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

Thanks for the in-depth review, [H]!

Hope to see lots of software optimizations for TR in the near future...this many cores/threads at these clock speeds have the potential to own the field for years to come.
 
^ Are you supposed to be posting that? Not sure I've ever seen an entire reviewers guide out in the wild. Good on AMD if they are letting that go public.

Awesome review, Kyle. These things are out of my reach for my personal use right now but I'm pretty stoked to see what this shakeup does for the market I do shop in over the next few years.
I will leave it there till they tell me to remove it. Figured you guys are welcome to see what I see. No more embargo.
 
So. Does the[H] agree with this headline?

Screenshot_20170810-115603.png
 


Funny how amd's mobo didn't have that plastic squeeze cover over the pins 0_o
 
Not bad. 7820x would be the better CPU to buy if one doesn't need all those PIC-e lanes. It is going to perform better then the 1900x and the price difference is going to be a wash. The 1950x tho is a beast for the price tho. Nice to have AMD around to start making Intel honest.
 
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.

Doesn't Game mode = NUMA, so I assume you meant UMA? UMA is slower than NUMA. NUMA is faster due to the ability of each Zeppelin die to prioritize memory access in games, but the flipside is you lose a lil bit in media creation tools.
 
I am torn between 7900x and 1950x, but with AMD making me this excited over the hobby in years, I am leaning toward the 1950x after proper coolers come out.
 
Waste costs money, now they've solved the waste problem too!, just make them a structural support and stop throwing any chips.
*AMD Masterplan completed - One chip to do it all!

That's a damn good point. They could do two separate runs on one set of equipment, one with Epyc and another with 2 of 4 die feed machines with binned dead chips....

Do wonder what they are, because if they're slabs of silicon that seems like a waste in itself but i guess it's sharing the same line as Epyc cpus so no retooling needed, i guess i'll wait till someone delids one of these suckers and takes a microscope to it.

This please. Someone has to confirm the mystery!

Sleep well Kyle, you did well today. I hope you have great blackout curtains.
 
The high clocks on that intel kinda surprised me how much they were able to deliver performance wise vs the TR. I'm sure there's a performance per dollar graph someone will come up with for a full kitted out system, which would be interesting to see. I'm really looking forward to some serious cooling and overclocking potential even if it's some disabled cores for gaming scenarios. Would love to have 4 Cores ~5ghz on days I don't need to render.
 
That's a damn good point. They could do two separate runs on one set of equipment, one with Epyc and another with 2 of 4 die feed machines with binned dead chips....



This please. Someone has to confirm the mystery!

Sleep well Kyle, you did well today. I hope you have great blackout curtains.

It's been documented many times before but ppl still believe or have hope.

https://www.extremetech.com/computing/253416-amd-explains-threadripper-cpus-four-die-hood

The explanation is straightforward: The reason Threadripper has four dies under the hood with only two in use is because leaving them off would create mechanical instability when the heatsink is mounted. AMD apparently needs four dies in that arrangement to stabilize the packaging. This does make sense, given some of the esoteric cooling solutions with high pressure attachments that overclockers use...

The other important thing to know about Threadripper is that the other two die apparently aren’t actually chips at all. They’re apparently structural inserts required for support, but do not represent “bad” Epyc cores or any kind of core recycling. OC3D also states that the two active die in Threadripper are on the diagonal from each other.
 
Most impressive. Those are some pretty solid numbers. It's good to have a reason to be excited about AMD again.

LOL @ the unboxing. Somebody needs to do some strength training.:p
 
Even if you can OC a 7900 to be comparable to the 1950x you are losing out on 20 PCIE lanes and ECC memory. Anyone planning on using this for gaming or gaming and streaming is just wasting their money.
 
  • Like
Reactions: ecktt
like this
It's been documented many times before but ppl still believe or have hope.

https://www.extremetech.com/computing/253416-amd-explains-threadripper-cpus-four-die-hood

I'm not one with hope/belief, just curious for the truth with a microscope. Have looked into the original source of this before and the wording was pretty vague, what Kyle has stated is the most definite answer so far and contradicts it a little. As I said in past, I'm just curious from an industrial POV, don't care/think that it would be 'unlockable' more so than if they actually are failed Ryzen dies or simple structural (glass?) pieces as stated by OC3D.

