NVIDIA "Ada Lovelace" Streaming Multiprocessor Counts Surface

[zlobby]

I was joking. Satori and Vitalik are the respective inventors of Bitcoin and Ethereum

Yeah, but they are also regarded as hackers. But I didn't catch your drift towards crypto.

 

[Punkenjoy]

I hope lower end parts have huge frequency gains since the SM count isn't that much higher. They probably just want to stay competitive on the high end and milk the bottom end. (and they aren't alone).

 

[DanielF823]

I will still be incredibly upset if you need a 1200-1500Watt PSU for these

As someone who just dropped good money on an 800Watt SFF PSU

 

[Chrispy_]

I will still be incredibly upset if you need a 1200-1500Watt PSU for these

As someone who just dropped good money on an 800Watt SFF PSU

You'll be fine. There's only so much heat you can remove from an SFF case in the first place. You won't need a 1200W PSU because any GPU requiring it won't be coolable in an SFF PC.

It depends on your ambient temps and how much noise you are willing to tolerate but something in the sub-15L size is going to really struggle to evacuate more than about 750W of waste heat without sounding like a hair dryer. Anything more than about 400W is going to be clearly audible under load.

If you're using an SFX PSU in a full-sized case, then I can't help you there, but presumably you're smarter than that!

 

[Keullo-e]

I will still be incredibly upset if you need a 1200-1500Watt PSU for these

As someone who just dropped good money on an 800Watt SFF PSU

I doubt, those were practically only for multi-GPU users with a heavily overclocked HEDT platform.

 

[AusWolf]

I wouldn't be so sure. Ampere was built on both 8nm Samsung and 7nm TSMC depending on the market segment. With Ada we might get the entire line on some Samsung process and Hopper getting TSMC 5nm due to higher profit margins in the professional/server markets. Unless NV officially confirmed 5nm for Ada, but I'm not aware of such announcement.

Some process change has to happen. I'm dreading to see what higher core counts on the same node would do to TDP.

 

[Valent117]

big numbers, big gpu, big price...

 

[ModEl4]

If you look the previous TSMC designs (GM200, TU116 for example) and their die sizes and the new SM counts, if 144SM 4090ti is around 600mm2 then we may have a 60SM design at around 284mm2 achieving 3090ti performance level at $599-499.The cut down version of the 60SM part should be a little bit slower (max -5% at 4K) than the 6nm navi 33 which rumoured to be 440mm2 on 6nm and have 6900XT performance at QHD at a $499-449 price range. The thing is that 6nm vs 7nm TSMC gives only -15% area scaling for only the logic part so 6nm 440mm2 can fit at best very similar number of transistors (or less) vs navi 21, so regarding matrix/tensor performance on Navi 33 (if there is any, but realistically should be present) should be uneventful, probably based on their MI200 effort adapted accordingly for the gaming sector, while Nvidia probably will double down based on the rumored 144SM die size.Also if there isn't going to be a Navi 34 and AMD has only 6nm Navi 22 & 23 versions for the below $400 sector then at the end of 2023 we may see AMD's AIBs market share to be at just 15-16% and the other 5-6% that will be lost probably split between Intel and Nvidia...

 

[watzupken]

They said the same about Samsung 8 nm vs TSMC 12 nm, yet I see nothing of it. All I can say at this point is, fingers crossed.

You really think so? You don’t have to look far yo figure out Samsung nodes are really lagging behind TSMC. Just look at this and the last gen Qualcomm Snapdragon 8 series chips and you get an idea how bad they are. In fact, the SD 888 improvement over the TSMC made SD 865 wasn’t that far off as it was only sounded great mostly on paper. But it’s a hot and throttling mess, just like Samsung’s own Exynos. You can almost say that despite the similar 7nm, Samsung may be a full node behind TSMC. So the 8nm which is essentially a refined 10nm, I actually don’t find it better than TSMC’s 12 nm (a refined 16nm). While there is a slight improvement in power efficiency going from Turing to Ampere, but I would attribute it more to the architecture improvements and going wider with Ampere. Overall, we see quite a lot worst power consumption and lower clock speed.

 

[R-T-B]

I was joking. Satori and Vitalik are the respective inventors of Bitcoin and Ethereum

I think you mean Satoshi. There is no one named Satori in Bitcoin.

It's also almost certainly not the persons real name.

 

[Vayra86]

I think you mean Satoshi. There is no one named Satori in Bitcoin.

It's also almost certainly not the persons real name.

Wow indeed. Thx for that correction

 

[Cutechri]

I'm interested in the VRAM capacity. Will NVIDIA finally offer sane VRAM amounts? Who knows.

 

[MarsM4N]

Glad they use now the full name Ada Lovelace (was an English mathematician and writer).

Was kinda confused back then why they picked the name "Lovelace", because that's what came to my mind:

Fantastic movie. Still need to watch the "original", though lol.

 

[arni-gx]

so, in the late 2022..... all RTX 4080 20gb series and RTX 4090 24gb series will have TDP 500-600w minimum ???

 

[DanielF823]

I doubt, those were practically only for multi-GPU users with a heavily overclocked HEDT platform.

That was my thought... Those huge PSUs were usually only for Quad-SLI rigs... It's hard to imagine a single GPU needing 1200-1500

 

[Cutechri]

Let's hope not. Just because it's a desktop it doesn't mean companies should be allowed to cram all the wattage they want in it. Efficiency is still important. Personally, I'm not buying a GPU over 250w.