AMD Ryzen Threadripper 1950X and 1920X CPU Review @ [H]

Unigine Heaven 640x480! Love it, that's something probably neither AMD nor Intel ever optimized their CPUs for, so fun to see how they perform on an unexpected workload :D
Looks like a great test to see how latency affects the outcome, also memory speeds.
 
So I suspect, based on the one 3200 Cinebench score that Intel leaked for the i9-7960 (16c) that what it's going to come down to is if you are hardware price constrained, the AMD's TR 1950x is possibly a better way to go for about $700 less in the CPU (everything else being roughly the same) and if you're software price/time constrained (software is MUCH more expensive or that 10% or so speed increase pays for the hardware difference) then Intel's i9-7960 or i9-7980 will be the better way to go. And if you're not in the market for stupid expensive, workstation class, "seriously wife/husband/gf/bf/sig other this $1k+ CPU was ABSOLUTELY NEEDED FOR ME TO SURVIVE" level of computer, we're going to FINALLY be back into a situation where we, as enthusiasts, will actually have a value to our friends and families in recommending hardware because AMD actually has CPUs out that can challenge anything Intel is currently shipping in the "retail" space (and pokes at the server/workstation space pretty well too), meaning once again, we'll actually have to start paying attention to benchmarks, usage scenarios, price, weird driver conflicts, and motion of the planets and stars to find the best solutions again.

AMD gets a well deserved "Attaboy" from me for this. Now let's see what Intel can come up with/drop price on.

The real question is: Will EITHER of these (1950x or 7900x) play Crysis?
(Seriously, someone had to ask)
 
Not huge gains on o/c, even when using a multi thread app. Curious of thoughts on that. Is it because the app is not using all threads and 4c boost and XFR are offsetting the gains?
 
Unigine 640x480 was on lowest or maxxed out settings? I got 352 and ~240 on my 4790K respectively w/1080FE
 
Unigine 640x480 was on lowest or maxxed out settings? I got 352 and ~240 on my 4790K respectively w/1080FE
These are very much "benchmarks." These are good in helping us understand how well CPUs are at performing calculations in 3D gaming engines. These benchmarks in no way represent real-world gameplay. These are all run at very low resolutions to try our best to remove the video card as a bottleneck.

LOWEST SETTINGS....
 
Does anyone know what the max RAM speed is for the X399 / Threadripper platform? I notice in the review the AMD system only used 3200mhz... I heard there is an advantage for the Ryzen platform to use the fastest RAM possible but not sure if that is even substantiated, or if that applies to TR as well.

PS: Great review!!
 
People still dragging up the Segfault issue for the small fraction of a small fraction of Linux users in other threads is quite funny, making it out as if it's not being addressed and that it's a huge issue, while conveniently side stepping that 2600k and other CPUs also suffer from this bug. It seems it has been solved now, as people have reported RMA CPUs fixing the problem, whilst others have not replicated the fault on some existing CPUs, while others have temporarily resolved by disabling ASLR. At least AMD isn't pulling a L2 cache patch screw job..
Theory is early batches had some issue, later batches (RMA replacements) including TR don't. So if we can get more data on processor dates from anyone affected, this could help narrow things down.

Threadripper and Epyc CPUs appear unaffected also, so spreading that BS in this thread, is total FUD shilling at this point.

On the more positive front though, my latest CPU from my second RMA has now survived over 24 hours of compiling without failure. That includes about 14 hours of compiling gcc with make -j16, 8 hours of two parallel compiles of gcc with make -j9 each (on both cases on a tmpfs mounted drive) and 4+ hours now of 4 parallel compiles with -j4 in a zram mounted compressed ramdisk.
https://community.amd.com/thread/215773?start=630&tstart=0
 
Hi, does anyone know which of the X399 boards support ECC ram in ECC mode? I'd like with build a server with TR. There's been some posts that suggests ECC ram can be used, but as non-ECC. Thanks!
 
Hi, does anyone know which of the X399 boards support ECC ram in ECC mode? I'd like with build a server with TR. There's been some posts that suggests ECC ram can be used, but as non-ECC. Thanks!

all of them should support ECC in ECC mode as well as non-ecc.
 
Hi, does anyone know which of the X399 boards support ECC ram in ECC mode? I'd like with build a server with TR. There's been some posts that suggests ECC ram can be used, but as non-ECC. Thanks!

Do any of the motherboards have this wording (ECC supported / operates in non ECC mode)? I mean I know that X370 boards had this on a few boards but I have not seen it on X399 although I have not extensively looked at the specifications.
 
Last edited:
ECC should have made native support on all UEFI to my understanding.
 
