9980XE review shows power draw going up to over 500 watts with that processor overclocked, around 140 watts higher than the overclocked 3950X...

https://hothardware.com/reviews/intel-core-i9-9980xe-review?page=6
 
And here we see the reason why intel needs 10nm sooner rather than later.
I'm not convinced 10nm offers real benefits for Intel in the Desktop market. It brings clear density and power improvements, which will offer tangible gains for the mobile and server spaces. It really appears, however, that peak clockspeeds on 10nm are going to be lower than Intel have been able to squeeze out of their mature 14nm process. 10nm desktop parts then, may well require Intel to make sacrifices to their only clear advantage over AMD at present... frequency at the top end.
 

PaulAlcorn

Managing Editor: News and Emerging Technology
Editor
Feb 24, 2015
876
394
19,360
Right, but that's true of every 3rd Gen Ryzen CPU... and it's true of every modern GPU and has been for some time.

The statement in your article is that the extra 4 active cores on the 3950X require more robust cooling, yet the data you present shows the 16 active cores drawing the same power as the lower binned 12 core part. I can't see how both statements can be true.

I appreciate the additional data you just posted, however wouldn't you expect to get similar jumps in performance when running the 3900X under the beefier custom loop? My understanding is from all the data I've seen so far that the better binned 3950X runs on lower voltages across its frequency and boost range which (and I acknowledge I'm generalising here) tends to completely offset the power draw from the additional four cores. Am I mistaken here?

Isn't it more accurate to state that the 3950X, like all 3rd gen Ryzen CPUs, will boost higher and for longer with better cooling. However with 3950X, AMD have been able to offset the added power draw of the additional cores through binning and lower voltages?
The chip draws more power at mid-loadings than it does under full load, as mentioned in the article. Look no further than AMD for proof that the additional cores equate to a higher cooling requirement: The company says the minimum cooling solution is a 280mm radiator, yet doesn't set nearly as high of a cooling requirement for the 12-core model. Yes, you can cool it with a lesser solution, but throttle-fests are no bueno, and due to AMD's boost algorithms these micro-throttles aren't readily apparent to the user, and they aren't exposed through external hooks to software utilities. Instead, the chip simply runs slower if there isn't enough thermal dissipation capacity, adjusting its performance to stay under a full throttle triggered by excessive temps.
 
  • Like
Reactions: joeblowsmynose
Nov 14, 2019
1
0
10
I'm hard to believe in Cinebench 20 single-core stock 3900x able to achieve 532 points! (must be golden chip?)

I got the 3900x on July 15 and already been a lot of bios updates (MSI X570 Creation). I tried all the bios, even the new beta bios with 1.0.0.4b, my 3900x with PBO only able to get 509 max in C20, the highest single-core clock recorded in Ryzen Master was 4.394. In multi-core C20, all core boost was 4.09GHz with PBO (7035 points), 4.05Ghz without PBO (6992 points). What a joke compared to your test 3900x.

However, my 3900x able to do manual all core 4.3Ghz with 1.331v, in the C20 multi-core test achieved max 7560 points, and single-core 512 points. Also, with 1.33v, my 3900x run much cooler around 70-71C.

Cooler: Deepcool Gamerstorm Castle 360 RGB V2
PSU: EVGA SuperNOVA 1200 P2 1200W 80+ Platinum
RAM: 64GB (16x4) 3200Mhz cl16
 
This review did the opposite for me, it was counter productive, it actually pushed me to wait for the i9-10908XE.
In almost all benchmarks the 4.4Ghz last gen 18 core won over PBO 3950x, in the final page where it shows combined Geomean stat, it also wins by a lot, the new 18 core will do higher overclock so we can only go higher, sadly even with custom loop AMD wont overclock much.


