Discussion Nvidia Blackwell in Q4-2024 ?

[jpiniero]

NVIDIA GeForce RTX 50 announcement may slip to 2025 - VideoCardz.com

NVIDIA RTX 50 in 2025 Kopite7kimi shares disappointing prediction. Shortly after Computex, journalists who attended the popular technology expo shared their insights from conversations with multiple board partners. However, unless they spoke with executives, this information is rarely accurate...

videocardz.com

Looks like it really is going to be CES now.

 

[SolidQ]

From RTG. It's on your opinion, believe or not in this "leak"

 

[jpiniero]

I'd be impressed if they manage to get the 5070 to be 26% faster than the 4070 Super despite the core count cut.

 

[MrTeal]

That formatting is brain melting.

The 5080 being 29% over the 4080S seems like a big jump with basically equal SM numbers.

 

[gdansk]

The gap between xx90 and xx80 grows larger?
I dislike this trend.

 

[TESKATLIPOKA]

That formatting is brain melting.

The 5080 being 29% over the 4080S seems like a big jump with basically equal SM numbers.

That would be equivalent to RTX 4090.

 

[amenx]

If accurate, dont like this. For the 5080 card to only match the 4090. 3080 and 4080 were quite a bit faster than previous flagships. As well as historically with xx80 cards for the most part.

 

[AdamK47]

Titan AI?

What dat?

 

[jpiniero]

If accurate, dont like this. For the 5080 card to only match the 4090

It's most likely less than the 4090. 4090 perf is plausible but that's the upper range.

Titan AI?

What dat?

5090 looks like it'd be the second cut of GB202. Titan AI would presumably be the first cut although still with the full 512-bit memory bus.

 

[SolidQ]

Titan AI?

What dat

There was "leak" like 56gb for AI
NV now is AI company

 

[blckgrffn]

There was "leak" like 56gb for AI
NV now is AI company

Historically Titans were large memory cards. So that's keeping inline with that tradition, at least.

 

[SolidQ]

Historically Titans were large memory cards. So that's keeping inline with that tradition, at least.

“The big thing exists"
From kopite

 

[Saylick]

That would be equivalent to RTX 4090.

With only 16 GB of VRAM, I bet, since GB203 tops out at 256-bit memory bus. I recall there being discussion of 3 GB/memory chip for GDDR7 but I don't believe that would be ready in time for this launch...

 

[KompuKare]

Titan AI?

What dat?

RTX 5090 + $2,000?

With big spenders, once you've hit near $2,000 you might as well stretch to $4,000, right? Plus, no VRAM anxiety according to rumours.

Meanwhile for the rest of us, a few crumbs maybe?

 

[Grooveriding]

Nvidia is going for it again, they always find a new way. Titan AI branding will be $5K for a decent, but not massive performance uplift over the 5090. At least that may allow the 5090 to go no higher than the 4090 is, who am I kidding? $2K USD for 5090? I just care about price right now. I can sell my 4090 right now for a little more than I bought it new for up here.

 

[APU_Fusion]

Does nvidia really care about desktop anymore?

 

[jpiniero]

Does nvidia really care about desktop anymore?

They still make a ton of money off of Client in general,. so yeah.

 

[MoogleW]

Does nvidia really care about desktop anymore?

I doubt they don't care about the aproximately $12 Billion a year. Just looking at Siggraph papers shows the latest in their funded research still has improvements in graphics. Like Area reSTIR for RT

 

[gdansk]

It's still good money and it can be even more as Nvidia increases prices.

 

[Heartbreaker]

I doubt they don't care about the aproximately $12 Billion a year. Just looking at Siggraph papers shows the latest in their funded research still has improvements in graphics. Like Area reSTIR for RT

Even NVidia knows that someday their AI profits will decline from some combination of increased competition, decreased AI companies, market saturation, etc... So it would be nice to still have a 10Billion+ gaming GPU business.

 

[MoogleW]

Even NVidia knows that someday their AI profits will decline from some combination of increased competition, decreased AI companies, market saturation, etc... So it would be nice to still have a 10Billion+ gaming GPU business.

Besides, omniverse is their current next big project that they are pushing and that will definitely require people to have GeForce GPUs to be successful.

 

[beginner99]

Even NVidia knows that someday their AI profits will decline from some combination of increased competition, decreased AI companies, market saturation, etc... So it would be nice to still have a 10Billion+ gaming GPU business.

One would hope so as the hype is certainly slowing down as people played around with the stuff but are left wondering how it is actually helpful? I mean yeah just fro the sake of it I now add AI generated title images to my slides for some wow effect but it doesn't make my work and better or more efficient.

 

[jpiniero]

英伟达 RTX 40 系列显卡产量被曝削减 50%，黄仁勋：本周发送 Blackwell 样品 - IT之家

消息源经济日报昨日还曝料称英伟达为了准备 Blackwell 产品，已经大幅削减了消费级显卡的供应，最高降幅为 50%，导致显卡供应出现短缺和新一轮涨价。

m.ithome.com

Speculation that NV is cutting supply of Ada to begin the transition to Blackwell. Interestingly they suggest 4070 and above is getting the cuts so perhaps the initial desktop launch might be wider than I had thought.

 

[Mopetar]

The gap between xx90 and xx80 grows larger?
I dislike this trend.

As long as the pricing is in some proportion to the silicon you get, does it really matter? I actually like the idea of a truly halo card that's well above the rest of the pack even if it has a premium price.

I thought what NVidia did with Ampere where the 3080 had a reasonable price was preferable to ADL and a 4080 nearly double the MSRP at $1,200. I'm not going to count the pandemic and mining boom against Nvidia and what it did to actual prices with Ampere, but they made an affordable xx80 card that gave great performance per dollar.

 

[gdansk]

As long as the pricing is in some proportion to the silicon you get, does it really matter? I actually like the idea of a truly halo card that's well above the rest of the pack even if it has a premium price.

I thought what NVidia did with Ampere where the 3080 had a reasonable price was preferable to ADL and a 4080 nearly double the MSRP at $1,200. I'm not going to count the pandemic and mining boom against Nvidia and what it did to actual prices with Ampere, but they made an affordable xx80 card that gave great performance per dollar.

Hmm, I think it means the 5080 will continue to get worse mm² per $ than the ultra expensive 5090.

You're looking at it the other way? But really I'd rather have a good affordable 5080 than a 5090.