Question Zen 3 Threadripper in August according to a leaker

inf64

Diamond Member
Mar 11, 2011
3,697
4,015
136
That is great news for the buyers of HEDT but there is no competition to current TR parts so it's like beating a dead horse. I get that AMD wants to push new parts for more $$$$ but when you lead intel by such a HUGE margin then it's kinda pointless to get anything new out of the door. Zen3 is severely starved for memory BW and is also power limited (this can be seen even with Epyc) - Zen4 is the core which will unlock the full potential and perform the fatality even on the poor SR.
 
  • Like
Reactions: Drazick

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,656
136
That is great news for the buyers of HEDT but there is no competition to current TR parts so it's like beating a dead horse.
That's a terrible take to make as long as there are valid use cases profiting of the MT performance TR delivers. If it can speed up workloads it should be available, excessive amount of noncompetitive dead horses notwithstanding.
 

Mopetar

Diamond Member
Jan 31, 2011
7,831
5,980
136
I get that AMD wants to push new parts for more $$$$ but when you lead intel by such a HUGE margin then it's kinda pointless to get anything new out of the door.

From AMD's perspective they don't really gain anything by holding back since they'd just be using TSMC 7nm to manufacture any Zen 2 chiplets that would need to go into the older TR models. The die sizes are similar so it's pointless not to put out a new model as soon as they can.

Maybe the have a lot of excess stock of Threadripper parts to sell (it looks like they're all in stock on sites like Newegg) but anyone paying attention would know that Zen 3 models will come out eventually as well as how much of a performance bump that will offer. I don't think they sell too many older CPUs as a result of this.
 

eek2121

Platinum Member
Aug 2, 2005
2,930
4,026
136
From AMD's perspective they don't really gain anything by holding back since they'd just be using TSMC 7nm to manufacture any Zen 2 chiplets that would need to go into the older TR models. The die sizes are similar so it's pointless not to put out a new model as soon as they can.

Maybe the have a lot of excess stock of Threadripper parts to sell (it looks like they're all in stock on sites like Newegg) but anyone paying attention would know that Zen 3 models will come out eventually as well as how much of a performance bump that will offer. I don't think they sell too many older CPUs as a result of this.

Margins are lower on Zen 3 parts due to larger dies.
 

fkoehler

Member
Feb 29, 2008
193
145
116
"I get that AMD wants to push new parts for more $$$$ but when you lead intel by such a HUGE margin then it's kinda pointless to get anything new out of the door."

OK, so you want AMD to start acting like Intel has for the past 10 years?

Slacking off is the worst thing AMD could do, and I'm quite sure they are not going to.

Doing as you suggest means, AMD should cut its engineering staff in half, right? I mean, if you're going to slow-drip new product advances out, you certainly don't need the resources you currently have working at 100%....

I hated when Intel slow-dripped new technology so they could maximize profits at everyone else's expen$e. Can't believe people are on here crying for more of that shiz. Turn in your geek card.
 

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136
Anyone plan on building a TR3 system? I am curious if we will see a 16c/32t unit at launch, or just the same initial chip config as TR2 at launch.
 

Endgame124

Senior member
Feb 11, 2008
955
669
136
Isn’t that a bit late to release zen 3 threadripper? With Zen 4 coming mid next year, it would probably be better to cancel Zen 3 threadripper and launch Zen 4 threadripper a couple months early.
 

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136
Isn’t that a bit late to release zen 3 threadripper? With Zen 4 coming mid next year, it would probably be better to cancel Zen 3 threadripper and launch Zen 4 threadripper a couple months early.

Well, it is a different product, HEDT, with a different release cycle; products (CPU's in this case) are planned years in advance because the still have to be physically manufactured and gotten into the hands of consumers.

Zen2 TR came out 5 months after Zen2 desktop, but, that was a non-pandemic time :)


August 2021 seems like a reasonable time period for Zen3 TR, as, AMD looks to have had to fill large OEM mobile and APU orders for Zen3 products these past few months.

I am guessing demand will be quite high for it, if a lot of content creation shops are still on Zen1 TR and are now looking to replace them.

AMD also has to make the whole CPU before they can sell them, and, they need to bin the best / better chips than what is used for 5600X - 5950X as they need to use less power and clock really well.
 
  • Like
Reactions: Tlh97

dr1337

Senior member
May 25, 2020
331
559
106
Can't believe people are on here crying for more of that shiz.
Didn't sound like he was crying, besides zen 3 has a minimal enhancement to multi-core workloads. Really a company competing with itself is a bad thing and while yes 64 zen 3 cores are slightly better than 64 zen 2 cores, Its still 64 cores and still dominates intels HEDT lineup. Seriously If you don't understand the difference in perspective of intel being stagnate for a decade and being 5+ years behind on roadmaps, and AMD waiting to release zen 3 threadripper because they have no competition and already have extremely strong products on the market, then you can keep your geek card but I think you should turn in your logic, economics, and rationale cards should you have any.
 