So it goes like this, I almost ordered x570 board [ASUS X570 Crosshair VIII Hero (Wi-Fi) or MSI PRESTIGE X570 CREATION], even the lack of PCIe and the fact that i have 8 HDDs and capture card and optane PCIe card, that ill need to remove M.2 heat sinks and install M.2 to PCIe adapters to utilize my Optane.
x570 has 2 PCIe slots and one Untouchable that halfs the graphics slot to x8, so I can install one 2080ti + x4 Capture card, but what about my Optane?
I also have 4TB U.2 SSD that can be installed with adapter into M.2 or PCIe slot, but i wont be able to use it since most mobos have x2 m.2, and only asus boards have 8 SATA which i need for 2 for my SSDs and 6 for my HDDs.
So i found a solutions to all my problems, ill have to give up only NVMe drive [fastest but 2Tb] or use the M.2 slot for the 4TB U.2 SSD, i can live with it and my optane will be droped into m.2 slot using m.2 to PCIe adapter i already got them and tested, they work.
BUT after i finished reading the review i noticed that the 18 Intel with only 4.4Ghz overclock is beating 3950x in almost everything, and Intel has more PCIe lanes [solves all my issues] and if the truth is that the new 18 core can do 5Ghz on AIO [I plan on custom loop] anyway], and I already checked x299 mobos are cheaper then x570 [when compared by same manufacturer and mobo line], for me personally the Intel is kinda more interesting.
Of course ill wait for reviews to make final decision and I might even end up with TR3 24 core [It will cost me much more but because its new platform it has future, i wont need to replace my mobo until PCie gen5 comes since new TR3 has everything already]

P.S. I wish AMD had 8 more PCIe lanes on the chipset at least, even Gen 3 lanes, just more.
If next Nvidia GPU has PCIe Gen 4.0 then it means I will be able to use the second PCIe slot that shares bandwidth with the first one, since x8 Gen 4.0 = x16 Gen 3.0 and that im sure will be enough for 3080ti

If you have a real use case that requires more PCIe lanes than 3950x will provide it would be foolish to not wait another month and see what Threadripper brings. There's a reason AMD is still indicating 3950x is a mainstream cpu...Threadripper will remain at the top for AMDs HEDT space and I would be suprised if the i9-10908XE is faster.
 

AtotehZ

Distinguished
Nov 23, 2008
403
13
18,815
I know this would be a lot more work, but showing games in 1440p and 4k is relevant. I'm aware that 1080p limits the gpu bottleneck, but seeing how often one cpu over the other even matters is not without meaning for gaming.

Currently im looking at an i9-9900k being 5-15% faster at 1080p. That's a deciding factor when people go to buy a new cpu. If the advantage drops to below 5% already at 1440p, people may choose the 3900x or 3950x instead. That way they'll have an overall faster system.
 
Thank you @TCA_ChinChin, I was literally typing the following the reply when you posted your message!

I'm with @joeblowsmynose: I don't believe the claim that the 3950X requires a more robust cooling solution is supported by the data. Toms, Gamers Nexus and Hardware Unboxed all found the 3950X matching or even coming in under the 3900X for sustained power draw. Sure, cooling requirements don't correlate perfectly to power draw, but surely no one is arguing that a 3950X pulling 145W somehow requires more cooling than a 3900X pulling the same power?

In fact, I don't want to sound fan-boyish here, but isn't it pretty remarkable that AMD have managed to get 33% more cores running without measurably increasing the power draw... all with only a slight drop to the base clock? Any increase in core count on the same process node without having to resort to ramping power or gutting clock speeds is laudable progress in my book. Correct me if I'm wrong, but the 3900X was never presented as a power hog, in fact, from my memory it was generally deemed an efficient processor for its target market. AMD hit the same power draw with 4 extra cores... almost exclusively (as far as I can tell) from better binning. The 3950X is straight up a more efficient and superior product.

OC it, and of course, all efficiency goes out the window. On top of that, performance gains through OCing are minimal at best. I agree that those who value the experience of overclocking and tweaking won't be particularly excited by the 3950X... fair criticism. But criticising power draw, efficiency and cooling requirements??? That seems like a criticism we were all expecting to be able to level, but where the 3950X has actually and quite impressively surpassed expectations and should have, IMHO, been praised for... not criticised.

Again it is AMD themselves stating that it requires more robust cooling, not the reviewers. Go to the link I posted up top from AMDs site. It literally states that they recommend liquid cooling.

And yes even under the same power draw there are 4 more active cores which means more heat and heat in areas that the 3900X may not have.

I would rather state that instead of the 3950X being superior the 3900X is inferior. They both use two 7nm dies but the 3900X obviously has 4 cores turned off. If it and the 3950X are using similar power then the 3900X would probably be more inefficient. Best way to truly test it would be to take both and set them to the same clock speed and see what their power draw is and their heat output as well.
 
