GB203 is about the same size as AD103, so GB202 should be around 750mm².There's a bunch of fixed stuff that takes up a bit of space besides the shaders.
I am expecting GB202 to be close to the limit.
GB203 is about the same size as AD103
It's not from MLID.I seriously doubt that. I think MLID claimed something to that effect... but you think people would just ignore MLID by now.
The 1080 Ti was released in March of 2017 at $699. Accounting for inflation that would be $905 today. I hope we don't see a 5070 Ti releasing at $1K.The 5070 Ti is going to be the $999 model. Maybe slightly cheaper.
Nvidia’s Blackwell chip troubles, revealed in August, hurt relations with manufacturing partner TSMC, leading Nvidia to consider having Samsung make its latest gaming GPUs (non-AI related chips), The Information reports. Nvidia engineers testing initial Blackwell chips found they failed during some common data center usage, possibly from a design flaw.
Nvidia would never switch or even contemplate doing that as suggested. It would require rearchitecting moving to a different foundry. Even if on the same node.ZERO chance of that. Slow news cycle, so time for more nonsense rumors that go nowhere. Process libraries get increasingly more complex over time, making switching FABs more and more difficult. Next Gen GPUs are locked to TSMC at this point.
"We had a design flaw in Blackwell, it was functional, but the design flaw caused the yield to be low," Huang said. "It was 100% Nvidia's fault."
Looks like Jensen is keen on keeping good relations with TSMC. He accepted full blame for the issue in Blackwell and said that TSMC helped them fix it.
Sounds to me like it's a leader just being a good leader.This is the consequence of TSMC's dominance. Not great for the market.
Nvidia is not stupid.
The only time it makes sense to me that they would have switched, is when they decided to do 20X naming like 2nd gen maxwell, maybe this is 2nd gen blackwell design. However, not only would that have been a long time ago, but the process issues talked about here are unrelated to the GeForce department. We also had no leaked indications of this in a timeframe that makes sense.
This is not he first or even second time in a row that rumors of tension resulting in Nvidia getting dumped or 'having a fit' to go outside of TSMC for gaming instead of the decision being something that would have been chosen out of careful consideration at the start, and just like 600-1000W rumors, it's something I am sure we will see every gen
The faster this bubble drops and fails the better. It’s the next con after Crypto, Metaverse and now AI.
Bs, ai is extremely bad at most tasks that are considered high value. They are just making models bigger and bigger bigger. We are already hyper optimzed for the workload and results are terrible for the investment. AI is replacing the bottom end of the market and generally doing a bad job, that's why no ai company is making any money.No - crypto and 'metaverse' were always inflated nonsense, AI is not. It works and it keeps getting better.
It's true there is huge overinvesting and it's unsustainable in several ways
BUT
it actually works. they're rushing for AGI and it's not really that far
Plus now that they explained yields are so low, they can justify raising the price even more for BlackwellSounds to me like it's a leader just being a good leader.
Yeah, we imported some memory for development and testing at the end of September ain't mathing with a global launch mid-November.I'd like to think that if they were going to ship in time for Black Friday we'd have a lot more indicators pointing to that.
I don't think they feel the need to get anything out this year. They're not challenged at the high end now and they know AMD isn't going to release anything that could change that either.
Since they release the top of the stack first and know that those customers will buy those cards any time of the year, they don't gain a lot trying to hit a holiday launch. The average Black Friday shopper isn't in the market for a 5090.
They could also be intentionally holding off on a release until they clear out more channel inventory. Black Friday and some discounts or bundles on existing cards can certainly help with that.
Bs, ai is extremely bad at most tasks that are considered high value. They are just making models bigger and bigger bigger. We are already hyper optimzed for the workload and results are terrible for the investment. AI is replacing the bottom end of the market and generally doing a bad job, that's why no ai company is making any money.
It could just be me but I can't see the fundamental change that will allow agi , doing more of the same it's going to get us there.
There are so many fields where there simply isn't enough data to train on , how does an llm based anything handle that.
Even things ai is good at the second complex technical language enters they are terrible. They are terrible because spoken language is terrible for articulating complex problems. In my field you would expect ai to be awesome but it's isn't, I have found one really good use case for ai and the problem is I can't sell it because no one will buy it because the cost of doing the old skool way isn't actually that high( effectively fault tree development and analysis using live data sets).
Edit: the next problem is customers generally don't know what they want ,they just know they have a problem. Lots of times they actually don't even know what the problem is. Let's watch AI have a crack at that, in many technical fields that literally 1/2 your job.
Don't worry tech bros have it all figured out.
They could also be intentionally holding off on a release until they clear out more channel inventory. Black Friday and some discounts or bundles on existing cards can certainly help with that.