Threadripper 29xx, Ryzen (AM4) 2800 Discussion Thread

[eek2121]

Obviously, we are likely still a few months out. However, I'd like to get a thread going for the Threadripper 2xxx and the Ryzen 2800x

First, let me start out by specilating my own personal thoughts about the Threadripper lineup and 1800x:

• There was an 1800x, but the 2800x is MIA. AMD isn't going to ditch the x800 moniker because it was considered flagship and did NOT sell poorly.
• The 1700X -> 2700X had a +300 MHz jump for the base and an astounding 550 MHz increase for the boost. This may or may not include XFR 2.0. I expect it doesn't since most of the tests were run on X370, which means that with XFR 2.0, you'll see 3.7/4.5. However, this likely will not be all core boost. Expect a massive improvement in gaming thanks to improved cache and memory latencies, and a better IMC.
• So where does that leave the 2800X? I believe that the 2800x will pick a spot right below the 2900X TR4 chip. We'll see a 3.8/4.7 chip with 125 watt TDP or something similar. This is just speculation on my part. Note that my 1950x can run 4.2 GHz on 1 die at 1.425 GHz It can also run 4.3 stable at around 1.5V. Doing napkin math tells me that we'll see something like this:
• The 8 core 2900X chip I expect will have slightly better clock speeds this time around. Something like 4.3/4.7, 8 cores.
• The 12 core chip will follow with 3.8/4.7 GHz
• The 16 core chip will follow with 3.6/4.7 GHz.
• There might be a 24 core chip. It would be around 3.2/4.7
• All of these assumptions assume that the boost is not all core. If the boost is not all core, we'll see much lower boost clocks.
• All this is going to boil down to TDP. The 2800X I expect will be a mid year refresh. I also expect it to have a 125 watt TDP.

I could be completely wrong about all of it. I will be able to provide better answers once actual 2xxx chips start making their way to reviewers.

What I CAN tell you is this. These chips are going to manage clock speeds much better, so you won't see a 10% improvement in a workload, but rather, you will likely see 20-40% improvement in workloads.

 

[Markfw]

All I can say is that all sounds logical and I HOPE so, and that will be my Threadripper number 5 and 6 and....

 

[VirtualLarry]

All I can say is that all sounds logical and I HOPE so, and that will be my Threadripper number 5 and 6 and....

I dub you "ThreadRipper KING of AT Forums"!

 

[piesquared]

Nobody is expecting a 2800. The leaks indicate 2700 as the top AM4 SKU, which aligns nicely with the competing SKU's. Do we need another Ryzen thread though? There are plenty already. Maybe just get someone to rename one of the other ones, that's how the intel thread starters roll. It's all about the view and post count!

 

[Mopetar]

Nobody is expecting a 2800.

Yeah, I don't think we see one any time soon. Maybe one eventually gets released months from now as a bit of nice filler between now and the eventual Zen 2 release. That gives AMD some additional time to work with the process along with the opportunity for a nice mid-cycle press bump.

If it ever does get released, I expect it to coincide with price cuts. Even if it has the characteristics described by eek, I can't see AMD releasing it at the same price as the 1800X. My guess is that it slots in much closer to what the 2700X is retailing for and pushes everything else down.

 

[whm1974]

Supposedly AMD is going to release a 12c/24t AM4 Ryzen CPU, and I have to ask, but would dual channel 3200 memory have enough bandwidth to feed such a CPU?

 

[darkswordsman17]

I wish they'd have reworked their naming scheme so it would be straightforward. Make the even numbers base chips with lower TDPs, and then odd numbers the performance ones with higher TDPs. Let the consumer line have the full scheme (so two 8 cores this time, with one being a 65W 2800, and the other being the 105W 2900). Have Threadripper do something different. I'd say just have X2 at the end, but that would probably be viewed as making the lower Threadrippers seem worse than the higher consumer stuff (2400X2 vs 2800).

I'd actually guess the 2800 is going to be a Threadripper chip. It seemed odd to have all of them sharing the same 19x0 designation. Have the 2800 series be the base chips, maybe they'll make ones that run at a lower TDP or something even (140-160W?), and then let the 2900 series go to 200W? Assuming there's extra clocking to be had with the higher thermal headroom). Something that might be more suitable for OEM configurations, and then the 2900 would be the gaming/overclockers/robust OEM config setups.

Rampant speculation: maybe get 6 chips, with a low 8 core one starting at $499, but aggressively sits in a lower TDP. Then a higher one that is $150 more, and it comes with an AIO cooler, and 24MB L3. Then a 12 core with again aggressive TDP (but a bit higher than the 8 core) at $650, and $150 for the higher part (putting it at $800), with a beefier AIO (maybe 240mm with dual fans). Then the 16 core is the full fat one at $800, and the 2900 version being $1000 but with 360mm triple fan radiator. Maybe make it a bundle option at launch and then stop offering it but drop the price later (thus keeping margins up). Or maybe at launch not include a cooler but then add it in instead of dropping the price later.

