3rd Gen Ryzen based Threadripper could be released in October - up-to 64 cores

#5692145 Posted on: 07/18/2019 08:03 AM

---

Intels reaction

#5692146 Posted on: 07/18/2019 08:15 AM

---

Reserved for meme.

I am on the toilet, I'll post meme asap
Amateur. Real men create meme during the process.

#5692147 Posted on: 07/18/2019 08:19 AM

---

I'll bet quite a few people who dont need it will buy it (gamers).

#5692158 Posted on: 07/18/2019 09:31 AM

---

Reserved for meme.

I am on the toilet, I'll post meme asap
Waiting for it.

#5692167 Posted on: 07/18/2019 10:10 AM

---

AMD already hinted that 3rd gen Threadripper going up-to 64C, listen Lisa Su and other AMD stuff - they all use : "....Move Up, Up" this refers that this Gen will move 2 steps above current 32C limit on TR, thus there will be 48C and 64C.


The interesting part will be the Chiplet/Cache config for 16c~32C parts:
16C 64MB or 128MB with 4 chiplets
32C 128MB or 256MB with 8 chiplets
In Epyc lineup they offer all the configs with diff cach sized due to num of chiplets used.

Also- looks like the Chipset will use the large I/O die from Rome- how they gonna cool 25~50W chipset?

P.S: Intel is already reacting in secter to upcoming EPYC/TR with "new" 240W TDP parts for extra 50% price increase:
https://ark.intel.com/content/www/us/en/ark/products/193956/intel-xeon-platinum-8284-processor-38-5m-cache-3-00-ghz.html
https://ark.intel.com/content/www/us/en/ark/products/192487/intel-xeon-platinum-8274-processor-35-75m-cache-3-10-ghz.html
https://ark.intel.com/content/www/us/en/ark/products/192458/intel-xeon-gold-6269y-processor-30-25m-cache-3-20-ghz.html

#5692209 Posted on: 07/18/2019 01:10 PM

---

*Does the happy dance*

#5692219 Posted on: 07/18/2019 01:53 PM

---

Imagining how wonderful those cores would look in taskmanager

#5692259 Posted on: 07/18/2019 03:26 PM

---

If AMD wants to battle Intel in HEDT, they just have to keep doing what they've been doing with AM4: disintegrate the HEDT market by making many-core CPUs affordable on mainstream platforms. As long as Intel uses a monolithic die, they can't really do this. Besides, with the rate AMD is going, there's hardly any incentive to go for TR anymore, since whatever you get will become obsolete in a couple of years.

#5692268 Posted on: 07/18/2019 03:51 PM

---

I'd say that's the way it should be. As an enthusiast, I despaired when I saw the poor upgrade options three years after buying a 4790K (the 7700K, which was about as mediocre an upgrade as one can imagine). It's this frustration which ultimately led me to abandon Intel for AMD, and I do not want AMD going the same route. Like Matisse, I expect a major jump in performance with third-gen TR and a worthy upgrade option, making previous gen products completely obsolete.

#5692270 Posted on: 07/18/2019 03:54 PM

---

I don't think we'll see AMD raising the number of cores for a while now. The current design is already pretty robust, and aside from another die shrink, there's only so many cores that can fit the physical space of a die. Zen 3 will (probably) be a refresh aimed at better performance per core and better latency, with probable improvements to the chiplet design. And the future after that is a mystery really.

So I think investing in a TR CPU that uses Zen 2 would be good for the next 3 years at least. They might release a TR based on Zen 3, but will probably only have better cores, not more.

#5692277 Posted on: 07/18/2019 04:14 PM

---

It's one thing to want upgrade options, but to me, it leaves a bitter taste when you buy a product that is trounced in almost every conceivable way for a fraction of the price in a very short timeframe. Since TR was never really intended to be a budget option, I'm sure owners of it don't care as much about obsolescence, but I guess what I'm getting at is knowing AMD's recent track record, it doesn't really make sense to go for TR considering you could just wait a couple years for something faster and more efficient at half the price. If you need more cores now and price doesn't matter that much, might as well go for Epyc.