Does anyone have a hypothetical "maturity" timing for TR performance and what any final comparisons may be in areas where TR is currently lacking? That is to say, as it seems that Intel X99 / X299 chips have some performance benefits on single (or few) threaded applications due to the higher frequencies when OC'ed. I know myself for instance my 5960X runs at 4.5ghz (turbo'ed) across all threads. However, while this initially seems to be a little unfortunate for things like gaming, despite TR's strong - sometimes exceptionally so - performance in other situations. However, it seems to be launching from a stronger initial position than Ryzen and I am wondering if with maturity TR may end up minimizing any of these potential shortcomings.

If I'm correct it seems there are a handful of ways that Threadripper can mature. First of all is regarding overclocking and cooling itself. I am curious to see what happens when proper TR4 socket-designed coolers come onto the scene (I saw that X399-ready EKWB Supremacy EVO were announced) and for that matter, how existing Socket 2066 / 2011-3 style blocks like the EKWB Supremacy EVO and Swiftech Apogee XL2 are doing, regarding enabling better cooling and if this will affect OC potential. Next seems to be drivers and software optimization. Ryzen seemed to fall behind (especially in gaming) at first, but as time went on and many bugs were fixed, AMD was given equal treatment to many Intel-specific optimizations etc... the few gaps on Ryzens' already strong showing began to close. With Threadripper from a stronger starting point I wonder if this will also be the case. Is there any other potential points of enhancement due to maturity that I'm not thinking of?

Finally, in the areas where TR may not hold the crown on benchmarks, I guess what it really comes down to is how big a gap there might be when it comes to real world performance? I think many of us were pleasantly surprised that AMD managed to get TR to essentially become top dog (for equal/lower price!) in the number of metrics and circumstances they did, when the possibility for a more traditional " slightly lower performance all around, yet cheaper than Intel" would still have been a win for AMD. So for things like gaming and other situations etc... will there really be a significant difference in real world performance, especially after a little maturity to the platform? Not that long ago many people were saying of the mainstream boards that "forget Ryzen for gaming" based on some similar clock speed / single core performance kinda things, but that has more or less evaporated and lots of people are building performance Ryzen gaming rigs, so I wonder if the same thing will happen with the stronger-starting by comparison Threadripper. If it turns out that TR has the strong lead in many situations and where it lags a little it doesn't lag much in the real world (made even smaller by maturity), perhaps it is still the overall best choice for a HEDT used for many activities including gaming?
 
Do any of the motherboards have this wording (ECC supported / operates in non ECC mode)? I mean I know that X370 boards had this on a few boards but I have not seen it on X399 although I have not extensively looked at the specifications.

Yes, in a Reddit post, an Asus rep said this was the case, but it was unclear which specific motherboard he/she was referring to. That's why I wanted to confirm before purchasing.
 
Funny enough, I'm actually kind of excited to see what they do with the new tech in the APU department. It'll never replace my "proper" gaming desktop, but I'm someone that would love to put more usable (yet small form factor) PCs in some more rooms of the house. My kids and I are doing more and more LAN play as time goes on, and I've been slowly building more desktops to get them off their laptops for playing shooters. A decently fast modern APU would do nicely for bolstering that further.

Personally, I doubt it will be that much better than the best iGPU that Intel offers. Maybe it might add a feature or two, and be better at tesselation/geometry. I hope I’m wrong, but this is my gut feeling.

The other problem that AMD has, is that the Vega architecture is (yet again) very inefficient and produces a hell of a lot of heat, and that’s really not good for integrating in to a CPU. So whatever Vega tech they use in the iGPU has to be drastically cut down, and clock reduced compared to the desktop GPU.

AMDs GPU department has been badly underperforming for many years, and still shows no sign of being able to pull ahead of any competition anytime soon.
 
Last edited:
Personally, I doubt it will be that much better than the best iGPU that Intel offers. Maybe it might add a feature or two, and be better at tesselation/geometry. I hope I’m wrong, but this is my gut feeling.

The other problem that AMD has, is that the Vega architecture is (yet again) very inefficient and produces a hell of a lot of heat, and that’s really not good for integrating in to a CPU. So whatever Vega tech they use in the iGPU has to be drastically cut down, and clock reduced compared to the desktop GPU.

AMDs GPU department has been badly underperforming for many years, and still shows no sign of being able to pull ahead of any competition anytime soon.

I think that's probably a reasonable assessment. I'm not expecting miracles, but if they do manage to top Intel's even by a little bit though, then it may be worth it for the mid-level tasks I have in mind for them.
 
I wonder if the fact that the substrate has space for 4 dies means that the socket can support 4 cpu dies if AMD wants to release a 32core/64thread CPU in a TR4 socket. They probably won't, but who's to say what the future will bring. AMD has a history of keeping the same socket designs for a long time.
 