  • Like
Reactions: Tlh97

Endgame124

Senior member
Feb 11, 2008
955
669
136
Well, it is a different product, HEDT, with a different release cycle; products (CPU's in this case) are planned years in advance because the still have to be physically manufactured and gotten into the hands of consumers.

Zen2 TR came out 5 months after Zen2 desktop, but, that was a non-pandemic time :)


August 2021 seems like a reasonable time period for Zen3 TR, as, AMD looks to have had to fill large OEM mobile and APU orders for Zen3 products these past few months.

I am guessing demand will be quite high for it, if a lot of content creation shops are still on Zen1 TR and are now looking to replace them.

AMD also has to make the whole CPU before they can sell them, and, they need to bin the best / better chips than what is used for 5600X - 5950X as they need to use less power and clock really well.
There is very little (note, I'm not saying zero) value proposition for a zen 3 TR that is released like <9 months before Zen 4. First, lets note that Milan EPYC chips have been available since Dec, so if you were already dying for Zen 3 in a many, many cores format, you probably already have moved to Milan.

Next, the Zen 4 top chip (lets call it a 6950) will have higher IPC and likely a small bump in clockspeed over Zen 3. On a clock basis, this likely means that the 6950 will perform about the same as a 5960. Zen 4 will also have DDR 5, so the 6950 will have higher memory bandwidth than a Zen 3 Threadripper (based on projections that DDR 5 is appx 2.2x faster than DDR 4).

Basically, you'll pay a lot of money for a Zen 3 threadripper only to have even the 5970 nearly eclipsed by the 6950 in around 6 months for 1/3 the cost.

Given the low value prop of Zen 3 Threadripper, if AMD could skip it and bump up the release of Zen 4 Threadripper to sell it as higher margin HEDT that releases before the consumer Zen 4, that would probably be best. It gets higher margins for AMD, and doesn't sell something to people that is obsoleted at a rate faster than we even saw in the late 90s.
 

blckgrffn

Diamond Member
May 1, 2003
9,123
3,056
136
www.teamjuchems.com
There is very little (note, I'm not saying zero) value proposition for a zen 3 TR that is released like <9 months before Zen 4. First, lets note that Milan EPYC chips have been available since Dec, so if you were already dying for Zen 3 in a many, many cores format, you probably already have moved to Milan.

Next, the Zen 4 top chip (lets call it a 6950) will have higher IPC and likely a small bump in clockspeed over Zen 3. On a clock basis, this likely means that the 6950 will perform about the same as a 5960. Zen 4 will also have DDR 5, so the 6950 will have higher memory bandwidth than a Zen 3 Threadripper (based on projections that DDR 5 is appx 2.2x faster than DDR 4).

Basically, you'll pay a lot of money for a Zen 3 threadripper only to have even the 5970 nearly eclipsed by the 6950 in around 6 months for 1/3 the cost.

Given the low value prop of Zen 3 Threadripper, if AMD could skip it and bump up the release of Zen 4 Threadripper to sell it as higher margin HEDT that releases before the consumer Zen 4, that would probably be best. It gets higher margins for AMD, and doesn't sell something to people that is obsoleted at a rate faster than we even saw in the late 90s.

Let's assume that the Zen 4 TR launch is relatively fixed as AMD roadmap is set for that time period and expensive/maybe near impossible to bring in.

A product refresh that has been long planned and is probably already in silicon and impacts EOL roadmaps down the road is just what you do.

The well oiled machine churns forward.

I mean, the marketing slide decks are already made?!? What about those poor marketing interns that put their heart in souls into them? You just going to put them in the digital recycle bin!?! ;)
 

Endgame124

Senior member
Feb 11, 2008
955
669
136
Let's assume that the Zen 4 TR launch is relatively fixed as AMD roadmap is set for that time period and expensive/maybe near impossible to bring in.

A product refresh that has been long planned and is probably already in silicon and impacts EOL roadmaps down the road is just what you do.

The well oiled machine churns forward.
Sure, they certainly can and will release it. It's just bad value for almost anyone to actually buy it, unless you just absolutely have to have it now and can't wait a few more months.
 

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,656
136
There is very little (note, I'm not saying zero) value proposition for a zen 3 TR that is released like <9 months before Zen 4.
And what was the value proposition of the Zen 2 based Threadripper Pro 3000WX series that launched to the open market only this March, so two months ago?
 

Vattila

Senior member
Oct 22, 2004
799
1,351
136
what is the market segment for threadripper pro that can’t use Milan Epyc?