  • Like
Reactions: Soaptrail

joeblowsmynose

Distinguished
If you chart out multi-core boosts, you'll see that the chip sustains higher nT clocks for a longer period of time with better cooling. Again, at stock settings. AMD's boost algorithms respond directly to cooling.

Ok that's somewhat valid.

Maybe you could do a review on the impact of different levels of cooling across a few CPUs -- 9900k/ks, 3900x 3950x, 9980xe?

Lower end (but still reasonable) air cooler, high end air cooler, lower end aio, high end 360 aio, and a custom loop with your dual 360s :)

That would actually make a pretty great investigative article that I think would be useful to many people.
 
Last edited:

joeblowsmynose

Distinguished
...
BUT after i finished reading the review i noticed that the 18 Intel with only 4.4Ghz overclock is beating 3950x in almost everything, and Intel has more PCIe lanes [solves all my issues] and if the truth is that the new 18 core can do 5Ghz on AIO

5.0 on 18 cores just able to boot windows and run a 20 second benchmark or a CPU-z capture before the CPU overhaats and throttles? ... or sustained 100% workload on 18 cores for an hour? I assume that if 18 cores interests you, so do heavy workloads.

Massive difference between those two - also note the 9980XE pulls ~400w (at the socket! 660w at the wall) at 4.6ghz. While I do suspect that the 10980xe will be somewhat more efficient (just a better bin), we're still probably looking at 500-600w or more at anything close to 5.0 on all cores loaded; again, probably at the socket.

If I were you, I'd wait for proper reviews ...

https://www.legitreviews.com/intel-core-i9-9980xe-extreme-edition-processor-review_209016/8
"Overclocking the 9980XE on the ASUS PRIME X299 Deluxe-II motherboard was simple and we were able to get it up to 4600MHz. Benchmarks with all cores set to 4600MHz were impressive as we went from around 3750 to 4500 on the Cinebench CPU test that uses all available cores. The power consumption in Cinebench R15 went from 364 Watts on the 9980XE in stock form to 658 Watts with the overclock. This systems idles at 101 Watts, so just running the CPU at full load is pulling over 650 Watts from the wall. "
 
Last edited:

joeblowsmynose

Distinguished
... AMD hit the same power draw with 4 extra cores... almost exclusively (as far as I can tell) from better binning. The 3950X is straight up a more efficient and superior product.


I am also impressed that they are getting chiplets of this of quality in comparison to the other skus - as you said, not that they aren't efficient, just that the difference in the 3950x is impressive.

I wonder ... if AMD had the luxury of refining even this node for 5 years, what types of chiplets they would be able to get out after working on refinement for clock speeds and efficiency throughout that time?

Intel's 10nm delay was the best thing that ever happened to their 14nm node, if you want to see the lemonade from the lemons.

I see the speed in which AMD is moving nodes and adjusting architecture as perhaps losing out on the ability to really refine properly - it might be part of their strategy though -- run as fast as you can for as long as you can, and when you face obstacles to that, then refine the strides and steps if needed to keep up some speed. It certainly has worked well for Intel's 14nm. I'm actually impressed they still can get improvements.
 
Last edited:

TJ Hooker

Titan
Ambassador
You forget it has 4 additional cores which mean additional links to the I/O chips as well compared to the 3900X. That adds up and uses more power, one review I read showed the 3950X OCed to 4.3GHz was hitting 363W, just shy of the 9980XE, in power draw under load.
I really don't think this is true. The interface between each CCD and the I/O die is the same, regardless how many cores are enabled in the CCD. And we know both CCXs are at least partially enabled in both of the CCDs for the 3900X because it has 64 MB of L3 cache.
 
"Limited overclocking headroom"
Okay, should this REALLY be a con for this cpu considering its use case(affordable professional level cpu, sans the extra pcie lanes)?

No it shouldn't be on the Cons. In fact that AMD stated out loud that Ryzen 3xxx along with PBO already gives you the best version of the OC CPU out of the box. So whats the point?

Also, one thing I would love to know since there are no temps charts, What was the max temps when runing those test for the Intel Core i9 9980XE? Cause you said the Ryzen 9 3950X gets really hot and you need better cooling than what AMD suggest to get the greateast performance, so What about the others CPU like that Intel Extreme Edition one?

Did you also used the same Corsair 28mm AIO to cool all the CPUs used for the many benchmarks included? (sorry if this is wroten in the article and I didn't see it)
 
Last edited: