Discussion Nvidia Blackwell in Q4-2024 ?

[Mahboi]

There aren't any 3 Gb chips until at least the refresh.

Irrelevant. The bus makes no sense. Extra chips later down the line or immediately, it makes no sense.

You seem to be judging it as a gaming chip. It's not. It's a professional & prosumer card. If gamers want to pay out of the nose for it, that's great, but it's not what it's designed for.

Even as a prosumer card, it makes no sense. 512 bit bus isn't just "beefy bus boy", it's a literal oversize. It is a complete waste of an expense.

 

[jpiniero]

Irrelevant. The bus makes no sense.

It does when the memory capacity makes it look better as an AI card.

 

[gdansk]

Big bus is for more memory capacity for ML crap. Plus they can use the slowest bin of memory chips on the part that has the most of them.

 

[Ajay]

Irrelevant. The bus makes no sense. Extra chips later down the line or immediately, it makes no sense.

Even as a prosumer card, it makes no sense. 512 bit bus isn't just "beefy bus boy", it's a literal oversize. It is a complete waste of an expense.

Not prosumer. "Workstation" cards that SIs will put into servers for low(er) rent HPC/ML/AI systems. Then again, the 512b bus could be total BS.

 

[Mahboi]

It is. They might do it under desperation though, but I don't think there's any despair to fear from Radeon.
Technically, there is despair to feel from ARC.

 

[gdansk]

Desperation? If there's a 512b bus it's for increasing ASP by offering big VRAM. Not because they're scared of the competition (remember there is none for ML).

$2500 sounds nice to Jensen.

 

[jpiniero]

$2500 sounds nice to Jensen.

More than that. RTX 6000 Ada's MSRP is $6799.

 

[MrTeal]

Desperation? If there's a 512b bus it's for increasing ASP by offering big VRAM. Not because they're scared of the competition (remember there is none for ML).

$2500 sounds nice to Jensen.

LOL... For the 24GB cut down gaming version of it maybe. 48GB RTX6000 Ada is already a $10k (edit: CAD) touch, workstation won't be getting a full 512bit with 64GB for anywhere south of 5 figures.

 

[gdansk]

LOL... For the 24GB cut down gaming version of it maybe. 48GB RTX6000 Ada is already a $10k touch, workstation won't be getting a full 512bit with 64GB for anywhere south of 5 figures.

Yes, for whatever cutdown they sell as a 5090.

 

[Mahboi]

Desperation? If there's a 512b bus it's for increasing ASP by offering big VRAM. Not because they're scared of the competition (remember there is none for ML).

$2500 sounds nice to Jensen.

They can easily sell an AI monster like Blackwell with a 384 bit bus without needing to go below $2500. With any luck, they'll even push to $3000 with it.
Again, Nvidia is the stingiest corpo when it comes to VRAM. And they love to see the customers return to buy more because of it.

Jensen will sooner sell you a gimped card and tell you "soz, you have to buy more, teehee" no matter how much you pay. He's been doing it for 15 years, it really shouldn't come out as a surprise.
And if somehow "just making more 384 bit bus cards and shipping it all" doesn't work anymore, if for some reason there is an absolute imperative to get a 512 bit bus, sure, they can throw another one in the pond. I just think it'll happen only in case of absolute necessity. So pretty much desperation. And I don't see that necessity so I'm expecting NV to NV: "least VRAM is best for our clients".

 

[Mopetar]

That ain't commercial, and honestly with GDDR7 having beefier chips, I'm not sure it makes sense either.
Also I bet 160 SMs. Nvidia is legendarily paranoid, if there isn't an AMD to beat into the ground, they'll just shadow box the wind until they die. They always go hard.

It's not a consumer card, but they will use those dies for high-end professional cards that make GeForce prices look like a bargain.

The consumer card will probably be salvage dies where the bus is cut down. Maybe towards the end of the product lifecycle you'll get a full (or close to full) die, but NVidia won't make a $4,000 gamer card when they can sell it for twice that to CAD users.

 

[SteinFG]

So the recent news are that NV will launch 5090 and 5080 in Q4. Hope that 4080 and 4070 series will get a good discount then. 4070 Ti Super is a good card with decent amount of VRAM, just has to cost like 600-650 dollars to sell out. Unless NV makes DLSS 4 an exclusive feature of next gen cards

 

[Mopetar]

I wouldn't be surprised if they have some new exclusive features that the older generation doesn't get, but I would expect prices to drop as they try to clear out inventory. It really depends on how long they wait to release the lower parts of the product stack though.

NVidia has been fairly good at managing channel inventory, so I wouldn't expect to see massive price cuts. Especially if they increase prices on their new cards instead of using the existing price points from last generation. If they offer enough of a performance uplift, I think that's the route they'll go with.

 

[igor_kavinski]

Especially if they increase prices on their new cards instead of using the existing price points from last generation. If they offer enough of a performance uplift, I think that's the route they'll go with.

That would be an awful strategy for them and I hope they suffer badly for going with it. The general gaming audience wants at least 40% better performance from newer gen at the SAME price point. Otherwise, what's the point of upgrading? Better to skip a gen if you are gonna pay more anyway.

 

[Mopetar]

The people who will buy a 5090 or 5080 at launch are not typical consumers.

 

[jpiniero]

That would be an awful strategy for them and I hope they suffer badly for going with it. The general gaming audience wants at least 40% better performance from newer gen at the SAME price point. Otherwise, what's the point of upgrading? Better to skip a gen if you are gonna pay more anyway.

That's gonna be unlikely. Both 2 and 3 dies should be a fair amount bigger... we'll see if NV opts for "Shrinkflation" or not for the others. As to whether GB203 ends up faster than the 4090 might depend on the GDDR7 speed NV opts for.

 

[Aapje]

The people who will buy a 5090 or 5080 at launch are not typical consumers.

And yet the 4080 sold quite poorly and had to get a price reduction, so what makes you think that they can increase the price again?

 

[Mopetar]

And yet the 4080 sold quite poorly and had to get a price reduction, so what makes you think that they can increase the price again?

Lack of high end parts from AMD for one. The 4080 had a terrible MSRP as the 4090 (if you could get one) was much better value. RDNA3 wasn't as good as people might have hoped, but AMD priced more competitively and the 7900 XTX was a slightly better 4K card for $200 less and more VRAM. You had to really value RT to pick the 4080 there.

There was also the price increase over the previous generation. Although practically no one actually got a 3080 for $700, it made the jump to $1,200 seem obscene. If NVidia prices the 5080 at $1,300, people wont complain nearly as much.

The final reason is that NVidia will have some new DLSS bells and whistles that the previous generation will never get. We might ridicule some of the ridiculous charts they put out, but the average consumer knows very little about technology and don't understand the card isn't really 3x faster than the previous generation.

 

[Aapje]

There was also the price increase over the previous generation. Although practically no one actually got a 3080 for $700, it made the jump to $1,200 seem obscene. If NVidia prices the 5080 at $1,300, people wont complain nearly as much.

I think that people will compare it to the 4080 Super and will be quite upset.

The final reason is that NVidia will have some new DLSS bells and whistles that the previous generation will never get. We might ridicule some of the ridiculous charts they put out, but the average consumer knows very little about technology and don't understand the card isn't really 3x faster than the previous generation.

This is just an assumption. They can't just keep doing this.

 

[Ajay]

This is just an assumption. They can't just keep doing this.

They will certainly try. Competitive differentiation has been a real win for them for years. AMD Radeon still has this more open source approach - I don't think it's ever really helped them increase sales.

 

[Saylick]

They will certainly try. Competitive differentiation has been a real win for them for years. AMD Radeon still has this more open source approach - I don't think it's every really helped them increase sales.

*puts on tinfoil hat*

I wouldn’t put it past Nvidia to double the throughput of their Optical Flow Accelerator and allow for one more interpolated frame for Frame Generation, so instead of a doubling of frames you could get 3x. Couple that with tensor memory compression and a market ploy where Nvidia pressures game designers to be more liberal with their VRAM usage, you get planned obsolescence of prior RTX generations.

*takes off tinfoil hat*

 

[jpiniero]

I think that people will compare it to the 4080 Super and will be quite upset.

Well.. it is only 15% more SMs for GB203, but you figure there will be changes to the SMs. Plus GDDR7 is gonna be decently more expensive. I figure part of the price cut of the 4080 Super was to make room for the 5080 to replace the 4080's price bracket... but as to the price, we will see.

 

[Aapje]

Well.. it is only 15% more SMs for GB203, but you figure there will be changes to the SMs. Plus GDDR7 is gonna be decently more expensive. I figure part of the price cut of the 4080 Super was to make room for the 5080 to replace the 4080's price bracket... but as to the price, we will see.

You ignore that Nvidia is already using the more expensive to produce GDDR6X, which uses PAM4, where GDDR7 uses the more economical PAM3. And with GDDR6X, they are stuck with a single supplier, which makes it harder to bargain for better prices.

GDDR7 is the opportunity for Samsung and Hynix to defeat Micron as Nvidia's GDDR supplier, which would be a major win, since everything is about AI now, and Nvidia is the AI hardware company. So it makes perfect sense for them to offer great deals to Nvidia, if only because they can leverage that into getting higher prices from other customers.

So I'm absolutely not convinced that GDDR7 is significantly more expensive.

And with 15% more SMs, no extra memory, and a bit of a boost from GDDR7, I'm not seeing the justification for a $300 price increase over the 4080 Super.

 

[Timorous]

And with 15% more SMs, no extra memory, and a bit of a boost from GDDR7, I'm not seeing the justification for a $300 price increase over the 4080 Super.

Since when have NV needed a justification to bump prices up? They do it because they think they can get away with it. If the 5080 is anywhere close to 4090 performance then they can sell it as '4090 performance for less' a bit like the 3080Ti was basically 3090 performance with half the ram for a lower price.

 

[Aapje]

The 4090 captures the 'money no object'-buyers, as well a professional buyers who use it as a 'cheap' AI card, so they can't demand the same premium from lower tiers. The 4080 non-Super already proved this.

You are just arguing for them to make the same mistake with the 5080 that they made with the 4080.

a bit like the 3080Ti was

A useless comparison since the 3080 Ti was priced at double the price of the 3080 because the mining boom exploded the market prices. Those prices had nothing to do what the gamer market would bear.