Marketing would be the reason to not say 'failed ryzen dies'. As pointed out earlier, it's the perfect way to dump waste dies if they don't want to melt them.
 
If you have a case that has a good place to put it, I have seen an inexpensive PCIe-to-PCI adapter on NewEgg. I, too, have a need for PCI slots...


I actually purchased that one (not sure if it was the same manufacturer but looks the same) months ago but have not tested it yet. My original goal was to add a PCI slot to an X370 board in linux however with the Ryzen instability in linux issue (seems like a possible solution finally) I have waited on the upgrade.
 
Did people just read the same review I did? Intel Stitched up threadrippers ass with steal string in all but Blender ,Cinebench and POV-Ray. Games and Premiere prefer Intel. Hell econding MP3 and compression still prefers Intel. As a desktop processor, which I believe it is marketed as, the x7900 is the better chip. Is the AMD bandwagon so full of shit that people can't see? I'm Glad that thread ripper exist to force innovation and competition but get real men(and women).
 
Nice review, performing basically how you would expect. I really wonder how the TR4 specific air coolers (Noctua mainly) will fair. We saw here that the biggest obstacle seems to be the size of the IHS, so maybe the dual tower with the appropriate size base will be sufficient.
 
I have to honestly say that I have not seen this much excitement over a computer part since the 90s.

That does not mean I haven't seen company manufactured hype which has been the standard. But to see this much excitement from this many outlets over the new ThreadRippers really brings me back to when Athlon hit the scene.

One can only hope AMD is prepared for the anti-competitive practices Intel will use to try to keep this chip down.

However I don't think the same tactics will work given how accessible the internet is now. Cats out of the bag.

If this is the start of how things were in the 90s then Hardware enthusiasts are going to be doing double backflips.

Even more if Vega can pull off some cool things..
 
Did people just read the same review I did? Intel Stitched up threadrippers ass with steal string in all but Blender ,Cinebench and POV-Ray. Games and Premiere prefer Intel. Hell econding MP3 and compression still prefers Intel. As a desktop processor, which I believe it is marketed as, the x7900 is the better chip. Is the AMD bandwagon so full of shit that people can't see? I'm Glad that thread ripper exist to force innovation and competition but get real men(and women).

It is marketed as a High End Desktop CPU - read that as WORKSTATION CPU. As a WORKSTATION CPU, Threadripper is an incredible value with its (official) support of ECC memory and 64 PCIe Lanes (both areas where the Intel system is lacking). No one expected it to rule the gaming roost, and Intel has been favored for optimizations for some time now. Kyle clearly stated that in several of the benchmarks the CPU was under-utilized (anywhere from 20-30% idle time) whereas the Intel stayed near 100% utilization. Despite this deficit, the performance of Threadripper is still very impressive AND it has a lot of room to improve through simple software optimization.
 
I'm not one with hope/belief, just curious for the truth with a microscope. Have looked into the original source of this before and the wording was pretty vague, what Kyle has stated is the most definite answer so far and contradicts it a little. As I said in past, I'm just curious from an industrial POV, don't care/think that it would be 'unlockable' more so than if they actually are failed Ryzen dies or simple structural (glass?) pieces as stated by OC3D.

Marketing would be the reason to not say 'failed ryzen dies'. As pointed out earlier, it's the perfect way to dump waste dies if they don't want to melt them.

SP3 and TR4 sockets are similar but not electrically compatible. TR4 is 4 channel memory whereas SP3 is 8 channel memory. Thus for arguments sake if AMD did secretly throw away money in dies and those two dummies are actually full dies, once enabled they can't be used in TR4 sockets anymore or could they? The plot thickens?
 
Did people just read the same review I did? Intel Stitched up threadrippers ass with steal string in all but Blender ,Cinebench and POV-Ray. Games and Premiere prefer Intel. Hell econding MP3 and compression still prefers Intel. As a desktop processor, which I believe it is marketed as, the x7900 is the better chip. Is the AMD bandwagon so full of shit that people can't see? I'm Glad that thread ripper exist to force innovation and competition but get real men(and women).

Depends on the review and what they test.


It's always best to look at a number of reviews and their tests and decide what works best your your use case.

