QuoteGaming Rubin could use Nvidia's custom TSMC 3NP node
Correction: TSMC's custom NVIDIA 3NP node.
QuoteUnfortunately, there is no telling what improvements the RTX 60 series will bring over Blackwell.
Going from a 4NP (which is basically their 5nm node) to a 3NP node, is going to improve, as usual, the raw compute per Watt for any core types by approx. 25 to 33%. This is the base on which every other improvement is going to be based on: NV could increase the proportion of tensor cores, which would then lead to a better FPS per Watt of DLSS performance.
QuoteDoing so would also risk worsening the power connector problems that have already affected current-generation RTX 50 series graphics cards, as well as some last-generation models.
Or NV could simply add a second connector and make a 500 to 2*575W TDP GPU (of course, a 2*575W GPU would be hilarious), or design a new 800W connector and make e.g. a 800W TDP GPU (for people who prefer to heat their room with (expensive) electricity).
QuoteThat leaves us with a bigger question: where is the RTX 50 Super lineup? An earlier report stated they're gathering dust in warehouses because Nvidia and distributors can't zero-in on a price. Will it be unveiled alongside the RTX 60 series as a 'cheaper' alternative? Has it been mothballed for good? Will it get a stealth release when nobody is looking? Only time will tell.
If, let's say, the RTX 60 series is still 1 year away, NV could still release a Super series soon and cache on the 3 GB GDDR7 densities once and then on the RTX 60 series again.
Because if they release it alongside the RTX 60 series GPUs, then people would get the RTX 60 series GPUs straight up and ignoring the Super ones.
Quote from: Sin on Today at 11:07:52What's the point of launching it, if most of your GPUs go to data centers and the one that don't - are crazy overpriced..
See news, people are still buying them more than the previous year (maybe it's the AI folks that are buying them), despite the crazy price increases over the previous year.