It wouldn't have made sense to do a monolithic model if they were intending to release single chiplet models. The RDNA2 refresh will be just fine below Navi 33.
The chiplet-based models will be modular and can support a broader product line. I'm assuming it's entirely possible for there to be single chiplet GPUs that would be less powerful than a monolithic GPU targeting 6900XT performance levels.
I suspect it would still be beneficial to update the lower end stuff to TSMC N6 and RDNA 3 to reap any potential cost benefits and architectural improvements, but I agree that there is no real rush.It wouldn't have made sense to do a monolithic model if they were intending to release single chiplet models. The RDNA2 refresh will be just fine below Navi 33.
https://videocardz.com/newz/nvidia-...raphics-card-with-full-next-gen-ada-ad102-gpu
900W what the hell. That said, 48GB VRAM would be of interest to me.
https://videocardz.com/newz/nvidia-...raphics-card-with-full-next-gen-ada-ad102-gpu
900W what the hell. That said, 48GB VRAM would be of interest to me.
In a few weeks we will have a Flux Capacitor variant with 1.21 Gigawatts, comes with free DeLorean (Nuclear Power Plant sold separately).
So AD103 for RTX 4080 then, intersting, that means RTX 3080Ti/4090/Ti will be way faster this time compared to RTX 4080 not like RTX 3090 vs RTX 3080.
RTX 4080 84SM 16GB 256bit 64MB l2 cache
RTX 4070 60SM 12GB 192bit 48MB l2 cache
https://videocardz.com/newz/nvidia-...raphics-card-with-full-next-gen-ada-ad102-gpu
900W what the hell. That said, 48GB VRAM would be of interest to me.
Or Dual-Die?
Lol, "extremely highly clocked" might mean an extra 100 MHz only, at the cost of +300W.No. Just extremely highly clocked I guess. The regular 4090 is "only" 600 W so this would be a big jump.
Lol, "extremely highly clocked" might mean an extra 100 MHz only, at the cost of +300W.
That makes sense when you're going from 350W to 450W. Going from 600W to 900W is an entirely different matter. That's likely so far from the efficient zone that the F-V curve is practically flat (assuming voltage is on the horizontal axis and frequency is on the vertical axis). Heck, I'm more impressed that Nvidia can actually pump 900W of total board power into the thing. That's unheard of for a consumer part. It's even more than the TBP of their H100 server part.Well, the "official" boost clock for the 3090 Ti is 300 Mhz over the 3090 for an extra 100 W. And you'd have to account for having double the ram chips if it has 48 GB.
I look forward to laughing at the perf/W charts from various reviewers if this thing really does come with a 900W TBP. It's going to be within the same range as GPUs from last generation or possibly even two generations ago. I'd laugh even harder if the original Fermi GTX480 somehow gets better perf/W (not likely, but still something to chuckle about if it did happen).No matter what it takes, Nvidia must win by 1 fps!
My prediction/guess is that Nvidia will be very competitive at the top end and might beat AMD there, but is less competitive at the mid/low end.
I look forward to laughing at the perf/W charts from various reviewers if this thing really does come with a 900W TBP. It's going to be within the same range as GPUs from last generation or possibly even two generations ago. I'd laugh even harder if the original Fermi GTX480 somehow gets better perf/W (not likely, but still something to chuckle about if it did happen).
I hope they at least offer us a GT 4030 for under $800 with maybe performance similar to a GTX 1060.
Are we looking at 7800 XT having 20 GB? I'm sure the VRAM advantage factored into a lot of gamers' decisions in getting the 6800 XT over the 3080.
It wouldn't have made sense to do a monolithic model if they were intending to release single chiplet models. The RDNA2 refresh will be just fine below Navi 33.