Nvidia Ampere Discussion [2020-05-14]

[troyan]

With mobile SoCs Samsung 8nm LPP seems as effcient as TSMC's standard N7 process. The Kirin 980 used the same Mail configuration with 2% higher clocks (720Mhz to 702MHz) like the Samsung Exynos 9820 (8nm LPP). Exynos 9820 has ~10% better perf/watt: https://www.anandtech.com/show/14072/the-samsung-galaxy-s10plus-review/10

The Kirin 990 has a wider GPU but 16,7% lower clock rates and is ~20% better: https://www.anandtech.com/show/15099/the-huawei-mate-30-pro-review-top-hardware-without-google/4

 

[DegustatoR]

I'm having a very hard time believing that Nvidia totally fucked their power power draw advantage over AMD while doing a node shrink. 5700xt drew almost as much power as a 2080FE, and Turing was 12nm while 5700xt was 7nm. I know people are saying Samsung 8nm isn't great, but it seems very weird to me that a 2080FE would be 210W max and suddenly the 3080 is some kind of thermal monstrosity even with a node shrink that should give some advantages.

3080 FE cooler is almost exactly the same as that of 2080 FE.
It's also supposedly based on a cut down 102 die now instead of 104 so there's certainly some reason for power increase in that.

 

[BRiT]

Thread Reply-Ban issued to user incapable of keeping AMD out of Nvidia threads.

 

[Scott_Arm]

3080 FE cooler is almost exactly the same as that of 2080 FE.
It's also supposedly based on a cut down 102 die now instead of 104 so there's certainly some reason for power increase in that.

I thought the 3080 and 3090 shared the same cooler design?

 

[ninelven]

I'm having a very hard time believing that Nvidia totally fucked their power power draw advantage over AMD while doing a node shrink.

Indeed. To be fair, I think a portion of it probably is Nvidia expecting AMD to catch up somewhat, since AMD finally managed to turn the corner last gen and closing the gap is much easier than extending it. However, I suspect the primary reason is of a different nature: decibels. If you want a certain amount of cooling and to keep it below some audio threshold, then a larger heatsink / fan makes sense. And I am sure aesthetics also plays a role.

Edited due to uncertainty regarding fan size.

 

[DegustatoR]

I thought the 3080 and 3090 shared the same cooler design?

Design is similar, sizes are different.

 

[PSman1700]

I hope all 3000 series will look like that, i somehow really like that design ( i think the turing desing sucks ) Hope the design is near silent also.

 

[Scott_Arm]

Design is similar, sizes are different.

Yah, different size but same design with fans on opposite sides of the card. I mean, people saying you don't just make an exotic cooler unless you have to, but there are cards that push near 300W that use standard 2 fan designs. Do we really expect this thing to be a 300W card? If it is, it should be like 70% faster than a 2080FE.

 

[hurleybird]

Do we really expect this thing to be a 300W card?

Rumour mill is saying 350+

 

[pharma]

Overall size on a 12 pin doesn't seem to be much more than an 8 pin.

 

[Scott_Arm]

Rumour mill is saying 350+

For the 3080? That's 135W+ more. Doesn't make any sense to me. That's a 63% power increase. So are we expecting the 3080 to be maybe 80% more performance than a 2080?

 

[trinibwoy]

For the 3080? That's 135W+ more. Doesn't make any sense to me. That's a 63% power increase. So are we expecting the 3080 to be maybe 80% more performance than a 2080?

For the 3090.

 

[pharma]

Must have missed this rumor ... Tensor Memory Compression.

Tensor Memory Compression: NVCache is interesting, but Tensor Memory Compression will be on Ampere, and will reportedly use Tensor Cores to both compress and decompress items that are stored in VRAM. This could see a 20-40% reduction in VRAM usage, or more VRAM usage with higher textures in next-gen games and Tensor Memory Compression decreasing that VRAM footprint by 20-40%.

https://www.tweaktown.com/news/7466...per-small-on-geforce-rtx-30-series/index.html

 

[BRiT]

Would be nifty if it was using Kraken.

