Anyone who has been round in PC gaming for a very long time will know that each time a brand new high-end graphics card will get launched, provide is rarely sufficient to satisfy demand. In the case of the GeForce RTX 5090, board companion MSI says that its playing cards can have restricted availability and it is all the way down to an inadequate variety of GB202 GPUs from Nvidia.
While this information in all probability is not a shock to any graphics card fanatic, it is price noting that this is not some hearsay or leak—it is an official assertion by MSI, certainly one of Nvidia’s key GPU companions, as reported by IT Home (by way of Wccftech).
As with any manufactured product, the general availability can solely be pretty much as good because the weakest hyperlink within the provide chain. In this occasion, in line with MSI, it is the variety of GPUs being supplied by Nvidia.
MSI, Asus, Gigabyte, and all different AIB distributors buy graphics processors from Nvidia, which then distributes them principally from its centres in Hong Kong. However, one cannot merely rock up and ask for 100,000 chips—orders must be positioned properly prematurely after which Nvidia will allocate processors based mostly on numerous elements, similar to the scale of the order, relationship with the companion, what offers are in progress, and so forth.
It’s not simply MSI that’s struggling to satisfy demand. Zotac Korea says that there is not any likelihood of any 5090s being accessible (by way of Videocardz) till early February and that, as issues at present stand, there isn’t a confirmed date for the discharge of its RTX 5080 fashions.
This all tallies with the comment from UK retailer Overclockers saying that it’s going to solely have a ‘single digit’ variety of 5090 playing cards accessible when the GPU is launched for buy.
So what to make of this? Is Nvidia intentionally constraining the availability of its GB202 chips, to cost board distributors extra for every tray of processors they order? Or is it all the way down to the truth that at 750 mm2 in dimension, every silicon wafer is not going to supply many absolutely working dies, even when the yields are excellent, so TMSC simply cannot make sufficient of them?
Even if the previous is not remotely true, the latter might be to a sure diploma, and I can not think about that Nvidia is sitting on an enormous pile of Blackwell chips. We picked up by means of the grapevine varied mutterings that Nvidia commenced manufacturing of RTX 50-series GPUs fairly late in 2024, although we could not pin down such rumours nor decide any doable clarification for such a choice.
However, one issue that will have performed a job is the consumer-grade Blackwell chips are made on the identical course of node as datacentre Blackwell, i.e. TSMC N4 (Ada Lovelace GPUs are made on a customized model of that node).
Given simply how a lot cash Nvidia makes from promoting its large AI processors, I ought to think about TSMC order books had been chock filled with GB100s, leaving little spare capability to begin RTX 50-series chips in earnest.
And it is price noting that TMSC’s N4 course of node is in sizzling demand. As properly as Nvidia’s total Blackwell vary, the Taiwanese fabrication big additionally produces AMD’s Zen 5 chiplets, in addition to its Strix Halo and the most recent Hawk Point APUs on the identical node. They’re all a lot smaller than the GB202, so fewer wafers must be allotted for AMD’s orders, however it could actually solely churn out so many every month.
Not that that is consolation in the event you had been hoping to snag an RTX 5090 at its MSRP once they hit retailers’ cabinets (if they even attain them) on the finish of this month. The 575 W monster is probably the most highly effective gaming graphics card cash should purchase however it might appear that for the following month or so, no sum of money could possibly get you one.
