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

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,534
AMD Ryzen Threadripper 1950X and 1920X CPU Review - The day is finally upon us that many CPU enthusiasts have been waiting for. We get to see what AMD's new Threadripper CPU is all about in terms of performance, and in attempts to cool the beast. There has been no lack of hype for months now, so let's see if it is all justified.
 
The 2 additional dies are allegedly dummy dies that provide structural support.

To make a long story short, the two extra dies in Threadripper are merely structural inserts, with the active dies being placed in a diagonal configuration in all Threadripper CPUs to give them all the same thermal profiles. Threadripper CPUs are not "failed EPYC" CPUs and they certainly are not made in a wasteful way by AMD.
 
It's always nice being pleasantly surprised, TR is kicking serious ass at this price point, very impressive, great review
 
Have you thought about using another benchmark other than Hyper-Pi and WPrime? y-cruncher is really nice, if you want to give it a look. It'll use all your cores and even rate how effective is your hyper-threading.

Nice writeup. For sure seems like a production piece with gaming chops.
 
The 2 additional dies are allegedly dummy dies that provide structural support.
All I could get out of AMD was that the extra dies are "non-functioning" so who knows.
 
Have you thought about using another benchmark other than Hyper-Pi and WPrime? y-cruncher is really nice, if you want to give it a look. It'll use all your cores and even rate how effective is your hyper-threading.

Nice writeup. For sure seems like a production piece with gaming chops.
Hyper Pi will do all cores, I just use it for the metric that it is. Don't need any more synthetics IMO.
 
if only i had a reason for a setup like this, either way it was worth waiting til 6 am to read the review.. competition is finally getting spicy in the cpu market.
 
And now, I almost wish I had waited for Threadripper to do my build. Oh well, not exactly suffering with a cheap 8 core box, but still, most impressive. Very curious to see the overclocking results with the new waterblock. I'll be waiting!
 
Wow! Thanks for your review! The only thing I'm concerned about is that, given the price point, the Intel 7900 OC'ed to 4.6 seems to be so competitive and in some cases come in significantly above on benchmarks, perhaps due to its faster frequency.

I am very interested to see your future overclocking investigations to see what kind of performance we can really get out of a 1950X, OC'ed across all threads. Does it seem possible to "catch up" to Intel's performance in single/low threaded workloads with an enhanced OC? If so, then TR could really be an absolute stand out across all workloads and usage, as one would expect of something at this price point. No matter what I'm glad to see AMD competitive again and especially at the HEDT leve, but it would be nice given the price point if it could equal/beat Intel's performance in single/lower threaded workloads as well as thrive in the usages cases that can make use of the additional cores/threads.

It does seem that most coolers aren't designed for/up to the task, for the moment though; good to know. Do you think you could please look into how common higher-end waterblocks from EKWB (ie Supremacy EVO), Swiftech (Apogee; both the current edition on which the block included in the "X2" AIO's is based, and the new upcoming next generation taking pre-orders), and more? Just to know if some other major waterblocks are up to the task or if we really absolutely need purpose-built, socket-specific blocks.
 
It's definitely interesting. I hope they can kick on to some big sellers.

Wish they did a lower power consumption one with a lower clock speed as I'd take a punt. The intel 1567 and 1577 are witchcraft in that regard, different market but one AMD should really be pitching Epyc or Threadkiller towards.
 
Question for Kyle:

I noticed that PCPer and Anand ran the 1950X in both "Game mode" (UMA vs NUMA modes) and regular mode. Which mode were your benchmarks in?
 
Question for Kyle:

I noticed that PCPer and Anand ran the 1950X in both "Game mode" (UMA vs NUMA modes) and regular mode. Which mode were your benchmarks in?
Mine were done in "Game" mode. Memory access modes were not available to change on my setups. Distributed vs Local no worky from what I was exposed to.

Edit: To fill everyone in...

upload_2017-8-10_8-45-37.png
 
This thing looks like all that was promised! I wish I had the cash and the need to use one. I do think our next virtualization hardware in the data center I work at will be Epyc based though.
 
Wonder how well this CPU will hold four or five years from now.
That's the question I'm always trying to answer when I go to build a new PC.

Kyle, you packin a time machine anywhere? :p
Let's go back and buy some Google/Apple stock, too.
 
