Dubious; discuss in a Ryzen thread. They'd need another at least 2 new dies with incredible low volume for either of them.
A hypothetical bigger GB203 would have very good volume.
Everyone who enables Nvidia; be it Intel or AMD or gamers.
Dubious; discuss in a Ryzen thread. They'd need another at least 2 new dies with incredible low volume for either of them.AMD could easily make a 24/32 core Ryzen but they choose not too.
Everyone who enables Nvidia; be it Intel or AMD or gamers.whose fault is that? Both consumers for buying such expensive cards and AMD/Intels for not competing and Nvidias as well for being a greedy pig.
That's the excuse. But it'll be more popular than the RTX 5080 on Steam Hardware Survey just like the RTX 4090 is more popular than the RTX 4080. Possibly even by a larger margin. They're chasing whales like Cpt Ahab.It should have been obvious that NV beefed up GB202 to chase AI markets... cards which will go for multiple times the 5090 goes for.
They're chasing whales like Ahab.
You really think GB203 at 1/2th size with parametric yields is costly to make? Come on. Stop simping for the world's most profitable company.Which is the way to go with node prices going the way they are.
uh no, most still would go for GB202.A hypothetical bigger GB203 would have very good volume.
yep, why do think AMD is moving to chiplets again for GPUs after RDNA4.You really think GB203 at 1/2th size with parametric yields is costly to make?
Agree.GB203 being half the size is just a joke.
You really think GB203 at 1/2th size with parametric yields is costly to make?
I doubt it. They're made money shipping even lower priced dies to Sony.Yes. If AMD doesn't get with the program, they will quit... with the way node prices are going.
Doubt it. The only reason AD102 did so well is because AD103 sucks. Nvidia could make a 48GB GB202.5 384 bit version and it would sell better (omg it's AI) than a 32GB GB202 and be cheaper to make. But they don't want to do that for some weird reason. Margin-maxing, paypig-porking, whale-hunting. Whatever you want to call it.uh no, most still would go for GB202.
Isn't AMD a software company now?They'll leave the consumer graphics card market because of software development costs, not node pricing.
So they say. But that's because Nvidia already is. And despite ordering billions of Hopper they still spent more on software... so good luck AMD!Isn't AMD a software company now?
It is actually worse. AMD can give you more cores at minimal cost. By reusing the same I/O die and using tiny chiplets, AMD saves on R and D and manufacturing which allows them to add cores for cheaper.It's not equivalent. Namely that's a CPU. It also is slower than its Ryzen equivalents for anything except professional work so it is undesirable and AMD will never be able to convince people that's a good deal even if they priced it lower.
GB202 will be faster at everything. It will be easy to convince pay pigs to buy it. Look how good of a deal it is compared to GB203. Amazing! But if there was a reasonably sized intermediate between GB202 and 203 many would choose that instead. Nvidia makes sure to not have an option which is a new and more anti-consumer choice.
There is no reason except profiteering and lack of competition for the 370mm² gap between GB203 and GB202.
Not when it requires a new socket (or a new CCD and IOd pair such that it can somehow fit in AM5) for a product that will never show up at nearly 1% of the Steam Hardware Survey even if priced at an (amazingly cheap for Nvidia) $2000. Because a $480 CPU would still be just as good and easier to use.AMD chiplet design allows them to scale cheap in comparison and save on R and D.
Zen5 is cheap to make, period. The dies are very small compared to GPUs and they reuse the IOD for desktop. So if anything AMD makes bank on Zen5 be it EYPC or Ryzen.Not when it requires a new socket (or a new CCD and IOd pair such that it can somehow fit in AM5) for a product that will never show up in the Steam Hardware Survey even if priced at an (amazingly cheap for Nvidia) $2000.
It is not cheap to make if the volume is low and it needs - like I said - two new chips to fit in AM5. Fixed cost don't scale. But remember: Nvidia's gaming revenue is 70% higher than *all* consumer Zens including TR, mobile, desktop. So they can justify these fixed costs much more easily. And it wouldn't even need a new chip to have a part halfway between GB202 and 203.Zen5 is cheap to make, period. The dies are very small compared to GPUs and they reuse the IOD for desktop. So if anything AMD makes bank on Zen5 be it EYPC or Ryzen.
Nearly all the 50 series is a wash. Only two cards worth buying are the 5090 and 5070 TiWell, the only good thing I have seen is that Nvidia finally realized that the 5070 is most probably DOA due to 12GB RAM, and has slipped in the 5070 TI at launch that will make use of a stripped down 5080 chip, but keep the 16GB memory from that design.
You seem to think 80 and 90 series cards are brought in high volume but outside the tech bubble it’s the 60 and 70 class cards that most people buy.And it wouldn't even need a new chip to have a part halfway between GB202 and 203.
Nearly 1% of the Steam install base is running 4090 or 4090D. That is a lot of machines. And it is more than 4080, which is my point. The gap widens.You seem to think 80 and 90 series cards are brought in high volume but outside the tech bubble it’s the 60 and 70 class cards that most people buy.
Interesting. With AI bull it will get worse too, a card with 32GB gddr7. oh boy its gonna hit 4K+ AUD. Not only is die cut but the vram too for the 5080. avoid it lolNearly 1% of the Steam install base is running 4090 or 4090D. That is a lot of machines. And it is more than 4080, which is my point. The gap widens.
Next card that I buy will definitely have more than 16GB of VRAM. Another option would be APU with 64GB of RAM that is faster than my current card...I do think a 5080 super needs to come out. Most probably it will be at CES 2026.
These low VRAMs amount vs price won’t cut it till Ruby. A super refresh is a must this gen.
Interesting. With AI bull it will get worse too, a card with 32GB gddr7. oh boy its gonna hit 4K+ AUD. Not only is die cut but the vram too for the 5080. avoid it lol
A smart person would go for the 70 ti.
Not from the leaks I saw. The 5080 is the GB203-400, the 5070 TI is a GB203-300, the 5090 is the GB203-600. So that would make both the 5080 and 5070 TI cut down ones with only the 5090 being a close to fully implemented card (I suspect there is still a little room on the table for a 5090 TI or 5090 Super, but those might also be a process node change, but I doubt it).The 5080 is the full GB203 and the 5070 Ti the cut one.
Not from the leaks I saw. The 5080 is the GB203-400, the 5070 TI is a GB203-300, the 5090 is the GB203-600. So that would make both the 5080 and 5070 TI cut down ones with only the 5090 being a close to fully implemented card (I suspect there is still a little room on the table for a 5090 TI or 5090 Super, but those might also be a process node change, but I doubt it).
Desktop 5080 and Mobile 5090 are supposed to be uncut GB203. Desktop 5090 is GB202.