The end of THREADRIPPER?

[LePhuronn]

Sound reasoning, smilingcrow. I disagree though. It's worth clarifying here that I am talking purely about Ryzen 3000 vs Threadripper 3000 and the opening point of this thread that somehow a 16c Ryzen marks the end of Threadripper.

I'm also not talking about HEDT users downgrading to AM4. I'm saying that the crossover point between a top-end 16c Ryzen 3K and entry-level 16c Threadripper 3K is incredibly small, and will only be a consideration for those customers who only want a high core count for their hobbyist/amateur/prosumer work. That is where 16c Ryzen will eat into Threadripper sales because Zen 2 now makes the platform choice for you: if you're a hobbyist or prosumer video creator (for example) then there's no need to drop thousands of pounds on HEDT because mainstream AM4 has you covered.

But that is the only instance.

If you don't want 16 cores, you won't even be looking at HEDT. If you do more with media creation than be a YouTuber, Twitch streamer or make wedding videos then you know what hardware you need to work with that 4:4:4 8K RED footage, the 20 minute 3DS Max, Maya or Renderman sequence, the multi-layered Avid NLE timeline with real-time colour grading, the fluid dynamics and physics simulations (on the cheap ). Any of this and you won't be looking at AM4.

Now I kinda agree with you when you start bringing previous generation Threadripper into the equation the waters aren't so clear. Do we drop £500 on a 16c Ryzen 3000 and £200 on a X570 board to run it, or £500 for a 12c 2920X and £350 for a board with a boat load of extra kit which may be useful in the future? Again, if it's only the cores I want then it's Ryzen all the way - 4 more cores and cheaper - and a lost Threadripper sale. But that's a previous generation Threadripper, so is it really relevant to the "end of Threadripper" debate?

And as an aside, I don't think AMD would need to sacrifice Threadripper in any way in order to annihilate Intel on the mainstream platform. I've argued many times that you can make up a beastly Threadripper using comparatively duff chiplets. The good-to-great chiplets remaining after EPYC has cherry picked the frugal ones can be used to make 4.6-5GHz Ryzen desktop monsters and batter Intel evry which way from Sunday, and those "duff" chiplets with only 4 cores working can still get strapped together into a 32 core Threadripper.

Such versatility is the beauty of this chiplet design: AMD almost literally do not have to make concessions on their product lines.

 

[LePhuronn]

I think they'll keep TR but have a slower release cycle. Releasing a new TR chip every year is probably not very profitable, the market for HEDT chips is tiny in comparison to desktop chips so developing/marketing a TR chip probably isn't great if they have to do it every year. However, a 2 year release cycle would mean that the development/marketing budget then gets two years worth of sales. If you want a 32 core CPU you can still get the 2990wx and that can happily remain the highest core count CPU until the 4xxx ryzen series.

What development? Threadripper is (largely) just EPYC with half the bits turned off. Whether Threadripper 3 uses an EPYC Rome I/O die with half the RAM channels disabled, or has a bespoke 4-channel IO remains to be seen, but it's hardly a herculean effort to do the latter. And also, it's been reported that a lot of EPYC's chipset features are coming down to Threadripper too, so there's even more similarity between the 2 platforms, and therefore less "development" to do for the latter.

 

[Stephanie Peterson]

Same could be said for the Xeon workstation HEDT platforms though, they are just Xeon server stuff..

Still TR will continue, i would like to think it gets some 48 and then next revision 64 cores and eventually maybe...6 or 8 channel ram. amazing how such a major processor lineup nearly didnt get made as it was never in the original plans

 

[smilingcrow]

I'm saying that the crossover point between a top-end 16c Ryzen 3K and entry-level 16c Threadripper 3K is incredibly small.

I'm sure TR is a fairly small market anyway in the grand scheme of things so without doing a focused poll or two we can't say for sure if you are correct or not.
I don't claim to know the percentages but you clearly have a strong opinion on this and laughed when I suggested a poll.
So that is where we disagree in that I don't take subjective guesses/opinions too seriously when it comes to statistics but prefer objective data.
At least I can see the misunderstanding now. Have a good one.

 

[LePhuronn]

Oh I didn't laugh at the notion of a poll, I raised issue with the contradiction of wanting to base an argument on data and then immediately made a second point with nothing more than assumption.

In any case, it's nothing more than a difference of opinion that remains nothing more without empirical data to back it up.

 

[smilingcrow]

In any case, it's nothing more than a difference of opinion that remains nothing more without empirical data to back it up.

Well that's your take on it.
For me it's more about seeing the difference in how people relate to their own hype.
Some take it as being real and others treat it more as mere possibilities with degrees of probability which are currently unknown.
The difference is night and day but too nuanced for some or probably a lot based on my current perceptions.
Although I don't have a probability quotient for that.

 

[pmc25]

Threadripper fills the HEDT platform niche. They're hardly going to can it when Zen2 TR4 will likely expand their absolute performance lead over Intel 'X' to virtually every realm, all the while with increasing core counts meaning a further widening in their cost and yield advantage. The differential in power draw and heat output will be comical too.