Seeing as both HEDT platforms are a waste of money for gaming its best to just get a 7700k if that is your focus.
 
great review, looks like it should be good for virtualization. wish i had a need to justify it lol, my i5 7600 is all i need for games sadly and this doesn't seem like a worthy purchase for gaming really.
 
SP3 and TR4 sockets are similar but not electrically compatible. TR4 is 4 channel memory whereas SP3 is 8 channel memory. Thus for arguments sake if AMD did secretly throw away money in dies and those two dummies are actually full dies, once enabled they can't be used in TR4 sockets anymore or could they? The plot thickens?

Doubt it indeed, they wouldn't waste time/traces wiring it up on boards. One thing of note is the (decoupling capacitors?) and their arrangements on both sides of the chip. They are still wired in on the 'opposite' side of the interposer and supposedly the interposer/substrate is identical, so electrically they can't be too different. Perhaps they use some of the unused memory pins for additional power delivery?



Some OC results from around the net

4.1 @ 1.4V (still under 1.45 max safe recommended) on a thermaltake water 3 at 85 degrees..
4.0 @ 1.25V 65deg

4.8-5.18 on LN2 at 1.6V win 10/cinebench stable
4.8GHz is 3688 and 5.18 is 4122 on cinebench
 
Umm I guess as a package deal with more lanes, cheaper motherboards, and being able to use ecc RAM this is good but TR is certainly losing to the 7900x.

Petty disappointed tbh. The 1900x seemed great.
 
Doesn't Game mode = NUMA, so I assume you meant UMA? UMA is slower than NUMA. NUMA is faster due to the ability of each Zeppelin die to prioritize memory access in games, but the flipside is you lose a lil bit in media creation tools.

Game mode also turns SMT off, Creator mode has SMT on but UMA set for the memory. I

t's unfortunate they don't have a 4 settings, so you could have NUMA and SMT directly from the Ryzen Master software, not a big deal though since you can always do it manually in the bios.
 
Umm I guess as a package deal with more lanes, cheaper motherboards, and being able to use ecc RAM this is good but TR is certainly losing to the 7900x.

Petty disappointed tbh. The 1900x seemed great.
TR MB are not cheaper. The cheapest one I seen was $350. x299 board can be had at $200-250. I am sure there will be TR boards at sub $250 also but would make it a push at best.
 
TR MB are not cheaper. The cheapest one I seen was $350. x299 board can be had at $200-250. I am sure there will be TR boards at sub $250 also but would make it a push at best.

The x399 Giga Aorus is 385 while the x299 Aorus is 499.
 
Did people just read the same review I did? Intel Stitched up threadrippers ass with steal string in all but Blender ,Cinebench and POV-Ray. Games and Premiere prefer Intel.
Um TR won in Premiere.
1502322393x097rt9n5y_5_2.png

Essentially if it's productivity TR is pretty much the winner, which is exactly why you buy such a platform. Throw in ECC which is pretty much mandatory when you have your project sitting in memory and I would question your understanding of the market for these chips. They are not for games. You want to game? Get a 7700K.
 
Um TR won in Premiere.
1502322393x097rt9n5y_5_2.png

Essentially if it's productivity TR is pretty much the winner, which is exactly why you buy such a platform. Throw in ECC which is pretty much mandatory when you have your project sitting in memory and I would question your understanding of the market for these chips. They are not for games. You want to game? Get a 7700K.

I guess the question depends on how well TR overclocks because I would not call the 16C a clear winner in this chart at all. I mean 583 versus 589 is not that big of a deal. Although the 589 is overclocked and I expect a few hundred extra MHz from the TR when overclocked.


Edit: I am sorry I missed the fact that the 4GHz was an overclock. I see now what you mean. I did not have a lot of time to read reviews today (since I am at work)..
 
Last edited:
  • Like
Reactions: ecktt
like this
The x399 Giga Aorus is 385 while the x299 Aorus is 499.
There are like 3 x299 Aorus boards, which one do you mean? The Aorus 9 which is $500 even in Z270 version?
I guess the question depends on how well TR overclocks because I would not call the 16C a clear winner in this chart at all. I mean 583 versus 589 is not that big of a deal. Although the 589 is overclocked and I expect a few hundred extra MHz from the TR when overclocked.
Uhm, 583 is overclocked result itself, and i would not expect more than 100Mhz from TR compared to Kyle's sample.
 
Back
Top