Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram ThePornDude
Home Artists Posts Import Register

Expectations for AMD 2023 - Telegrams for Die Shrink (Patreon)

Published:
2021-05-06 19:40:02
Edited:
2021-05-06 19:44:42
Imported:
2021-12

Content

Recently MLID did an "Ask me anything" about Intel's upcoming products after the Xe DG2 leak....and now there is an RDNA 3 leak.  Let's talk openly about what we expect from AMD over the next few years. 


You have ~1.5 hours to quickly put down your comments, thoughts, and questions below!  Sorry for the short notice!  Be thoughtful, use good grammar, and be concise - Tom may just read your questions on air!


Video about RDNA 3 and Nvidia's response: https://youtu.be/E8JCSTPdwHs

Comments

Anonymous

Hi Tom, I think AMD should consider creating some extra functionality with chiplets to make APUs be more like mobile SoCs. Think adding something like NPUs or other more specialized things. You've mentioned that they could add the GPU Compute units in the IO chiplet on Zen 4. Well for Zen 5 (assuming it's 2023), I think more heterogeneous computer models siimlar to what Apple is doing with M1 will really help especially in the mobile space (laptops and other form factors). How likely is it that AMD will try to push further into this design direction? Much love from Brooklyn.

KarbinCry

Secret Milan. Come on. You know you want to. :P

Anonymous

Can we expect to see a revolutionary new architecture for Zen 5?

Anonymous

Hi tom, Even if AMD was to capture the performance crown by a fairly large lead, (a clear cut 15-20% faster than nvidia) do you think the mindset of amd gpus would change at all? Putting the pandemic and the get what you can get mentality aside, What would AMD need to do to start changing the average consumer perception of their graphic cards? As it still seems that nvidia has the marketing share and mindset of the average computer user.

Anonymous

I think that with the recent success of AMD in taking performance crown from intel and soon to be from nvidia as well (if rdna 3 is as good as expected) is going to allow AMD to compete better outside of just hardware. As in I think that AMD will soon start to implement features BEFORE Nvdia instead of just making their own version shortly after. Lastly I also hope to see AMD continue the path of performance per watt, allowing mobile performance to rise. As it has been disappointing to see the mediocre performance increase from ampere mobile. All I want is a gaming laptop that has the battery life of a decent class thin and light!

Anonymous

So RDNA3 isn’t 6 times faster than RDNA1??? Joking aside, I am really intrigued for what AMD can do moving forward. They have a massive set of dominos setup. With the jump Zen 4 and RDNA3 is looking to bring, that’s already really impressive, but that’s not what I am looking at. I am looking at the insane size of programming talent that is going into the Zen and RDNA architectures. You have AMD themselves, xBox/Microsoft, PlayStation/Sony, Samsung and their mobile phones that now have RDNA IP. Any weakness that AMD had with software theoretically will very quickly evaporate. If you make a game for consoles, then we you port it to PC, you are halfway done for optimizing for AMD

Anonymous

Hey Tom, cool video as usual. As long as there's a 69% perf increase, i'm happy :D But on another note, have you heard anything about future RDNA architectures or others considering 3d stacked SRAM? I know the patents from a few months back about active bridges and whatnot are just that: patents, but It makes sense to do it when SRAM scaling is stalling and less power guzzly bandwidth is needed. Besides, Xe has off-die Rambo cache, kinda reminiscent of the Pentium pro, so anything wouldn't really surprise me.

Cleansweep

I feel like the supply the situation with silicon and other tech product components might take the wind out of AMD's sales and sails if things don't improve before 2023. For people who got burned waiting for RDNA 2 or trying to get any GPU in 2021, the sooner AMD announces and releases a product with good stock, the better things will be, especially if DG2 is at/near MSRP and easily available. RDNA 3 taking the crown won't matter much if AMD can't get GPUs into people's hands, but if they have good availability and beat Nvidia in performance at a good price, they should be in a good spot to rake in money and mindshare.

Anonymous

Tom, 2 questions: 1. If RDNA 3 releases on schedule, will AMD extend production of RDNA 2 to both fill the low end but also because scarcity may make it feel relevant to consumers? 2. What is AMD's reputation like outside of the gaming space? Will RDNA3 be competitive in the professional sector?

Anonymous

Are there any rumors that EVGA will start making AMD products? It's not like NVIDIA or Intel are working out too well for them right now. (One can always dream right?) EVGA has huge mindshare that AMD could easily tap into if they can work a deal out.

