Nvidia Ampere Discussion [2020-05-14]

[Deleted member 13524]

I think it's because there's a 12GB card in reserve against RDNA 2 (rogame has PCI IDs for a 12GB card). I mean, *maybe* the 20GB 3080s will have extra two SMs or so enabled but 20GBs will be solve at a large premium and made for prosumers or the guillable who see dat extra VRAM.

Activating more execution units is what I thought nVidia would do for a 3080 Super announced in October 28th, but that was before I knew the performance delta between the 3080 and the 3090 was so small.
If they add more SMs, the performance difference might go from 10% to 5% or less, and with 20GB the RTX 3080 Super will start cannibalizing the 3090.

 

[trinibwoy]

I wonder what changed their mind on their taxonomy.

They're leaving room for a $3000 48GB Titan next year.

 

[techuse]

So it's official:



https://www.nvidia.com/en-gb/geforce/news/rtx-3090-out-september-24/

The leaked review was accurate.

It's also interesting that nVidia is placing the non-Titan Geforce RTX 3090 as a prosumer graphics card, as their marketing material seems to focus heavily on productivity applications.
I wonder what changed their mind on their taxonomy.

I’d wager them needing to use the 102 die to offer a relevant performance increase at the 104 tier factors in.

 

[DegustatoR]

I’d wager them needing to use the 102 die to offer a relevant performance increase at the 104 tier factors in.

GA102 size is actually closer to TU104 than TU102 though. What "tier" are you talking about?

 

[techuse]

GA102 size is actually closer to TU104 than TU102 though. What "tier" are you talking about?

The xx80 tier of GPUs. They have no path to offer anything noticeably faster since they are already on the largest die. When was the last time this happened? Fermi?

 

[DegustatoR]

The xx80 tier of GPUs. They have no path to offer anything noticeably faster since they are already on the largest die. When was the last time this happened? Fermi?

Kepler actually. GTX780.
But the reason why they can't offer anything faster has nothing to do with tiers and die sizes and everything to do with power consumption.

 

[Cyan]

Two YouTubers given preferred treatment over every other NDA signees.
Here's the other guy:

edit: Yes, I fully realize those are not reviews of the card, thanks.

the guy is totally right, framerate is king! Talking of which, when smoke is on screen and some transparency effects it suffers quite a bit, specially in the winter setting of FH4. It's also obvious that contrary to Linus, this guy didn't use nVidia's recommendations taking into account how Control runs.

So it's official:



https://www.nvidia.com/en-gb/geforce/news/rtx-3090-out-september-24/

The leaked review was accurate.

It's also interesting that nVidia is placing the non-Titan Geforce RTX 3090 as a prosumer graphics card, as their marketing material seems to focus heavily on productivity applications.
I wonder what changed their mind on their taxonomy.

a bit disappointing to be honest, but at least nVidia aren't deceiving anyone. Some people were hyped about the RTX 3090 for gaming and waiting for it but I expect those expectations to diminish.

 

[pcchen]

a bit disappointing to be honest, but at least nVidia aren't deceiving anyone. Some people were hyped about the RTX 3090 for gaming and waiting for it but I expect those expectations to diminish.

It shouldn't be surprising though, as 3090 has only about 20% more resources than 3080 in almost every measure (a little higher in memory bandwidth, ~23% more). Therefore, people shouldn't be expecting 3090 to be something like 30% faster than 3080 anyway.
Of course, overclocking-wise it could be very different, but that'd depend on how much more power limit can be set to the card.

 

[PSman1700]

It shouldn't be surprising though, as 3090 has only about 20% more resources than 3080 in almost every measure (a little higher in memory bandwidth, ~23% more). Therefore, people shouldn't be expecting 3090 to be something like 30% faster than 3080 anyway.
Of course, overclocking-wise it could be very different, but that'd depend on how much more power limit can be set to the card.

Indeed, only goes to show how good a value the 3080 is then.

 

[arandomguy]

The GA102 whitepaper seems to have been slightly updated.

I remember awhile back I had a question regarding the ambiguity of the features GA104 lacks compared to GA102 as the GA104/RTX 3070 section had the following line -

GA104 retains most of the key new features that were added to NVIDIA’s GA10x Ampere GPU Architecture and ships with GDDR6 memory.

it has been changed to -

Note that the GA104 GPU includes all GA10x architectural features except NVLink, and while GA104 supports GDDR6X, the RTX 3070 GPU uses GDDR6, not GDDR6X.

I don't know if there are any other changes yet.

 

[Voxilla]

The GA102 whitepaper seems to have been slightly updated.

I remember awhile back I had a question regarding the ambiguity of the features GA104 lacks compared to GA102 as the GA104/RTX 3070 section had the following line -



it has been changed to -




I don't know if there are any other changes yet.

They fixed the Titan RTX mixed precision spec too, from 65.2 to 130.5 TFLOP, Nvidia is reading this forum

 

[Voxilla]

I have no GT 1030 handy, but at least 1060 still had (a few) DP-units.

Fun fact: DP-throughput on Radeon HD 5870 still beats RTX 3080. 3090 will finally overtake it though.

The HD 7970 still is unchallenged in the consumer space at ~1 TFLOP double precision.
The only use I ever had for DP was a fractal zoomer I made:

DX11 DirectCompute Mandelbrot and Julia viewer

 

[CarstenS]

The HD 7970 still is unchallenged in the consumer space at ~1 TFLOP double precision.
The only use I ever had for DP was a fractal zoomer I made:

DX11 DirectCompute Mandelbrot and Julia viewer

Yep, I know (also your mandelbrot-thingie, but I like the bandwidth limited waves3D better). Was going back as far as I could though, saving "the best" for later. One could argue about the first Titan being a consumer card though.

 

[Deleted member 13524]

It shouldn't be surprising though, as 3090 has only about 20% more resources than 3080 in almost every measure (a little higher in memory bandwidth, ~23% more). Therefore, people shouldn't be expecting 3090 to be something like 30% faster than 3080 anyway.
Of course, overclocking-wise it could be very different, but that'd depend on how much more power limit can be set to the card.

Eeerm you quoted me but I didn't write what you quoted..

 

[Deleted member 13524]

Indeed, only goes to show how good a value the 3080 is then¹ ².

1 - If you can find one;
2 - At MSRP.

 

[Jawed]

Eeerm you quoted me but I didn't write what you quoted..

I think there is a forum bug. I've noticed recently that lots of replies are quoting someone who did not originate the text being quoted.

 

[Voxilla]

There are emerging numerous reports of crashing/freezing issues with the 3080, like
https://www.nvidia.com/en-us/geforc...x-3080-hitching-and-crashing-across-all-titl/
No official statement yet from Nvidia.

 

[BRiT]

Eeerm you quoted me but I didn't write what you quoted..

I fixed that misquote as it was quoting Cyan's post (that had quoted you). I have no idea how it ever got your post attributes. Most of the time its from someone multiquoting but missing an end quote tag, but that want the case here.

Very odd situation this time around.

 

[Wesker]

Nice reviews, 8k60 is reality in 2020. Now DLSS and ray tracing in that mix and see how far you can come? The details and fidelity that comes available in Eternal @8k is impressive, and that at 60fps/hdr.

Nope:

 

[CarstenS]

3080-launch reloaded? I thought Nvidia said, they'd taken measures...