Every 2 years, Nvidia launches a new generation of graphics cards for our PC systems, but in the wake of supply constraints and a global chip shortage, the fact that Nvidia will still be releasing their next-gen GPUs in 2022 is at least a sigh of relief. More interestingly though, it has been officially confirmed to use the TSMC 5nm process node.
Whilst previous rumors have come from (admittedly, reliable) leakers, the latest news comes directly from the Taiwanese factories, where the graphics cards will be made. The report comes from DigiTimes behind a paywall, but thankfully a brief segment was revealed on Twitter by user RetiredEnginner (@chiakokhua).
“Nvidia gaming GPUs getting major refresh next year, Taiwan factories all-out to support 5nm RTX40 generation,” the article says. “Nvidia's biennial GPU refresh coming in 2022, riding on metaverse and gaming. Following H100, based on Hopper architecture, using TSMC's 5nm + CoWoS, aimed at datacenter/AI, gaming GPU RTX40 series, based on Ada Lovelace architecture, will also tap TSMC's 5nm...”
That might be a bit difficult to read, but essentially that quote is confirming that the next-gen successor to the RTX 30 series based on Ampere will officially be the RTX 40 series based on Ada Lovelace, and will be using the TSMC 5nm process node. The H100 is an AI/datacenter GPU by Nvidia, which will also get a similar upgrade.
Obviously these will go head-to-head against AMD’s RX 7000 series graphics cards based on RDNA3. But what can we expect in terms of performance? Well, based on rumors at the moment (so take this information with a grain of salt), we could see a similar performance increase that we saw with the RTX 20 series to RTX 30 series.
The rumors suggest the RTX 40 series flagship graphics card (AD102 SKU) will have 18,432 CUDA Cores, which is roughly 70% more than the flagship GPU of the RTX 30 series, the RTX 3090. So we can assume this particular GPU will be called the RTX 4090. Rumors also suggest it will have a clock speed of 2.3-2.5GHz.
With a 2.3-2.5GHz clock speed, we could see compute performance (FP32) up to 85-92 TFLOPS. In comparison, the RTX 3090 achieves 36 TFLOPs, and the RTX 2080 Ti achieved 13 TFLOPS. Obviously TFLOPS are not an exact indication of gaming performance, as in reality the 3090 was only 50-60% faster than the 2080 Ti, despite the 170% increase in performance the TFLOPs data would suggest.
Again, take it all with a big grain of salt, but if these rumors are true we could be seeing another impressive leap in GPU performance next year. And based on the comments made in the DigiTimes report, along with lessons learned from the past 2 years, we can hope there will be much more supply this time round.
What do you think? Are you excited for the RTX 40 series graphics cards? How do you feel about the current (and early) performance expectations? Are you expecting a bigger or smaller leap in performance than outlined above? And what do you think the supply situation will be like? Let us know your thoughts!