Classic Nvidia... The Way the Consumer is Meant to Be Played.™
Classic Nvidia... The Way the Consumer is Meant to Be Played.™Oh boy. There's an entire 3 year+ thread on limitations of Nvidia's preferred VRAM quantities.
| Model | SM | Clock Speed | TDP | Memory | Memory BW |
|---|---|---|---|---|---|
| RTX 4070 S | 56 | 2475 MHz | 220 W | 12GB GDDR6X 21 Gbps | 504.2 GB/s |
| RTX 5070 | 50 ? | ? | 250 W | 12GB GDDR7 28 Gbps | 672 GB/s |
| N48 | 64 | ? | ? | 16GB GDDR6 20 Gbps | 640 GB/s |
| RTX 4080 S | 80 | 2550 MHz | 320 W | 16GB GDDR6X 23 Gbps | 736.3 GB/s |
| RTX 5080 | 84 | ? | 400 W | 16GB GDDR7 32 Gbps | 1 TB/s |
| RTX 4090 | 128 | 2520 MHz | 450 W | 24GB GDDR6X 21 Gbps | 1 TB/s |
| RTX 5090 | 170 | ? | 600 W | 32GB GDDR7 28 Gbps | ~ 1.8 TB/s |
It's DOA if it's priced at $700 or more. That's close enough into 4070 Ti Super territory, where there's more raster performance and 16GB of VRAM.And dead on arrival if it is only 12GB.... Current games already use more than that.
And dead on arrival if it is only 12GB.... Current games already use more than that.
Here I charitably read DOA to mean "Probably shouldn't be bought except by an extremely limited number of users who will accept an experience worse than consoles in many new games despite having paid for an state-of-the-art GPU die that could easily do better"Sure just like 4060 and 4060 Ti were DOA at 8GB... Both of which are now in Steam Survey top 5.
Too many people on this forum, substitute their wishful thinking in place of reality.
I'm very doubtful the 5090 will be 600W, there is simply no need.
There might be if it's hitting pretty high frequencies or there are just too many extra transistors to feed.I'm very doubtful the 5090 will be 600W, there is simply no need.
Here I charitably read DOA to mean...
Yes indeed! I'm not inclined to limit myself to the reality where we have $400 8GB GPUs in 2025. But here we have multi-year uncompensated cheerleaders for it because it sells billions. Who are we to call it DOA because it is DOA to most people paying attention.So substituting your alternate reality, for his alternate realty...
Really? The SM count, TDP, bus width, memory increase and memory bandwidth all suggest it will be a pretty huge increase at 4K. Like 50-60% faster than the 4090.5090 won’t change much, probably just a few games at 4K that will use the additional power for ray traced games.
Dang. Ewww. DangIt *should* be DOA. But smiling coprophagists will keep it alive.
Yeah, if they feel they need to target a full 70% they might need 600W. But since GDDR7 and 80% more memory bandwidth is probably helping a lot with performance already, I suspect the slight node shrink and tweaked architecture will do the rest and give us a comfy 50%+. NVidia simply doesn't need to deliver more now IMO. My opinion is of course not worth much though, but there is PR value in not requiring ridiculous power as well.It might be 60-70% faster than the 4090.
Considering how this lineup as rumoured I think Nvidia planning a Super refresh for the 5070 & 5080 with 3GB Modules (so 18 & 24GB VRAM respectively) and higher clock speeds against RDNA 5. Hell maybe a 5090 Super with more SMs enabled (say to 180-182) and clocks go up a bit with 48GB VRAM. Like the "400W" and "600W" is probably headroom for Super variants.
There isn't a sign of Quadro Blackwells in the next few months, suspect they are waiting for 3GB Modules sometime in 2025 to whack them on and have them use 96GB of VRAM on both sides of the PCB.18 and 24GB 5070s and 5080s respectively are possible mid generation in 2026. I suspect they will come with a price premium like the 16GB 4060Ti does. When you get that much VRAM, NV wants to get paid handsomely for the fact that you'll likely not upgrade in a long time. It might be a mistake to wait for them unless you're ready to pay a very solid premium.
I don't believe in a 48GB 5090, there will be dedicated AI cards for that market, probably made from the same chip, and those will be priced way above the 5090 price, which already got a 33% increase in VRAM. In the current lack of competition situation, NV will be very careful and not setting themselves up for less sales of the 6000 series because people are content with VRAM.
I don't see this happening. My expectation is that for a refresh they will only increase the memory for the deficient models, and will keep using 2 GB modules for some cards.18 and 24GB 5070s and 5080s respectively are possible mid generation in 2026.
Minor typo probably, but the 5070 Super would be 18GB (12GB * 1.5) with 3Gbit VRAM modules.I don't see this happening. My expectation is that for a refresh they will only increase the memory for the deficient models, and will keep using 2 GB modules for some cards.
So:
5080 Super 16 GB with 2 GB modules
5070 Super 16 GB with 3 GB
5060 (Ti) Super 12 GB with 3 GB
And then perhaps a 5090 Ti or a Titan with 3 GB modules, while keeping the 5090 the same.
Of course, all of this assumes that the 3 GB modules are available in volume and for a good enough price.
You can bet on it. No matter the price of the 5090.I worry about individuals and companies snatching up consumer level boards to be put into AI training farms.
With that and the scalpers it's going to be the 3090 all over again. AI replacing the cryptocurrency mining craze.
If it's anything like the 3090 release then I better get in line at Micro Center at midnight. There were none available at retail after the first day. The only way to get one was though a scalper.You can bet on it. No matter the price of the 5090.
My solution is to skip the rat race this time.If it's anything like the 3090 release then I better get in line at Micro Center at midnight. There were none available at retail after the first day. The only way to get one was though a scalper.