Anonymous

Will AMD's chip architect skills, including packaging, continue to keep pace with Intel? Will ARM chips significantly blunt their x86 success?

Anonymous

RDNA3 might be showing significant performance uplift, but wouldn't the design include a target process node (such as 5nm). What happens if AMD doesn't have enough capacity to manufacture the GPUs on 5nm (similar to what's going on right now). Is it possible for them to launch or design a version of RDNA3 for 7nm/6nm instead? Wouldn't AMD's priority for 5nm be Ryzen > CDNA > RDNA? Aren't any performance leaks pointless if there isn't any 5nm capacity to actually build the cards?

Anonymous

A lot of people seem to think that top non-cutdown RDNA 3 will be two 80 CU "chiplets" together on one board. I am concerned at how much power this could use. Do you think that the RDNA 3 "chiplets" could have a smaller CU count to be more compact or have less power draw? Also, will these separate "chiplets" require more expensive cooling than a regular monolithic gpu die?

Anonymous

What is the expected performance increase in ray tracing for rdna 3? If its not going to be very impressive then when will AMD release a card that makes ray tracing actually viable, and will it likely have a separate die just for ray tracing tasks?

Anonymous

Do you think that the 2.5-3X leaks are really in reference to the 5700 XT and being misinterpreted somewhere in the leak pipeline? That would line up much closer with what you are saying in your video and AMD's public target of NAVI 3X. Second question if you care to speculate: If 3X the 6900 XT did turn out to be true, and the rest of the 7000 lineup had similar gains, how would you price that as AMD?

Kirbyrules

Could the new chiplet based architecture for GPU's be put on a CPU. If the CPU has its own cashe AND say some GDDR or HBM2 on the die, could the APU have an up to 80 compute units chiplet. I could see them mass producing 80CU GPU's and cut them down for APU's or add two 60/72/80 CU's for the new RDNA 3 or 4. When we get down to 3nm, everything will be so small you could have a ton of stuff stuffed on one CPU die.

Mia

Clearly CPU is still the main money-making division for AMD, but have they reached the point (with some big profits recently) where the GPU division can get properly funded? It seemed like RTG was stuck in a cycle for a while of being underfunded and then underperforming/having strange lineups/major problems, which in turn made them less profitable (at least on the desktop GPU side).

Anonymous

Considering AMD is aiming for chiplet based GPUs with RDNA 3 which approach do you believe is the better one? - Manufacturing a bigger die (example 80 CU) and having most of the SKU being I/O die and cutdown GPU with only the top SKU having 2 GPU chiplets -Manufacturing a smaller die (example 40 CU or even less) and going with I/O die + 2 or more GPU chiplets across most of the lineup

MelodicWarrior

Hi Tom and crew! Is there any hints or info that points to an increase (or at least better support) in workloads that would help a person who does video editing or content creation with RDNA 3?

Anonymous

Do you think AMD will update it’s video encoding engine? At this moment results are worse than Intel’s QuickSync.

The Immortal Cameraman

I expect AMD to improve upon Z3N and RDNA 2 by no less than +15% and +45% performance increases, respectively. That is all.

Anonymous

Hello Tom and Dan, At this point, we are somewhat sure that Navi 31 has two compute die on it. But how many CUs do you think will be on each die? Personally, I think 60-CU makes more sense than 80. The yield would be better since a 80-CU die is still pretty big. And 60 sounds like a more versatile number. You could have a lineup consist of 120, 108, 96, 80-CU card, etc. What do you think?

Anonymous

Hey Tom and Dan, I think in 2023 Ryzen will maintain dominance but be on the verge of being threatened by Intel’s attempt at a true comeback, and I think they will have the true definitive raster crown but RT may be up in the air (per game basis at least) , that’s what I think will be happening, what I hope is that AMD doesn’t become complacent but seeing the price increases in Ryzen this gen, I think the forecast isn’t looking great for that.

qhfreddy

When will AMD start moving away from the processor-centric paradigm? Tech of the likes of processing in/near memory is long overdue and it's clear AMD (and intel) is struggling with the data movement problem in their architectures. What's the noise as far as the more long term goals of AMD?

kjm015

Hey guys, How does AMD plan on addressing the increased competition for TSMC process nodes moving forward? I imagine that as more and more large companies fight for allotment, it will harder for them to be able to reserve capacity for Ryzen/Radeon silicon.