NVIDIA GeForce RTX 5070 SUPER Possible Specs Emerge: 24 Gbit Memory Takes Centerstage

[Tigerfox]

The VRAM will eat LESS compared to non-Super variants.

What? How? Can you explain to me, how the same number of memory-ICs at the same frequency with 1.5x the densitiy is supposed to use less power? Are they made on a smaller process or do they use less voltage? If not, they will at least eat the same amount of power.

Hence my prediction of +10% rather than +5% you woulda expected from just +2 SM. 5070 has probably the craziest OC potential in the whole line-up so I won't be surprised to see 5070S running 3+ GHz outta box.

I didn't know about the high OC-potential, because I would not consider a 5070, but ~300MHz is a lot. But that's due to an effect we have seen at least since Maxwell, if not since Turing: All Cores reach about the same OC-speed, only very TDP-limited SKUs (like the original 4070) or 4090/5090 are limited. So a 5070S will reach about the same speed as a 5070, but 5070Ti and 5080(S) will, too.

So, 5070S getting higher TDP doesn't mean it will reach higher clocks out of the box, if you don't overclock. If you overclock, you would have done the same with a regular 5070, so the difference is still only +2SM and 50% more VRAM. If the official clock Speed stays the same, I don't see more than 4% improvement, more only in 4K with high RT or PT due to increased memory.

You do realise 9070 and XT are like 10 percent apart? It's too close to call. I consider them effectively the same GPU.

The tests I read it was more like 13-14%. Since 5070 was 5-7% slower thant 9070 in those tests, 5070S will be ~ on par with 9070 and slightly above in RT (more with PT), but still far from 9070XT.

We life in a time were 10%+ difference in performance is by far enough to warrant an additional SKU. All RTX 4000 down to 4070 were apart by just ~15% and 5070Ti and 5080 are, too.

We're far more likely to see 4+ GB GDDR7 modules than 1.5 GB ones.

I didn't get the beginning of this discussion, but who thinks about 1.5GB-modules? They should have been a thing with GDDR6 but AFAIK never made it to the market. We have been at 2GB-modules ever since RTX 3070 and RXT 3090Ti, and still are. 3GB already made their debut on RTX 5090 mobile, which is a 5080 with only 82SM, but 24GB VRAM, so it was only a question of time and naming when we would see this ony desktop.

I had doubt there would be 5070Ti with 24GB, since they could have made the 5080 with 24GB more exclusive. And I would have wished it would have gotten some additional CUs. But I still think the 5070Ti(S) is the most attractive card in NVs current lineup, becaus it offers 16/24GB@256Bit at a lower price than 5080(S).

 

[Macro Device]

do they use less voltage?

They actually do. It's a more refined process. Same node but architectural improvements made it more efficient.

We life in a time were 10%+ difference in performance is by far enough to warrant an additional SKU.

I know but when you can't tell them in a blind test what can I do.