Review Threadripper 3rd Gen Review and Availability Thread

[Hitman928]

Since the other thread got merged and I am not the OP as well as the announcement thread being so long, I figured it would be better to have a separate review and availability thread.

Once reviews are posted I will update this post. If you see a review you would like added to this post, please PM me.

Video reviews
Hardware Unboxed
Gamers Nexus
Linus Tech Tips

Print reviews
Phoronix
PCWorld
PCMag
Guru3d
Tweaktown
Anandtech
Computerbase (translated)
Servethehome
Phoronix (3990x)
For sale
3960x, 3970x, and 3990x listed for sale on Newegg.

 

[JasonLD]

Yes, and since it will support ECC and its faster than EPYC and could be more powerful VM's that way.

Well, for the ECC support. It is almost as if it is not supporting it. There is a reason reviews are saying Threadripper 3 is limited on 256GB of RAM for now. AMD wants you to use EPYC for RDIMM and LRDIMM support, since Threadripper supports neither. I can see why AMD is doing that. They don't want TR3 to cannibalize EPYC sales.

 

[VietOZ]

3990X in action: https://universeathome.pl/universe/show_host_detail.php?hostid=545169
worth every penny ... this is with crappy bios crippling my DRAM to 2133mhz atm

I want to say 2x 3970X won't produce the same number, but I'll wait a few days to get a more definitive answer.

 

[jalyst]

I want to say 2x 3970X won't produce the same number, but I'll wait a few days to get a more definitive answer.

Nice, thank-you. Any further updates to this?

Best Regards.

 

[Gideon]

Probably not the best thread to bump, but I didn't want to make another one.

It seems that the rumored 8-channel workstation threadrippers might not have been a hoax at all :O

AMD Ryzen Threadripper PRO 3995WX spotted, could launch next week - VideoCardz.com

AMD is rumored to launch a new Ryzen Threadripper series for a high-end workstations. AMD to launch Threadripper PRO series? AMD Ryzen Threadripper PRO 3995X processor as leaked by Chiphell Just hours ago we teased an anonymous tip we received regarding a possible launch of AMD Ryzen...

videocardz.com

AMD Threadripper PRO is said to feature 8-channel memory and up to 2TB of capacity. For comparison, consumer Threadripper series are 4-channel only and they support up to 256GB.
We might finally see the sWRX8 socket coming to the market, it was rumored to feature such configuration more than a year ago.
The Threadripper PRO series would allegedly launch on July 14th, which is exactly a week after Ryzen 3000XT series.

I hope this is true, would be a very niche chip. The limit on max memory on Threadripper 3 was IMO a letdown compared to 2nd gen, the added memory channels are icing on the cake.

 

[Zucker2k]

Probably not the best thread to bump, but I didn't want to make another one.

It seems that the rumored 8-channel workstation threadrippers might not have been a hoax at all :O

AMD Ryzen Threadripper PRO 3995WX spotted, could launch next week - VideoCardz.com

AMD is rumored to launch a new Ryzen Threadripper series for a high-end workstations. AMD to launch Threadripper PRO series? AMD Ryzen Threadripper PRO 3995X processor as leaked by Chiphell Just hours ago we teased an anonymous tip we received regarding a possible launch of AMD Ryzen...

videocardz.com

I hope this is true, would be a very niche chip. The limit on max memory on Threadripper 3 was IMO a letdown compared to 2nd gen, the added memory channels are icing on the cake.

AMD unlocking overclocking on Epyc to sell more chips, if true. Is AMD not moving Epyc chips as much as they'd like to?

 

[coercitiv]

Is AMD not moving Epyc chips as much as they'd like to?

Is Intel's lower HEDT pricing indicative of their Xeon sales?

 

[DrMrLordX]

Is Intel's lower HEDT pricing indicative of their Xeon sales?

Heh! As if Intel HEDT chips were even available.

 

[coercitiv]

Heh! As if Intel HEDT chips were even available.

That's exactly the point: HEDT and Threadripper are niche products and their existence in the market says little about server chip sales.

It was only a few weeks ago that some Ryzen chips were hard to find and had higher than expected prices, is that indicative of EPYC sales?

 

[amrnuke]

AMD unlocking overclocking on Epyc to sell more chips, if true. Is AMD not moving Epyc chips as much as they'd like to?

I think even if sales were "good" they're still not moving as many Epyc chips as they'd like to. I don't think Intel is moving as many Xeons as they'd like either.

