DegustatoR
Legend
Should be doable going off what we know about RDNA2 right now. 3080 and 3090 though seem to be out of their reach again. I kinda hope that they'll beat 3080 though.
Nvidia Ampere Discussion [2020-05-14]
- Thread starter Man from Atlantis
- Start date
-
- Tags
- nvidia
Love_In_Rio
Veteran
Yes they can do. Look at XSX APU wattage.
8k performance of the 3090Well this is their slide
Since we have no breakdown or numbers yet, I think it's safe to assume that "multiple popular graphics intensive games" includes a mix of RT and non-RT titles, which means as an average it's probably winning in RT and losing in pure FP32.
The 3080 has a fairly significant bandwidth difference whereas the 3070 might be rather starved for bandwidth compared to the 2080ti.
As always, just trying to temper expectations vs marketing
Example, their 3070 page has this graph and BL3 compared to a 2080ti which can push 90fps easily at 1440p. But for $500? Yeah, there's going to be some cheap 2080ti's on the market soon...
View attachment 4545
17.09. At least that is the time nVidia will release the whitepaper...
Rootax
Veteran
They need more than 50% in some cases.
Although I thought so too, it remains to be seen in non-RT games, as Malo has pointed out. Apparently, it seems about equal based on the three non-RT games shown.
Indeed. Yet AMD (and maybe Nvidia? don't remember) has perpetuated this nonsense in their own official slides, I believe, so it's hard to blame people for using the term that way.
Damn was hoping earlier than card availability.
Love_In_Rio
Veteran
Yes, no dedicated units. Good marketing.
The most impressive SSD tech so far. 24GB/s with that new data path? Hot damn.
I'm blown away by the tech and graphics shown. Hope the prices are what we are going to pay finally also. Somehow i doubt that.
Without jet-noise at that
Indeed, the biggest generational leap ever. Never expected that to happen imo.
I would think they have a 3080Ti down the line somewhere, with added VRAM.
40TF what, rasterization performance?
Yeah, AMD seems to be competing with Turing. Not good, we need even lower prices i think.
I think you're going to be dissapointed.
The most impressive SSD tech so far. 24GB/s with that new data path? Hot damn.
I'm blown away by the tech and graphics shown. Hope the prices are what we are going to pay finally also. Somehow i doubt that.
Without jet-noise at that
Indeed, the biggest generational leap ever. Never expected that to happen imo.
I would think they have a 3080Ti down the line somewhere, with added VRAM.
40TF what, rasterization performance?
Yeah, AMD seems to be competing with Turing. Not good, we need even lower prices i think.
I think you're going to be dissapointed.
This is not 24 gb/s but 14 gb/s. This is just for reach 14 GB/s compressed you need 24 CPUs core with rtx I/O and some GPU power you need only half a CPU core.
I am blown away by the 3080 but finally the performance per flops is lower than Turing. I did not understand at the beginning how a 30 Tflops GPU can only be two times more powerful than the 2080 but at 699 this is a steal.
Last edited:
Lower prices? $699 is a bargin for nVidia. I think it is even disruptive for the whole market. I think the XBSX will be $499. A 3080 is so much faster.
And the best part is: PC gaming has now access to full Raytracing with ~60FPS in 1440p. Better deal than the 3070 with 8GB btw...
I think they are looking at new consoles.
A new graphic card or a full gaming system????
I recall AMD doing it with RDNA, may have started before that though. It seemed to become a normal trend for the industry when Intel started highlighting it with every new CPU architecture and it carried over to GPUs.
Pretty sure AMD stated awhile ago that RDNA2 was already at 50% perf/watt increase compared to RDNA1. Which is what they said they got going from GCN to RDNA1.
So 5700xt performance at ~130w, double that to reach ~1.4-1.5x 2080Ti at < 300w and realize that they may not be doubling everything in Navi10.
Sorry for the AMD OT.
Big Navi will probably be competitive in performance but they will need to do a big effort for the price of the GPU because 3080 price is disruptive.
Marketing slide for the moment.
Marketing slide for the moment.
Rootax
Veteran
This is not 24 gb/s but 14 gb/s. This is just for reach 14 GB/s compressed you need 24 CPUs core with rtx I/O and some GPU power you need only haf a CPU core.
I am blown away by the 3080 but finally the performance per flops is lower than Turing. I did not understand at the beginning how a 30 Tflops GPU can only be two times more powerful than the 2080 but at 699 this is a steal.
Well not everything is about tflops.
Trying to keep the discussion about RTX IO in one single thread, over here: https://forum.beyond3d.com/threads/blazing-fast-nvmes-and-direct-storage-api-for-pcs-spawn.61761
General texturing performance and ROP performance don't look to have changed much, and the bandwidth available per FLOP is definitely lower than before, even with GDDR6x. FP32 FLOPs are only one aspect of performance, and it's going to be tougher to keep them fed and occupied.
It'll be fun to benchmark the 3080 vs 2080Ti, especially with the 3080's VRAM clocked down to match the bandwidth of the 2080Ti.
3080 will have double the FLOPs and Tensor cores vs 2080Ti, but the exact same number of TMUs, ROPs, and RT units.