They aren't? It's a pretty tightly BOM-optimized part.
Is it? Standard ret limit is 26 by 33.
They aren't? It's a pretty tightly BOM-optimized part.I said AMD would be limited in what they can offer price wise
Is it? Standard ret limit is 26 by 33.Hold up, 31mm is outside of normal reticle aperture limits.
NV doing something funky.
Yeah I forgot it is a rectangle.Is it? Standard ret limit is 26 by 33.
They aren't? It's a pretty tightly BOM-optimized part.
We'll see about that. If the die size is 350-380mm2, Im not expecting less than $549 and $649 at best for the non-XT and XT parts, at best, which isnt exactly going to set the world on fire compared to the 5000 series offerings.
Scalpers reportedly selling 5090s for as high as $7000. A lot of them are at least much more reasonable at $3500-$5000. Lots of watchers as well and some have a lot of bids over $3000. That's why I think the next flagship needs to have an MSRP that is higher than whatever these buyers have shown they will pay on ebay for a scalped 5090. If these cards sell for $3500+ on ebay, then the 6090 should have an MSRP of at least $4500. The price should match what people are willing to pay, so I think Asus was right to charge $2800. People will complain about the price as they take out their credit card and add to cart.
It's 350 or less.If the die size is 350-380mm2
Probably ~800GB/s at that price (and that's assuming it remains the same) but yeah, if true that 275GB/s is not great. I guess what you'd be paying for would be the 128GB RAM then?Apple will offer ~1TB/s for $4K
You pay for CUDA.I guess what you'd be paying for would be the 128GB RAM then?
Scalpers exist only because the manufacturer temporarilly cannot satisfy the demand. Do not atribute any important role (as influencing the normal sale price) to them.Scalpers reportedly selling 5090s for as high as $7000. A lot of them are at least much more reasonable at $3500-$5000.
No, only the monopolist charges people what are they willing to pay, and monopols are subjected to all sorts of regulation, because it is simply WRONG.The price should match what people are willing to pay.
And power too if Qualcomm has anything to say about itIt wasn't. It was lot less dense, though.
Yeah TSM nodes are really good at density.
I think this needs to end, Nvidia should to stop selling flagship cards and make them available through subscription only. $200/month and you get your 6090 delivery, no more bids and wasted time. For a premium $300/month you can even get early access to new cards before the press does (24 months subscription only, terms and conditions apply)If these cards sell for $3500+ on ebay, then the 6090 should have an MSRP of at least $4500. The price should match what people are willing to pay, so I think Asus was right to charge $2800. People will complain about the price as they take out their credit card and add to cart.
Subscription GPUs or any hardware sub will never take off. It’s not a good business model.I think this needs to end, Nvidia should to stop selling flagship cards and make them available through subscription only. $200/month and you get your 6090 delivery, no more bids and wasted time. For a premium $300/month you can even get early access to new cards before the press does (24 months subscription only, terms and conditions apply)
Meanwhile everyone else can buy prepaid cards, with credit on them. Buy a $300 5060 with 300 hours of gaming on it and 6 months credit life. The more you buy, the more you save!
Gamers don't own their games anymore, why should they bother owning expensive cards that nowadays differentiate through their software features? Nvidia could even introduce a free tier in their subscription stack, using AMD hardware.Subscription GPUs or any hardware sub will never take off. It’s not a good business model.
When you look at the market position of Nvidia, I actually think that they should be regulated strictly and their excessive profit limited.No, only the monopolist charges people what are they willing to pay, and monopols are subjected to all sorts of regulation, because it is simply WRONG.
Jensen might need a cheaper leather jacket for CES 2026 showing 6090 made on N2 process in volume only for $999, I guess this speculation is for another thread...
Yup, they trained their model using a fraction of the hardware resources used by the western big boys. Investors are now worried that the huge datacenter capex increase is not really necessary. (or may be rendered obsolete by a simple open source release)Can someone ELI5 why DeepSeek being succesful would crush Nvidia prices? Is it not running on Nvidia GPUs?
It has potential crush NASDAQCan someone ELI5 why DeepSeek being succesful would crush Nvidia prices?
Lower training costs, turns out multi-billion datacenters for AI are not necessary.Can someone ELI5 why DeepSeek being succesful would crush Nvidia prices? Is it not running on Nvidia GPUs?
They are still necessary for inference which is getting much heavier hit with new "AI" models, but that is far less protected by CUDA moat (and NVLINK/Mellanox switching gear) and price premium for those isn't anywhere as high as for NVIDIA stuff.Lower training costs, turns out multi-billion datacenters for AI are not necessary.