Do you think Ford is selling as many F150s as it would like to?

Don't you think this is a silly question?

 

[Markfw]

AMD unlocking overclocking on Epyc to sell more chips, if true. Is AMD not moving Epyc chips as much as they'd like to?

Since this was remored many MONTHS ago, I doubt that has anything to do with it. And they are not unlocking overclocking, as the sockets are not interchangeable.

 

[Zucker2k]

Don't you think this is a silly question?

No. Artificial segmentation has always been a thing. Let's see how much it costs.

 

[Zucker2k]

Since this was remored many MONTHS ago, I doubt that has anything to do with it. And they are not unlocking overclocking, as the sockets are not interchangeable.

Repackaged, though? Would this chip not use epyc's io dies? So, basically, the only difference here would be binning and overclocking?

 

[Hitman928]

Repackaged, though? Would this chip not use epyc's io dies? So, basically, the only difference here would be binning and overclocking?

It's been a while since they launched, but from memory, Epyc and Threadripper use different IO dies.

Edit: Yes, Epyc uses different IO dies. They are even made on different processes. Epyc IOD is 14 nm and Ryzen/TR IOD is 12 nm.

 

[Zucker2k]

It's been a while since they launched, but from memory, Epyc and Threadripper use different IO dies.

Edit: Yes, Epyc uses different IO dies. They are even made on different processes. Epyc IOD is 14 nm and Ryzen/TR IOD is 12 nm.

Is this chip, with it's 8-channel memory support, going to use a non-epyc io die?

 

[Hitman928]

Is this chip, with it's 8-channel memory support, going to use a non-epyc io die?

No idea, first I'm hearing of it.

 

[StefanR5R]

It's been a while since they launched, but from memory, Epyc and Threadripper use different IO dies.

Edit: Yes, Epyc uses different IO dies. They are even made on different processes. Epyc IOD is 14 nm and Ryzen/TR IOD is 12 nm.

Is there any actual evidence that Threadripper's IOD is not actually the same as EPYC's, just with features unused? (And the TRX40 southbridge chip possibly too.)

Threadripper Pro's IOD most certainly is the same as the one in EPYC.

AnandTech	Cores	Base Freq	Turbo Freq	Chiplets	TDP	DRAM
3995WX	64 / 128	2700	4200	8 + 1	280 W	8 x DDR4-3200
3975WX	32 / 64	3500	4200	4 + 1	280 W	8 x DDR4-3200
3955WX	16 / 32	3900	4300	2 + 1	280 W	8 x DDR4-3200
3945WX	12 / 24	4000	4300	2 + 1	280 W	8 x DDR4-3200

I wonder about the chiplet configuration of 3955WX and 3945WX. Is the chiplet config confirmed by AMD, or merely an assumption by the author of the article?

(I would ask in the comment section of the article... if I had an account there.)

 

[zir_blazer]

It's been a while since they launched, but from memory, Epyc and Threadripper use different IO dies.

Edit: Yes, Epyc uses different IO dies. They are even made on different processes. Epyc IOD is 14 nm and Ryzen/TR IOD is 12 nm.

As far that I know, TR and EPYC uses the same 14nm IO die. Note that Matisse IO die and the X570 Chipset are based on the same design but manufactured on different processes, so even if manufactured on different processes they can be the same design, if not silicon.
I even recall reading somewhere that Rome IO die was actually pretty much a copypaste of 4 Matisse, which doesn't make sense since you don't want exactly four of the same things (Neither Zen 2 TR nor EPYC exposes the integrated HD Audio controller that Matisse has, reason why TR Motherboards with integrated sound must use some stupid USB-to-HD Audio/Azalia Bridges).

I'm interesed on the 16C one cause at some point I considered that if I had money for a new build, I would go for an EPYC 7282 before Ryzen 3900X (And this was BEFORE it was released) even if I lose a full GHz. With the TR Pro series it seems like it can get the best of both worlds, assuming price isn't much higher than either:

https://www.reddit.com/r/Amd/comments/cp049e

 

[Hitman928]

Is there any actual evidence that Threadripper's IOD is not actually the same as EPYC's, just with features unused? (And the TRX40 southbridge chip possibly too.)

Threadripper Pro's IOD most certainly is the same as the one in EPYC.


I wonder about the chiplet configuration of 3955WX and 3945WX. Is the chiplet config confirmed by AMD, or merely an assumption by the author of the article?

