Read more.Quote:
The flagship Navi 31 GPU is expected to offer 160 CUs and over 10,000 SPs.
Printable View
Read more.Quote:
The flagship Navi 31 GPU is expected to offer 160 CUs and over 10,000 SPs.
Interesting if true - seems a bit optimistic but who knows ;) That's if anyone will be able to buy them.
Seems a good place to leave a link to GN's recent video too on generational performance increase/$
https://www.youtube.com/watch?v=e_m2i5hU9jM
Imagine if they went all out and put HBM on the same substrate as the chiplets and IO die...
As they pull a Zen 3 and start charging 2-3 times the current prices of GPUs (because it's a lead product!).
AMD did rocket leap back into the CPU industry, they can do and am doing the same with the GPU Industry, in the long run it is good for us, well maybe, these advanced processes though for no matter what brand creates massive bottlenecks alone being able to get good yields.
What I am looking for, is when they invent something with a good yield on and maybe actually get the technology to create these things more on pair with what we want to make, so a huge pile of topend cards for all, from whichever team out there.
I can easily see this being plausible, using chiplets.
AMD however, have joined Nvidia in the pricing of GPUs. I expect these cards to continue the upward trend in cost, before taking into account crypto issues.
Ready availability of 5800X CPUs and a complete lack of 4 core die harvesting parts points to yield not being a problem on 7nm. The switch to EUV on 7+ and N5 nodes should in theory be able to make things simpler and hence more reliable.Quote:
Originally Posted by QuorTek
I saw a bunch of AMD GPUs go up on Overclockers and nearly bought one. Kind of glad I didn't as I ended up getting a 3060 Ti FE. I can't believe Nvidia was the cheaper option!Quote:
Originally Posted by Yoyoyo69
Using double the chipset and 5nm node its fairly possible that we will see a 2.5x performance increase, I doubt we'd see a 3x performance increase, its not realistic. Yeah 5nm is great, but they are also targeting power efficiency, I doubt they'll go all in for performance and leave energy efficiency behind.
Unless they are actually available, this is just another paper product on top of the current paper product.
If they're that quick, these are going to be swallowed up entirely by the mining industry. They'll be even harder to get than cards are now.
Only if they have really good memory bandwidth. The current gen with narrow memory and a big cache works for games, but really impacts mining :)Quote:
Originally Posted by Wozza63
Doesn't power efficiency go out the window for the flagship card, which they need to show max performance?Quote:
Originally Posted by SlickR
No, they won't be. The architecture may have that capability but there's no way they'd dump something that fast into the market. We'll get incremental increases - sizeable increments perhaps, but increments all the same.
We pretty much know at this point that if Nvidia can pull that sort of improvement they will. They will just charge through the roof for it knowing that enough people will, unbelievably, stump up the cash. AMD may as well add their product to the mix, or they sort of stop being relevant and Nvidia keep charging whatever they feel like.Quote:
Originally Posted by Spreadie
I don't really care, directly, what they do with or charge for flagship cards because even though I'm configuring a spec of a new machine right now, I'm not going anywhere near current flagship card price levels, let alone next-gen.
Where this interests me is indirectly. If these cars ae as fast as suggested, and are actually able to be bought, it is going kick the proverbial out of the desirability of current flagship products, which might make them accessible to mere mortals. I'm playing the long game. Sticking at the bleeding edge is seriously expensive, as my bank balanve over several decades showed. Now, I target the sweet spot of setting a medium budget, and getting the most bang-per-buck I can for it. Cards at 3x current flagship performance can only help. And tech history is littered with examples of where "wow, that fast?" becomes so ho-hum average, even a few years later. In fact, I struggle to think of a tech where it hasn't gone that way.
Double counting like that means you're also expecting a 600W TDP ;)Quote:
In the last story on HEXUS about the expected performance uplifts delivered by Zen 4 and RDNA 3, we quoted AMD's EVP of the Computing and Graphics Business Group, Rick Bergman, confirming that AMD was targeting a 50 per cent plus GPU performance per watt improvement from the previous generation. Add this to the doubling of CUs in the new flagship RNDA 3 GPU design, and we are staring down the barrel of a 3x performance increase – flagship vs flagship.