Level1Techs takes a look at the EPYC difference and usefulness of Threadripper Pro in the following videos. One particular difference noted is that the motherboard platforms for EPYC are ill-suited for a desktop workstation.

Personally, I want at least 16 Zen 3 cores in my upcoming upgrade, and I might consider Threadripper and higher core count, if available with Zen 3 cores. The Zen 3 attraction is the "no compromise" on single-thread performance.


 

dnavas

Senior member
Feb 25, 2017
355
190
116
Isn’t that a bit late to release zen 3 threadripper? With Zen 4 coming mid next year, it would probably be better to cancel Zen 3 threadripper and launch Zen 4 threadripper a couple months early.

I tend to believe that PCIE5/CXL and DDR5 is of more interest than XX% improvement in IPC. Maybe if floating point gets a huge jump in Zen4 it would be of interest, but heterogenous computing (GPU / TPU) seems like a fair workstation use-case. YMMV.
 

Shmee

Memory & Storage, Graphics Cards Mod Elite Member
Super Moderator
Sep 13, 2008
7,400
2,437
146
I my interest in the TR platform would be not a ton of cores, but rather a larger amount of lanes for more SATA and M.2 storage. Mainstream platforms are currently very limited. Ideally, I could see myself buying a 12 to 16 core Zen 4 TR, with DDR5 and PCIE5, and tons of SATA and m.2, and PCIe slots.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
I tend to believe that PCIE5/CXL and DDR5 is of more interest than XX% improvement in IPC. Maybe if floating point gets a huge jump in Zen4 it would be of interest, but heterogenous computing (GPU / TPU) seems like a fair workstation use-case. YMMV.
Under what circumstances is pci-e 5 going to actually make a meaningful difference? I think it is still up in the air how much the extra speed is going to help even HPC right now. Most highly optimized applications already overlap compute and data transfer and work on large enough chunks of data that it just isn’t as useful as people might expect. PCI-e 4 is already very fast.

At this point, Zen 3 based Epyc parts seem to be more available than Zen 2. This might just be that companies do not want to deal with brand new boards, bios, etc. I would expect AMD to put as much of their wafer allocation into making Zen 3 chiplets, so there may be a lot less Zen 2 chiplets available going forward. That is assuming that there is some overlap in production. I am assuming that it is the same process. They almost certainly make more money off a Zen 3 chiplet than Zen 2. Milan is more expensive that the equivalent Rome part.

I would expect a lot of threadrippers to be purchased as graphics workstations, not really just high end gaming machines. For enthusiast, buying an end of life platform doesn’t seem too good, but if it is a company or just someone who needs a workstation now, then that is different. Some enthusiast will probably want to wait for Zen 4. Myself, I generally don’t like getting revision 1.0 of everything, so I might just still get Zen 3 in this time frame. Zen 4 threadripper might also be a while, so zen 3 threadripper might still be the way to go if you can use a lot of cores for quite a while after Zen 4 comes out in Genoa or non-threadripper Ryzen parts. Threadripper requires a lot of high clocking chiplets with reasonable power consumption.

If it is launching due to lower production of Zen 2 chiplets (they have to start phasing it out at some point) then this might be launch it in august, or Zen 2 based threadrippers just go out of stock with no replacement.
 

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,656
136
Under what circumstances is pci-e 5 going to actually make a meaningful difference?
PCIe 5 alone is just an evolutionary improvement. It combined with CXL will be a paradigm shift as it will allow for cache coherence across all parts in the system as well as networked RAM etc.
 
  • Like
Reactions: dnavas

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136
I hate to invoke auto's, but too many posters are getting hung up on HEDT / Server vs AM4 / AM5.

HEDT / Server is like the F250 / F350 trucks.

AM4 / AM5 is like the F150, with the 5950X being like the Raptor version of the truck.

So, 3 different segements of CPU's with different market availability targets and release schedules.
 

Atari2600

Golden Member
Nov 22, 2016
1,409
1,655
136
I'd argue those saying Zen3 is a poor value proposition for Threadripper due to incremental MT improvements have clearly never worked in an environment where many Threadripper Pros see service.

When you look at a typical engineering application for these, CFD, not all the workflow is embarrassingly parallel. There are chokepoints, particularly in meshing and post-processing, where ST performance is critical.

Now - some would say you submit your work to the TR and do the front and back end stuff elsewhere - to which my response is; if your company is big enough for dedicated run stations, its unlikely your looking at a single socket TR anyway. Most of these would be user dedicated workstations IMO - which means ST performance is a massive component of overall productivity.
 

Asterox

Golden Member
May 15, 2012
1,026
1,775
136
If the information is correct, "Threadripper 5000 coming soon".