(I would ask in the comment section of the article... if I had an account there.)

As far that I know, TR and EPYC uses the same 14nm IO die. Note that Matisse IO die and the X570 Chipset are based on the same design but manufactured on different processes, so even if manufactured on different processes they can be the same design, if not silicon.
I even recall reading somewhere that Rome IO die was actually pretty much a copypaste of 4 Matisse, which doesn't make sense since you don't want exactly four of the same things (Neither Zen 2 TR nor EPYC exposes the integrated HD Audio controller that Matisse has, reason why TR Motherboards with integrated sound must use some stupid USB-to-HD Audio/Azalia Bridges).

I'm interesed on the 16C one cause at some point I considered that if I had money for a new build, I would go for an EPYC 7282 before Ryzen 3900X (And this was BEFORE it was released) even if I lose a full GHz. With the TR Pro series it seems like it can get the best of both worlds, assuming price isn't much higher than either:

https://www.reddit.com/r/Amd/comments/cp049e

The context of my reply was in reference to this post which was claiming that AMD was overclocking Epyc chips to use as Threadrippers. The fact that TR uses 12 nm for IOD and Epyc uses 14 nm proves that this is incorrect. It doesn't matter if the schematic design is the same or not in this context. I do doubt that they would have ported the IOD to 12 nm but left in a bunch of circuitry that would never be used, but I guess it's possible. In terms of the new workstation TR's, I haven't seen any info on the IOD they are using so it's difficult to say if it's the same as Epyc or not but it would be more likely than the regular TR because the feature sets between the Epyc IOD and the new TR IOD are so similar.

 

[zir_blazer]

The context of my reply was in reference to this post which was claiming that AMD was overclocking Epyc chips to use as Threadrippers. The fact that TR uses 12 nm for IOD and Epyc uses 14 nm proves that this is incorrect. It doesn't matter if the schematic design is the same or not in this context. I do doubt that they would have ported the IOD to 12 nm but left in a bunch of circuitry that would never be used, but I guess it's possible. In terms of the new workstation TR's, I haven't seen any info on the IOD they are using so it's difficult to say if it's the same as Epyc or not but it would be more likely than the regular TR because the feature sets between the Epyc IOD and the new TR IOD are so similar.

Do we have any confirmation about AMD saying that TR3 (Not including the new TR3 Pro) IO die is 12nm instead of 14nm like Rome? I hear people discussing this but was never confirmed, something that was for Matisse. Technically, the IO die size should be physically smaller, so end users could actually confirm this without AMD saying a word . Is there at least some delidded action?

 

[moinmoin]

Technically, the IO die size should be physically smaller,

Not if it's the same as with Zen/Zen+ using 14nm and 12nm respectively. The latter was the same size and mask, the additional spacing earned that way allowed for higher frequencies. In that regard it makes sense using 12nm for IODs in TR3. Haven't seen any confirmation of that either though.

 

[Hitman928]

I linked to Wikichips in my previous post, they're usually pretty reliable for this type of information. TweakTown also posted a product spec sheet showing it is 12 nm but I don't know if the spec sheet is official from AMD. It would seem strange to me that it would be so widely reported as 12 nm if this info wasn't from AMD but I'll see if I can find an official slide or comment.

 

[Hitman928]

Here's an official source, AMD's official tech briefing for the 3990x. At around 21:27 the slide shown confirms IOD is 12 nm.

 

[Markfw]

I think what is happening here from what I see, is that the new 3995wx and the like, use more power/.vcore/heat than EPYC. EPYC are pure servers, efficient in power and heat. The new 3995wx and the like are powerhouses made for speed. I mean 225 to 280 watts for essentially the same CPU and features and number of cores ? But they run faster, and less efficiently. Perfect for those that need the power and can handle the heat/power usage. Not in a room with 1000 or more CPUs like EPYC does. But perfect for a engineering firm with 20 computers.

 

[beginner99]

But perfect for a engineering firm with 20 computers.

Exactly and in such a context, eg workstation, single-threaded or very low threaded workloads sometimes also happen and here the much higher clock speeds compared to epyc will really shine. Such a machine at work but be perfect for me but will never happen...not standard compliant. argh...

 

[Atari2600]

Exactly and in such a context, eg workstation, single-threaded or very low threaded workloads sometimes often also happen and here the much higher clock speeds compared to epyc will really shine. Such a machine at work but be perfect for me but will never happen...not standard compliant. argh...

Fixed that for ya!