 

[Kaotik]

Must have missed this rumor ... Tensor Memory Compression.

https://www.tweaktown.com/news/7466...per-small-on-geforce-rtx-30-series/index.html

It's Moore's Law is Dead nonsense by origin. Just think how much bandwidth inside the chip it would take to pass everything through tensors in addition to all the other traffic, and for what I've heard (don't really have clue but I think the guy who said it does) tensors aren't even really suited for compression/decompression

 

[Remij]

It's Moore's Law is Dead nonsense by origin. Just think how much bandwidth inside the chip it would take to pass everything through tensors in addition to all the other traffic, and for what I've heard (don't really have clue but I think the guy who said it does) tensors aren't even really suited for compression/decompression

It was funny watching his little livestream today, he walked back the rumor he started about NVCache... He said now the 30 series might not include it but it's a thing in the Quadro, Data center, and Titan cards...

The guy just makes up bullshit that he thinks people want to hear to get them talking and clicking. Him and that Corteks guy or whatever are full of it.

 

[Frenetic Pony]

Design is similar, sizes are different.

Is the top top one in the right shot supposed to be a 3090? Looks kinda like it, maybe they're different bins of the same board as they look the same size underneath the cooler. One with 20gb of fast ram, the other 10gb of slower ram and disabled cores/lower clocks, thus the smaller cooler?

 

[Konan65]

Is the top top one in the right shot supposed to be a 3090? Looks kinda like it, maybe they're different bins of the same board as they look the same size underneath the cooler. One with 20gb of fast ram, the other 10gb of slower ram and disabled cores/lower clocks, thus the smaller cooler?

Apparently the 20GB model will be for card partners with the founders edition just with 10GB on the 3080.

 

[Scott_Arm]

For the 3090.

This thread is confusing. I'm posting about the 3080 and why the rumours about Ampere being a power hog don't make a lot of sense to me. Throwing a 3090 number in there means nothing. The comment that's been made is the fan design is different because it HAS to be, but I don't see why it would. The 3080 seems unlikely to be 300W, but it has the cooler with one fan on top of the card and the other fan on the bottom. It's most likely in the power range that could easily be cooled with a standard dual fan cooler. The only way the 3080 is approaching 300W is if it has a massive performance increase over the 2080.

The 3090 is an entirely different class of card. It might require some insane cooling if their through process was, "What's the most performance we can fit onto a card if we max out the pcie power delivery plus two power connectors. But I don't think that would tell us much about the architecture. It would just be cooling based on their TDP.

 

[pharma]

Apparently the 20GB model will be for card partners with the founders edition just with 10GB on the 3080.

Is it confirmed that the 3080 will only have a 10GB board?

What we know about NVIDIA SKUs so far:
Since we do not know the confirmed naming schemes yet, I will refer to these boards according to their board numbers and the RTX 2000 series card they are intended to replace.

1. The crown jewel of NVIDIA's lineup is the PG132-10 board with 24GB of vRAM. It is going to be replacing the RTX 2080 Ti and is currently scheduled to launch in the second half of September.
2. We then have the PG132-20 and PG132-30 boards, both of which are replacing the RTX 2080 SUPER graphics card and will have 20GB and 10GB worth of vRAM respectively. The PG132-20 board is going to be launching in the first half of October while the PG132-30 board is going to be launching in mid-September. It is worth adding here that these three parts are likely the SKU10, 20 and 30 we have been hearing about and the SKU20 is going to be targetted dead center at AMD's Big Navi offering (and hence the staggered launch schedule). Since AMD's Big Navi will *probably* have 16GB worth of vRAM, it also explains why NVIDIA wants to go with 20GB.
3. The PG142-0 and PG142-10 are both going to be replacing the RTX 2070 SUPER and will feature 16GB and 8GB worth of vRAM respectively. While the PG142-10 has a known launch schedule in the second half of September, the PG142-0 board has no confirmed launch date yet.
4. Finally, we have the PG190-10 board which is going to be replacing the RTX 2060 SUPER graphics card and will have 8GB of vRAM as well. The launch schedule for this board has not been decided yet either."

https://wccftech.com/nvidia-rtx-300...or-799-rtx-3070-for-599-and-rtx-3060-for-399/