How far back do you have to go for a 200W, 256 bit, 16GB, 86SM gap between 80 and 90?
It's a new segmentation. The 384-bit GPU that existed for over a decade is gone. They're leaving a very big GPU desert because they know they can get away with it.
How far back do you have to go for a 200W, 256 bit, 16GB, 86SM gap between 80 and 90?Guessing we see the same segmentation we always have from nvidia
They know they don't have competition.How far back do you have to go for a 200W, 256 bit, 16GB, 86SM gap between 80 and 90?
It's a new segmentation. The 384-bit GPU that existed for over a decade is gone. They're leaving a very big GPU desert because they know they can get away with it.
Yes and no. Even without competition in the past they generally had logical increments between GPUs. It's been 2? decades since the top end card was 2x the next card down like this. They know people will spend $2000+ on a GPU and want to pressure people into it by having no competition from within.They know they don't have competition.
I agree. The gap between the flagship card and the high-end card has been ~1.35x for the longest time. It seems like in this case, we're going to get something much higher, like 1.5x.Yes and no. Even without competition in the past they generally had logical increments between GPUs. It's been 2? decades since the top end card was 2x the next card down like this. They know people will spend $2000+ on a GPU and want to pressure people into it by having no competition from within.
This is FOMO marketing wankery from a company that has plenty of money. But hey, in 2026 they'll cut down a bunch of GB202 to fill that gap. But only after they milked the cattle.
Kepler in 2012/2013. GK104 was 1536 shaders, GK110 was 88% smore shaders though with only a 50% bigger bus.How far back do you have to go for a 200W, 256 bit, 16GB, 86SM gap between 80 and 90?
It's a new segmentation. The 384-bit GPU that existed for over a decade is gone. They're leaving a very big GPU desert because they know they can get away with it.
The 5080 is probably going to be at least $1300, if not $1500. There are likely very few consumers looking for a GPU above that price that wouldn't prefer a 512bit absolute unit of a 5090 over a perhaps more cost effective but slightly less performant 384bit card as long as the cost increase isn't comically out of proportion to the performance uplift. I don't see this product segmentation as Nvidia milking their 90 class customers, they are just giving them what they want. It is the price increase of the 80 class and under cards where Nvidia is milking people.Yes and no. Even without competition in the past they generally had logical increments between GPUs. It's been 2? decades since the top end card was 2x the next card down like this. They know people will spend $2000+ on a GPU and want to pressure people into it by having no competition from within.
This is FOMO marketing wankery from a company that has plenty of money. But hey, in 2026 they'll cut down a bunch of GB202 to fill that gap. But only after they milked the cattle.
Yeah, that's exactly the problem. They have priced a 256 bit card where the 384 bit card should be. In order to drive consumers toward the 5090.The 5080 is probably going to be at least $1300, if not $1500.
Yeah, that's exactly the problem. They have priced a 256 bit card where the 384 bit card should be. In order to drive consumers toward the 5090.
There are no logical increments. The entire stack is designed to make the 5090 look desirable and "the only reasonable upgrade".
You're complaining about the same thing but from the opposite direction. The strategy is to only significantly increase performance on the high-end. They'll increase 5090 performance by 60% but MSRP by only 25%. Wow, look at all that value. All their lower products end up worse. 5080 will be 20-30% more expensive than 4080S for 30-35% more performance. Pricing is all structured to push people up the stack.I get where you are coming from, but they don't need a 384bit 80-class card if a 256bit gddr7 card matches the 4090's bandwidth, and I don't care if they make the 5090 384 or 512 bit because I'm not buying a 5090, and anyone that is buying a 5090 would likely prefer the 512bit bus. All I want is for them to price the 5080 at $1200 or less, otherwise I'm not buying at release. I got my 3080 for $750 in December 2020. The 4080 two years later was LESS fps/$ than my 3080. If the 5080 ends up being 4090 performance for $1500, it will have equal fps/$ to my four year old 3080, and I will be skipping another generation. At $1200, the increase in value would still suck, but I would hold my nose and buy it. $1k would be a fair price in a rational market with competition, but that's not the world we live in.
On no timeline do I buy a $2k+ 5090 regardless of what Nvidia does to gimp the rest of the product stack's value, and I'm sure the vast majority of the market is like that. For the people that do buy a $2k+ 5090, most of them wouldn't care if the 5080 was being given away for free as long as the 5090 is a significant increase in performance.
I don't know where you're getting that idea. Who said anything like that?5080 for the same price as 4090 seems insane, no shift at all in price/perf seems too much, even for nvidia
The 5080 is probably going to be at least $1300, if not $1500. There are likely very few consumers looking for a GPU above that price that wouldn't prefer a 512bit absolute unit of a 5090 over a perhaps more cost effective but slightly less performant 384bit card as long as the cost increase isn't comically out of proportion to the performance uplift. I don't see this product segmentation as Nvidia milking their 90 class customers, they are just giving them what they want. It is the price increase of the 80 class and under cards where Nvidia is milking people.
I think they learned from the mistake of 4080, which is why the 5080 die is half the size of the 5090 die. They will be keeping this one to $999.
I also don't get the confidence in this idea. To me, 4080 super price is more likely. Its not like keeping that price won't maintain their current margins with less chance of trash reviews on day 1.5080 for the same price as 4090 seems insane, no shift at all in price/perf seems too much, even for nvidia
I also don't get the confidence in this idea. To me, 4080 super price is more likely. Its not like keeping that price won't maintain their current margins with less chance of trash reviews on day 1.
Its not like keeping that price won't maintain their current margins
One of the leaks claimed that nvidia is considering a price range of $1300-1500. As much as I would like the price to be $1k like the 4080 Super, I would be very surprised if it is less than the 4080 release price of $1200, and I think it is more likely they will try to increase prices again. $1300 would be about in line with inflation from when the 4080 released.
GB203 should be a fair amount bigger plus the extra cost of GDDR7 (however much that is)
GB203 should only be a small amount bigger (5080 only has ~5% more CUDA than 4080S)
They can't take up too much more space. AD102 is already 608.5mm², and GB202 is adding 50% to the SM count and 33% to the memory bus. If the units were much larger with Blackwell they wouldn't be able to do that and stay in the reticle.The new features should take up a fair amount of space plus I'm expecting them to juice the tensor cores and maybe the RT.
They can't take up too much more space. AD102 is already 608.5mm², and GB202 is adding 50% to the SM count and 33% to the memory bus. If the units were much larger with Blackwell they wouldn't be able to do that and stay in the reticle.