That's what we've been saying the last 2 years, and yet, we kept getting proven wrong. Nobody was expecting AM4 to have 16 cores, and yet, it is confirmed. I'm not at all expecting 32+ cores for AM4, but, I bet we will see that for AM5.

Possibly. There are enough differences with Zen2 (with the IF and chiplet design) where some of TR's shortcomings might be resolved. Since the socket already exists, AMD might as well use it. Besides, the additional PCIe lanes and more consistent support for ECC gives Threadripper a bit of an edge.

#5692287 Posted on: 07/18/2019 04:37 PM

---

But that was because of the jump from 14nm to 7nm, which only happened fast because AMD went 14nm when it was more or less already mature enough and 7nm was close. 7nm is relatively far newer, and will still see a few refreshes before 5nm starts to become reasonable to use. So it will probably take a while before we see another jump in number of cores like that. Again, die space is already well used, and each Zen 2 "core" probably won't change enough to allow more cores to fit in the existing space.

They may be able to fit more cores if they make the IO controller 7nm as well, but that's kinda overkill and I don't think will happen on Zen 3. Even if that happens, AMD is trouncing Intel in number of cores now, so there's little incentive to raise the bar that much more. I mean, come on, do we really need more than 64 cores on anything other than a server? I think AMD will calm down now :p


I think Zen 2 TR will have insane scalability, if current Ryzen is any indication. Just look at most multi-threaded benchmarks of the 3900x - it scales almost linearly when compared with the 3700x. So, the IF is much more efficient and doesn't really mismanage resources between cores as much (and maybe windows scheduler isn't fumbling as much?), meaning that things like "Game mode" won't even be necessary anymore. I would say it will be a very nice buy for those that need as many cores as they can get.

#5692289 Posted on: 07/18/2019 04:50 PM

---

Do I need it.... No.... But that's what they said in 2009 when I bought i7 970 6core (4ghz OC)and 24gigs!!! That i rocked till Ryzen Sidestep upgrade 2 months ago(2700x and 64gb)
It's fast, But will probably upgrade to TR Zen2 64core ASAP that i'll probably rock Foreseeable 5 plus years...

#5692292 Posted on: 07/18/2019 04:54 PM

---

Now AMD is really paving the way for the future, Intel has nothing to fight this short term.
Home, Workstation and Server, AMD CPU division has everything covered.
Now if only the GPU division would catch up...

#5692297 Posted on: 07/18/2019 05:02 PM

---

Transistor size is irrelevant (besides, AMD went from 12nm to 7). The 16 cores for 1st gen TR was a surprise. 32 cores for 2nd gen was a surprise, and there wasn't a major die shrink. 64 cores for Epyc was a surprise. 12 cores for AM4 was a surprise. All of these were surprises specifically because people had expectations just like what you said right there.
All that being said, normally, I'd totally agree with you, but just when we're like "nah, there's not enough room" they somehow make more room, and a die shrink isn't the only factor allowing them to do that. I'd like to agree with you that refreshes and expanding dies will slow down, but every time I said that, I was wrong.

Yes, I agree with all of this. Like I said before, I'm not expecting AM4 to get any more big core counts, because it's reaching EOL next year. I'm sure the last products released for AM4 will be the Zen2-based APUs. Frankly, I think 16 cores is still overkill for the average user. But, as long as people think more cores means universally more performance, it's a good marketing ploy and works against Intel.

Right, this is why I was saying the improvements in IF and the chiplet design might help overcome TR's shortcomings. But the key word here is "might". For the existing TR models, the greatest issue is how they're basically 2 separate CPUs with their own memory controllers crammed on the same die that can talk to each other. The chiplet design, in theory, should obsolete that (in which case you'd be right that "game mode" won't be necessary).
Also last time I checked, the Windows scheduler was still pretty numb. MS really needs to get their act together, because other OSes got their schedulers figured out.