Nice to see IPC keeping up with Intel for the most part. Will be looking forward to the followup on overclocking. It's going to be real hard deciding between a 1950X and 7940X now.
 
Was hoping TR would be a little more competitive gaming-wise, but I understand it continues the trend we've seen lower in the Ryzen line-up... moreover it's probably not the ideal use case for the processor anyway.
 
I wonder what perf would have been like if run in the game mode in tests where NuMA is issue.

Thanks for the unique review. Makes me want to get a TR and water cool it.
 
To be honest, i got a little bit bored reading reviews, it basically performed as i have expected it to.

Am glad to know that platform was nailed perfectly this time.
 
def going to wait for some more cooling options. (I'm lazy like that). also waiting to see if universe sandbox starts supporting
 
Wow! Thanks for your review! The only thing I'm concerned about is that, given the price point, the Intel 7900 OC'ed to 4.6 seems to be so competitive and in some cases come in significantly above on benchmarks, perhaps due to its faster frequency.

I am very interested to see your future overclocking investigations to see what kind of performance we can really get out of a 1950X, OC'ed across all threads. Does it seem possible to "catch up" to Intel's performance in single/low threaded workloads with an enhanced OC? If so, then TR could really be an absolute stand out across all workloads and usage, as one would expect of something at this price point. No matter what I'm glad to see AMD competitive again and especially at the HEDT leve, but it would be nice given the price point if it could equal/beat Intel's performance in single/lower threaded workloads as well as thrive in the usages cases that can make use of the additional cores/threads.

It does seem that most coolers aren't designed for/up to the task, for the moment though; good to know. Do you think you could please look into how common higher-end waterblocks from EKWB (ie Supremacy EVO), Swiftech (Apogee; both the current edition on which the block included in the "X2" AIO's is based, and the new upcoming next generation taking pre-orders), and more? Just to know if some other major waterblocks are up to the task or if we really absolutely need purpose-built, socket-specific blocks.
If you're talking about catching up to the 7900 in single thread workloads, I doubt it'll happen with overclocking. 4GHz was pretty much the ceiling with ryzen, and I expect you wont get much more with epyc/threadripper.

After optimization and with overclocking, you might get close... we'll have to wait and see how much they can actually do to optimize this beast of a cpu.
 
This is turning into maybe some of the best of times for PCs and thanks for bringing back the excitement in the review. Great integration of all the previous work with TR which gives the review a one stop power-base to work from. I would have total confidence in being able to build a rig based upon your work.

For me TR is probably a next year build, a serious platform which I would want good maturity to go along with it. Looks like AMD and AIBs has learned much from RyZen and has a better launch for TR - great news.
 
I can finally make a sound recommendation for our compile machine. Compiles usually take to 4 hours for 72 projects. I think this will nicely cut us down as Visual Studio's compiler uses all available threads (unless you specify otherwise in the compile options.)

Well done AMD.
 
To be honest, i got a little bit bored reading reviews, it basically performed as i have expected it to.

Am glad to know that platform was nailed perfectly this time.


that's why i like [H] reviews, straight to the point and without the 20 pages of fluff you see on some other sites..
 
Thanks Kyle, she looks like a winner for sure, looking forward to the follow up with the big block on it. If I were marketing to the TR crowd id name the WB 454/502/572, honoring the massive size. of the IHS! 4.2Ghz on this would be amazing...Is it possibly to do a clock for clock comparison in the future?
 
Good stuff!! I want! My 6600k isn't up to snuff for all the bluray ripping I've been doing.. Stupid 20 minute audio encodes.
 
I can finally make a sound recommendation for our compile machine. Compiles usually take to 4 hours for 72 projects. I think this will nicely cut us down as Visual Studio's compiler uses all available threads (unless you specify otherwise in the compile options.)

Well done AMD.
Dunno, mate, those compile tests i have seen were nothing to write home about compared to 7900X. ECC support tips it in TR's favor though (unless it starts segfaulting lul).
 
Another nail in intel coffin, courtesy of AMD. mwahahahaah

Nice review Kyle.
 
Can't wait to see how the 1900x fairs, I just see so much long term potential on this platform, even if the cost of entry is significant in comparison to the x370/1700 setup.
 
Back
Top