Those Adored leaks/speculation are inline with my 16c "dreaming" Ryzen scenario I posted in the other thread. My gut was telling me that was the aggressive AMD move and it's crazy that it looks to be possible. What a world we live in if it's true. I was in for 16c before. I wasn't sure about CCX, but I believe we will see an 8C CCX. That's my vote.
For me it was the 3700X and RX 3080. RTX 2070 performance at $250 would be awesome. I'm not crazy about their GPU naming though. It's looks an awful lot like what nvidia would use for their next gen (RTX 3080 vs RX 3080) .
You dont need 16/32 for that efect, 12/24-4.2ghz stock will do the same or eat i9 9900K for breakfast.For 330$ that kind of CPU, well if this is right and price is right, this CPU price-performanse is absurdly great.
Those clocks are remarkably unrealistic and I'm not convinced by the SKUs/pricing either. Looks like a modified version of that fake from reddit that was probably repackaged and sent to Adored, I don't know why he dedicated a video to this trash...
Looking at the AdoredTV... it could be possible that AMD is not only cutting the "i/o die by 4" but is also cutting the CPU chiplets to produce, 4C 6C and 8C chiplets? I really dont see AMD producing more than 1 CPU chiplet.
Those clocks are remarkably unrealistic and I'm not convinced by the SKUs/pricing either. Looks like a modified version of that fake from reddit that was probably repackaged and sent to Adored, I don't know why he dedicated a video to this trash...
He said to take with a pinch of salt. There are still a few very good ideas there and if you add PS5/XB2 to that speculation then it looks very good from a manufacturing point of perspective. Next gen consoles are high volume parts and if they both would use full 8C chiplets (with lower clocks) there would be many 6C chiplets (with up to 2 faulty cores) left to sell as lower end Ryzen 3000 6C or and higher end 12C.
16C (AM4) would use fully working high clocking 8C chiplets that could be used for EPYC/TR (or PS5/XB2 if they don't clock very high) so AMD might not want to sell too many of those 16C AM4 CPUs (at least not for too cheap) and is pricing them therefore higher.
If you look the already posted video at 7:23 it mentions that 8C with no iGPU would be in fact two 4C chiplets (with up to 4 faulty cores each):
You don't have to add PS5/XB2 to this equation but it would certainly help a lot to bin those chiplets much more effectively with higher volumes. It would make a lot of sense. And since PS4/XB1 has set the CPU performance bar so low, a single 8C chiplet would be a massive improvement.
The most interesting speculation, that I myself have also thought about, is the possibility to move memory controllers of a lower end Navi chips to a separate (12/14nm) IO die. HBM2 is still pretty pricey but that would still open many nice possibilities for scalability and customizability. I might get back to that later on with some illustrations. I really like the fact that AdoredTV went with this even if it turns out to be (partly) wrong.
Addition: I'm still happy with my original choices of a 16-core modular CPU (1-2x 8C chiplets (7nm) and an IO die (12/14nm)) and a 4-core monolithic APU (Picasso 12nm) but if AMD is really pushing this chiplet design so hard (as AdoredTV seems to think) then 8-core modular APU would make more sense with a separate Navi chiplet. I'm not changing my answer for now (because I already went with it) but all of this does make very much sense now if it turns out to be true. It's still just rumours and speculation but very interesting in any case.
Addition 2: Sure it might seem that AdoredTV is setting expections too high (clock speeds and pricing) but the basic idea behind chiplets looks better and better all the time. I really hope this turns out to be true even tough integrated memory controllers do have their benefits. We'll see soon enough.
Those clocks are remarkably unrealistic and I'm not convinced by the SKUs/pricing either. Looks like a modified version of that fake from reddit that was probably repackaged and sent to Adored, I don't know why he dedicated a video to this trash...
I'm also going 12 core. I think 16 core will hit their Threadripper (HEDT) sales too much. It would also be getting to the point about why people need 16 cores on a ryzen cpu. If your work load needs that many cores you should probably be on a HEDT platform anyway.
It's possible they may even stick to 8 core if they can get big IPC and frequency gains.
I'm also going 12 core. I think 16 core will hit their Threadripper (HEDT) sales too much. It would also be getting to the point about why people need 16 cores on a ryzen cpu. If your work load needs that many cores you should probably be on a HEDT platform anyway.
It's possible they may even stick to 8 core if they can get big IPC and frequency gains.
Do I think it will be an uplift? Yeah, and a very strong one. But people need to temper their expectations. At least then you can be pleasantly surprised if they are exceeded.
Cranking up expectations to "ZOMG! Over 9000!", then when they 'only' deliver 8500.... AMD sucks... Brrrr, all these tech wienies said they'd get more etc... Pay attention to what AMD says. Speculation? Not so much. AMD is expecting a %13 uplift over Zen+. Damn good. And hey, they may be sandbagging. They did on Ryzen. But don't base expectations on that. Leave room to be happy. ;-)
Those clocks are remarkably unrealistic and I'm not convinced by the SKUs/pricing either. Looks like a modified version of that fake from reddit that was probably repackaged and sent to Adored, I don't know why he dedicated a video to this trash...
if you watch his chiplet videos, he covers a study where the use of chiplets increases the yield and the ability to bin since fast chips with defects are still salvageable. the yield curves shift towards much higher frequency when you are able to combine more than one chiplet. the ability to use the slowest ones for ps5/xbone2 means they get much higher volumes of the cherry picked dice.
his sources + the reddit leak plays into his own theory, so even with the rumor caveats it is reasonable enough with the chiplet bin math. he doesnt want to believe it himself since it is so conveniently in line.
please note he indicates that the r9 16/32 will require a new socket. so the 12 core is likely the last am4 compatible chip.