 

[darkswordsman17]

Supposedly AMD is going to release a 12c/24t AM4 Ryzen CPU, and I have to ask, but would dual channel 3200 memory have enough bandwidth to feed such a CPU?

I don't see that happening, because its not like Threadripper where it actually has the socket of the full 4 module Epyc line, but can't imagine they could just slap another module in there without some hefty reworking (and not worth it for a single chip since it would likely not be able to clock as high due to thermal and/or power constraints). I suppose it would be possible (after all they had FX chips on probably worse designed boards), but just sounds like a recipe for disaster and too much work for such a configuration that likely won't be that big of a seller. Granted, I haven't been paying a lot of attention/seeking info, but I think that talk (of 12c/24t consumer Ryzen chips) had to do with Ryzen 2 on 7nm, and was in line with discussion about them adding a third CCX or upping the CCXes to 6 cores per. I would imagine they'd already be planning on finding ways of adding more memory bandwidth in future chips anyway (either they'll be pushing for higher bandwidth chips, maybe DDR5 support, or possible going to quad channel or maybe triple channel memory configs).

 

[eek2121]

All I can say is that all sounds logical and I HOPE so, and that will be my Threadripper number 5 and 6 and....

I suspect he eat's them for breakfast.

 

[eek2121]

Yeah, I don't think we see one any time soon. Maybe one eventually gets released months from now as a bit of nice filler between now and the eventual Zen 2 release. That gives AMD some additional time to work with the process along with the opportunity for a nice mid-cycle press bump.

If it ever does get released, I expect it to coincide with price cuts. Even if it has the characteristics described by eek, I can't see AMD releasing it at the same price as the 1800X. My guess is that it slots in much closer to what the 2700X is retailing for and pushes everything else down.

Oh the 2800X definitely won't be available at launch. However, a mid-cycle release would be my bet as well. Note that these clock numbers aren't pulled out of thin air either, my 1950x can do 4.2 @ 1.425, a 10% voltage improvement puts that at under 1.3V. The boost clocks may be a bit lower than predicted, but I think they want a flagship product to actually *be* flagship. I have yet to see a 1950X that can't hit 4.1 with proper cooling, LLC management, etc. so they ARE higher quality cores. AMD is likely just going to work on more effective binning this time around...and I fully expect them to charge a premium for it. If you end up with the fastest CPU on the market, you can charge what you want for it. My prediction for using the cores reserved for Threadripper comes from a) They probably had a lot more quality cores than sales volume would support and 2) While the 1800x DID sell, the best selling 8 core was actually the 1700, which could usually overclock pretty close to the 1800X. By creating an untouchable 2800X, you entice users to buy a higher quality product. The issue is that you won't have these 'quality dies' at launch, you build them up over release, hence a mid-cycle release. If the 2800X outperformed an 8700K by 10-20% would YOU buy one? I'm sticking with my TR4 platform of course.

 

[exquisitechar]

What I CAN tell you is this. These chips are going to manage clock speeds much better, so you won't see a 10% improvement in a workload, but rather, you will likely see 20-40% improvement in workloads.

What are you basing this on? Depending on the workload 20-40% improvement over first gen Ryzen CPUs is possible and some of the leaks indicate this as well, but I think the average improvement will be below 20%.

I think your predictions for the clocks might be a bit too optimistic, but I can see a 2800X model being released later on as well. I certainly hope so.

 

[LightningZ71]

Putting two real packages in the dummy die slots on a Threadripper as it is currently arranged won't be of any use to anyone. All of the I/O of the Zepplin dies in the current chips is taken up with the links between the two active dies. About the only thing that could be done to improve performance would be to use those die slots for some sort of high speed memory package that then routes to one of the memory channels on the Zepplin dies in the existing package. That is technically VERY difficult to make any faster than external memory as both channel controllers are configured from the same registers. I could MAYBE see some sort of embedded product that has ALL of it's memory on the package, and that has all of the memory channels pointed at the dummy dies and no external memory interface. It's incredibly restricting to be sure, but, with extremely tight tolerances on both the manufacture of the memory dies and the channels that connect the two, I could take a scientific wild ass guess and say that its in the realm of possibility that they could configure them to communicate at well beyond DDR-4000 numbers with very low CL numbers without stretching reality. There is a limit to what the memory controllers on the die are physically able to do and pushing beyond that would take major work inside the die that I can't see AMD investing in anytime soon.

 

[sandorski]

I think they realized that the 1800x didn't make much sense when the 1700x and 1700 pretty much were just as capable when overclocked to 1800x speeds. A 2800x makes more sense if it can bring something unique to the table.