I wonder if the fact that the substrate has space for 4 dies means that the socket can support 4 cpu dies if AMD wants to release a 32core/64thread CPU in a TR4 socket. They probably won't, but who's to say what the future will bring. AMD has a history of keeping the same socket designs for a long time.

they could the only problem is that with the quad channel 2 chips would be sharing memory controllers. in one of the reviews kyle linked on the front page they talked about it but i can't remember which one it is since i read all of them, lol. but yeah i doubt they'd ever do it even though it would be fun to see AMD do it just to cock block intel's 18 and 20 core i9's if they actually release them.
 
they could the only problem is that with the quad channel 2 chips would be sharing memory controllers. in one of the reviews kyle linked on the front page they talked about it but i can't remember which one it is since i read all of them, lol. but yeah i doubt they'd ever do it even though it would be fun to see AMD do it just to cock block intel's 18 and 20 core i9's if they actually release them.
Do we know for sure how the DIMM slots are wired?

If each DIMM slot has dedicated traces back to the CPU socket then 8 channel mode is not completely impossible
 
I'm entertaining the thought of getting a Threadripper. I'd like to set up one of those Linux systems with a virtual Windows gaming VM with passed through GPU. With a Threadripper, I could give 8 cores to each, which would be pretty awesome.

Before the whole Zen thing launched I was skeptical. I've always held that per core performance is more important than many cores, and while I knew more cores were in play, but I was expecting the individual cores to perform worse than they did.

This CPU may be behind Intel in per core performance, but it does better than I had expected:

1502322393x097rt9n5y_5_5.png


My hexacore i7-3930k has overclocked like a champ for six years at 4.8Ghz, but it's currently down to 4.6Ghz for the summer.

At 4.6Ghz it gets a Cinebench R15 single threaded score of 156, and at 4.8 it gets 163, so Threadripper - with 16 cores - is beating my Sandy-E at 4.8Ghz, and there is still overclocking to be done on TR once the bigger water blocks come in.

I switched to Intel after loving AMD during the K7 and K8 era when Phenom came out and they they just couldn't keep up anymore. I've wanted to go back ever since. Maybe this is the time.

Might have to wait a bit though, as it is a hefty investment. I'n need a new water block, new RAM (all my current ram is DDR3) new motherboard AND a new $1,000 CPU)
 
EK denied HardOCP a sample since we do not put all its PR spam on our news page giving EK free advertising. So when you guys come asking for EK results, you will need to discuss that with EK. We have others on the way though.


From: EK Atila Gobor [mailto:[email protected]]
Sent: Tuesday, August 15, 2017 2:10 PM
To: [email protected]
Subject: RE: EK is announcing dedicated AMD® Threadripper Supremacy EVO water blocks


Hello HardNews,
I see that you requested a water block sample for testing, and I see that you are getting the newsletters, but I see no EK news posted on the web page....

And we have sent out quite a few samples for far to close partners and medias, so for now, we are not sending out new samples.

BR,
Atila

adsfdasdsa.jpg
 
That is disappointing. These companies should learn that there are few things better than a good review from a reputable independent site like this.

I'm a huge fan of the EK products I've used, but I am glad the H stays independent and doesn't bow to the free advertising demands they make.
Far from the first company that has turned us down on samples, which is no big deal, but to tell me it is because they don't get free advertising (when that is our business model), is kind of crazy. I guess they think the reader exposure we would give EK is not worth the cost of the product and shipping.....and that is not a lot money. :(
 
Wow, kinda crazy. I would think that on a site like [H] where a large percentage of people who might actually use EK's products would be found, they'd be willing to send a sample. So they want to see "EK news" - is there any chance that's a language barrier kinda thing? Can you point them at the last time that you reviewed an EK product yourself or covered a major product launch? Even if it was awhile ago it would show that that you cover something and do it well when there is something to cover. I don't know if you've already done this but maybe more clarification would be helpful?

What does "EK News" mean to them - ie does it mean HardOCP listing "news" containing some review or post of an EK product (ie we don't see you reviewing our products that often), versus actually expecting you to list THEIR news items verbatim from their page? If there was any ambiguity at all, perhaps things could change with clarification.

Likewise, what are "Medias and Close Partners" and what does it take to qualify? After all, HardOCP / HardForum is one of the elder statesmen of enthusiast PC hardware and long have been reviewing watercooling hardware. Not to mention having a readership/community made up of people likely to actually be interested (with the means, knowledge, and interest) in a custom watercooled build - this is not something many places can say within a somewhat niche hobby. Many major hardware companies proudly display the [H]ard Gold or Silver badge when their hardware is reviewed favorably, and often volunteer hardware samples to [H], showing value in the site's reviews.

If they really want you to churn out every news item they make, replicated onto your page, that's unfortunate. I've been under the impression that EKWB makes some really great hardware so if they either expect free advertising or only "partner" with the kind of sites that give astroturfed glowing reviews, that would be disappointing.