I don't think there's any chance of that. Based on the AMD figures about the TSMC process, there will be high clocks, and there will be more cores, but not at the same time. They still need to fit roughly the same TDP, so "just" 50% improvement in power means that the clocks at 16 cores active will be similar as Zen1 clocks at 8 cores active. The higher clock speeds will only be there for you when you are using less cores.
I don't think there's any chance of that. Based on the AMD figures about the TSMC process, there will be high clocks, and there will be more cores, but not at the same time.
I don't think there's any chance of that. Based on the AMD figures about the TSMC process, there will be high clocks, and there will be more cores, but not at the same time. They still need to fit roughly the same TDP, so "just" 50% improvement in power means that the clocks at 16 cores active will be similar as Zen1 clocks at 8 cores active. The higher clock speeds will only be there for you when you are using less cores.
I think there is no chance AMD will be introducing a client socket different from AM4 before DDR5.
I think the max core count for mainstream desktop is going to be 12 with a chiplet design. The 12 core parts will include 2 chiplets with 6 cores enabled in each (the 16 core parts to follow if Intel gets really competitive or at the latest as part of the Ryzen 4000 series (with Zen 3 cores and 7nm+). Only the 12 core parts will feature 2 chiplets, everything from 4-8 cores would be a IO die plus one chiplet. The max boost/OC in the 4,6-4,8GHz range.
For APU's I predict one chiplet+ IO die+ a GPU die (either 14 or 7nm) or one CPU chiplet+ a 14nm IO die which has a GPU integrated.
Having 4 cores disabled in your top of the line mainstream product seems a waste of expensive die space; so highly unlikely.
16 cores OTOH would blow the power budget when there are already higher clocks and wider cores.
8 cores doesnt seem enough to battle the 10 core i9 which has similar IPC but higher clocks.
So 12 cores seems the only logical possibility to me. It might be 3*4CCX, 2*6CCX but the chiplet design doesn't make much sense on the client side.
People always tend to extrapolate towards a trend but I'm pretty much convinced EPYC / TR is diverging from Zen at this point.
He said to take with a pinch of salt. There are still a few very good ideas there and if you add PS5/XB2 to that speculation then it looks very good from a manufacturing point of perspective. Next gen consoles are high volume parts and if they both would use full 8C chiplets (with lower clocks) there would be many 6C chiplets (with up to 2 faulty cores) left to sell as lower end Ryzen 3000 6C or and higher end 12C.
I would see that exactly the other way around. The 6-core parts with 2 defective ones would go into consoles. I'm still of the opinion that a 4c8t Zen2@3ghz is more than enough for consoles. But I see your point. A top 12c ryzen could use 2 chiplets that have 1 or 2 defective cores but are high leakge eg. clock high so not best option for server or consoles. But if that much binning is needed, 7nm as tsmc must yield very poorly.
EDIT: In fact the next consoles most likely will be 14nm still. I could see the gpu being 7nm but even that would be an issue price wise.
The similarities (and differences) between RGT, ATV, and Reddit leaks, not to mention the timing, has me thinking that AMD might be in the middle of a mole hunt where they're handing out 90% accurate info to various partners, with some minor but unique differences here and there so that when something leaks they have an idea of the source. That might explain some of the oddities in the reddit leak. On the other hand, the Nvidia RTX leaks also looked like they had various red flags but turned out to be mostly accurate. Or maybe the situation is just very fluid and the lineup is in a state of flux. Grain of salt in any case.
16c mainstream also serves to kill future sales revenue. There comes a point where you've got more than you need for the next 5-10 years. Going full hog might prove to be disastrous longer term, not necessarily for just AMD but for the entire industry.
AMD needs to start going towards wider cores IMO.
16c mainstream also serves to kill future sales revenue. There comes a point where you've got more than you need for the next 5-10 years. Going full hog might prove to be disastrous longer term, not necessarily for just AMD but for the entire industry.
AMD needs to start going towards wider cores IMO.
Zen1 is already a wider core than Skylake (check Agner Fog;s optimization guide for his comments on this). Zen2 expands on that and I suspect Zen4 will be even wider core than Zen2/3.
AMD going 16 cores on "quasi mainstream" is not bad since these SKUs will be positioned way above any current SKUs on mainstream AM4 platform. You would still be forced to buy a new "speical" AM4 board with new chipset and new power delivery system that supports 16C for that socket (as rumors go around now) so it will not be "mainstream". It will be high end AM4 segment but still positioned below the TR3 since you will lack 4 CH memory controller and upgrade path to 32(64?) cores as TR3 will allow you to do.
This is a smart move to basically kill off the whole Skylake-X segment (except maybe the highest end 18C part) with one or two SKUs priced between 450 and 600USD.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.