Remember, TR4 also offers a lot more RAM bandwidth and totals, and vastly more PCI lanes than AM4. Plus, 16 core was only the mid tier on Zen+ TR4.

Likely it launches 6-8 weeks after Ryzen 3, just like the earlier ones.

 

[Panos]

Regarding several posts above.
AM4 supports ECC, and especially all the Asrock boards (maybe not the A320)
PCI-e 4.0 at 8x has as much bandwidth at the PCI-e 3.0 16x. So on the new series boards full fat dual 3.0 16x is possible.
Look at CVS (Colorfull) X570 motherboard design is interesting one.

If AMD was planning to kill TR4 with AM4, AMD needs to give (to AM4) multi raid NVME and raid boot. More than 16 cores (current design impossible), 64+ PCI-e lanes and we should have had started seeing ETX boards leaking.

Quad channel ram is irrelevant.

 

[Beren]

I suspect AMD are forecasting significant Ryzen 3k and Epyc demand and know that to really hammer Intel they need to be able to supply solid stock. TR is a niche product and they might not want to divert short term stock. I suspect TR will follow in Q4 or as soon as production ramps and they know they can make it. A 7nm TR may well enable AMD to offer a better performance and price in every single product stack and I dont see them missing that trick. We are just going to need to wait a little longer.

Also if they are going to release a 16c32t Ryzen 3800/50x those could be exactly the super top dies that TR also wants, if that top top segment is not putting out quite as many chips as they thought they may have chosen the high end gaming desktop chip to have the most marketing power and be building stock in that area.

 

[LePhuronn]

Also if they are going to release a 16c32t Ryzen 3800/50x those could be exactly the super top dies that TR also wants

You don't need super top dies for Threadripper. You could conceivably make an entry-level 16c/32t 4.5GHz all-core boost Threadripper from "junk" dies. And I'd be sceptical that you could get an acceptable TDP out of 48c/96t boosting to 4.8Ghz and above. Just because previous gen 'rippers used the best silicon doesn't mean the 3rd gen 'rippers will or even need to.

But yes, there are more lucrative markets to populate with chiplets first, Threadripper won't be a priority.

 

[sandys]

You would need top dies to maintain reasonable TDP with more cores, I would imagine EPYC is taking the cream, desktop getting what doesn't make the cut and TR as somewhere to sell your re-purposed server chips for lesser margin but not as low as desktop when demand tails off, I don't imagine TR is dead, I imagine the case is that they don't make enough 7nm die yet to satisfy their inroads in to the server markets which is where the money is.

Releasing TR early could cannibalise the EPYC 1P/2P workstation market.

 

[Deleted member 66701]

You don't need super top dies for Threadripper. You could conceivably make an entry-level 16c/32t 4.5GHz all-core boost Threadripper from "junk" dies. And I'd be sceptical that you could get an acceptable TDP out of 48c/96t boosting to 4.8Ghz and above. Just because previous gen 'rippers used the best silicon doesn't mean the 3rd gen 'rippers will or even need to.

Threadripper has always had the highest boost in the threadripper line, so it'd be a fairly significant shift in strategy if that wasn't the case going forward.

 

[LePhuronn]

Threadripper has always had the highest boost in the threadripper line, so it'd be a fairly significant shift in strategy if that wasn't the case going forward.

And it's entirely possible that "junk" 4 core chiplets can boost up there with the golden 8 core chiplets going into the 3700X and 3800X, so you still have entry-level 16 core Threadripper matching the top-end Ryzen's boost without diverting prized silicon away from more lucrative markets.

I don't think people are fully grasping the scope and versatility that chiplet design brings, especially with a highly granular binning process. The concept of "the best silicon" is much more nuanced now.

 

[smilingcrow]

I wonder what the difference in performance will be between a 2x8 v 4x4 chiplet design with TR3?
It needs to be small otherwise a Ryzen 3 2x8 might beat a TR3 4x4 which wouldn't be ideal.

 

[sandys]

I wonder what the difference in performance will be between a 2x8 v 4x4 chiplet design with TR3?
It needs to be small otherwise a Ryzen 3 2x8 might beat a TR3 4x4 which wouldn't be ideal.

It happened before, the Ryzen 1800x could be quicker than the 1900x despite 1900x having the best gen 1 clock speeds because of cross die issues, it was small though.

 

[Poneros]

Unless they increase PCI-E lanes on the AM4 platform, TR still has a place. Or if they kill it off then they'll send Intel customers back.

It would be a big & disappointing mistake if that's how they'd treat TR.

 

[smilingcrow]

I doubt many people really think it will be cancelled.
It's not a high priority I imagine so considering how many other new products they have coming out soon on 7nm something had to take the back seat.
They have already pushed it to 32C so it's not exactly lagging.
Nothing to see here.

 

[pmc25]

IT ISN'T BEING CANCELLED.

None of the sites printing the clickbait headlines think it is either.

 

[Scougar]

Threadripper has also been a massive marketing win for AMD. They would be unlikely to cancel such a strong presence for their brand building.

 

[Kelt]

Its got too cool a name to be cancelled.