Do you think its worth it for any of us - especially those with EK hardware sending pics along - to politely contact EK and express disappointment at not granting review samples to [H]?
 
Wow, kinda crazy. I would think that on a site like [H] where a large percentage of people who might actually use EK's products would be found, they'd be willing to send a sample. So they want to see "EK news" - is there any chance that's a language barrier kinda thing? Can you point them at the last time that you reviewed an EK product yourself or covered a major product launch? Even if it was awhile ago it would show that that you cover something and do it well when there is something to cover. I don't know if you've already done this but maybe more clarification would be helpful?

What does "EK News" mean to them - ie does it mean HardOCP listing "news" containing some review or post of an EK product (ie we don't see you reviewing our products that often), versus actually expecting you to list THEIR news items verbatim from their page? If there was any ambiguity at all, perhaps things could change with clarification.

Likewise, what are "Medias and Close Partners" and what does it take to qualify? After all, HardOCP / HardForum is one of the elder statesmen of enthusiast PC hardware and long have been reviewing watercooling hardware. Not to mention having a readership/community made up of people likely to actually be interested (with the means, knowledge, and interest) in a custom watercooled build - this is not something many places can say within a somewhat niche hobby. Many major hardware companies proudly display the [H]ard Gold or Silver badge when their hardware is reviewed favorably, and often volunteer hardware samples to [H], showing value in the site's reviews.

If they really want you to churn out every news item they make, replicated onto your page, that's unfortunate. I've been under the impression that EKWB makes some really great hardware so if they either expect free advertising or only "partner" with the kind of sites that give astroturfed glowing reviews, that would be disappointing.

Do you think its worth it for any of us - especially those with EK hardware sending pics along - to politely contact EK and express disappointment at not granting review samples to [H]?
Seemed very clear to me that since I do not post all the press releases they send HardOCP on the news page, that they were not interested in our readers. No biggy. That is the way the cookie crumbles. Plenty of other hardware to cover in the market.
 
I wonder if the fact that the substrate has space for 4 dies means that the socket can support 4 cpu dies if AMD wants to release a 32core/64thread CPU in a TR4 socket. They probably won't, but who's to say what the future will bring. AMD has a history of keeping the same socket designs for a long time.


Personally I was hoping for motherboard core unlock potential like we had with the Phenom II's and Athlons of that era :p
 
Have my 1950X/ROG Zenith up and running finally, and yeah, AIO water isn't going to cut it, folks.

Guess I'll drop even more money on a proper loop..

Good to know thanks.

I'm probably my htpc and game rig into one and replace the htpc with a nuc or something tiny just for playback. Now, how to stuff 18 drives into a Core X9.
 
Have my 1950X/ROG Zenith up and running finally, and yeah, AIO water isn't going to cut it, folks.

Guess I'll drop even more money on a proper loop..
Just do it! I am (though not on TR).
20170813223956.JPG 20170816203745.JPG
Currently BSing my copper tubes into shape, two "successful" bends so far. Maybe I should stop being lazy and take measurements for once...
Edit:Monsoon economy compression fittings are really nice, btw. They only come in humongous, but they're plated brass and come cheap in a six-pack. Haven't pressure tested mine yet, though.
 
Last edited:
Looks good. Pulled the trigger on an EK block and fixings. H105 works for now though. Have proper cables coming as well. Not bothering w/ GPU blocks this time around. Surprisingly, this works fine for my purposes. Volta time I will for sure.

PdSweoR.jpg
 
Looks good. Pulled the trigger on an EK block and fixings. H105 works for now though. Have proper cables coming as well. Not bothering w/ GPU blocks this time around. Surprisingly, this works fine for my purposes. Volta time I will for sure.

PdSweoR.jpg
Need to update that sig. ANd let me add, "Good god man!"
 
Looks good. Pulled the trigger on an EK block and fixings. H105 works for now though. Have proper cables coming as well. Not bothering w/ GPU blocks this time around. Surprisingly, this works fine for my purposes. Volta time I will for sure.

PdSweoR.jpg
Four in there just cause you can or are they staying there long term? If so, I'm curious how you will employ them
 
Four in there just cause you can or are they staying there long term? If so, I'm curious how you will employ them
Mostly, yeah. It's a 4K editing/live production workstation some of the time. I've also been experimenting with a whole-house game streaming setup with Nvidia Shield boxes.
 
Going to get my AMD system put together this weekend. Got the parts sitting on my workbench.

Threadripper 1950X
Vega 64
ASUS ROG Zenith Extreme
Samsung 960 Evo M.2 1TB NVMe
EVGA Supernova 1300G2
8x 8GB Corsair 3333 RGB
Thermaltake 360 RGB AIO
Inwin 805 Infinity
 
Back
Top