EVOLUTIONARY GREEN GPUS
Is Nvidia’s Volta architecture the revolutionary GPU we’ve all been waiting for?
IT WOULD BE FAIR TO SAY that Nvidia has demonstrated a good show of force when it came to the grand GPU war of the last two years. Pascal has been out for well over 18 months now, and still continues to reign dominant over its red competitor. Radeon’s Technology Group simply hasn’t been able to compete, either at the mid range, with its Polaris-based RX 400 and 500 series cards, or even more recently at the very high end, with Vega falling flat on its face, far from the hype that surrounded it. But with little competition, what’s the next step for the great green giant?
Well, that would be Volta. We’ve long been hearing rumors about Nvidia’s latest and greatest architecture finally making it to the consumer market, and it’s only now that we have any inkling as to what it might actually include. We know that TSMC is manufacturing the core, we know that it’s operating off the 12nm FinFET production node, and we also know that there are already enterprisegrade accelerator cards in existence—powered, of course, by the magnificent might of the GV100 core. So, what can we expect from this new GPU lineup when it actually does launch consumer-side?
We can’t say for certain, but looking back at the past may give us a clue. We know the number of CUDA cores found on both the current Pascal P100 GPU accelerator and the GTX 1080 Ti are the same. If that’s the case, it’s likely that Volta’s V100 and the consumer-side 2080 Ti (whenever that launches; think 2019) will be identical. We can then calculate the percentage difference between the CUDA cores on the GTX 1080 Ti and the 1080, apply that to the V100’s spec, and come up with a rough estimate of just how many additional CUDA cores Nvidia’s next-gen flagship (the GTX 2080) might hold. The short answer is that it’s a 33 percent increase in CUDA cores, and therefore pure GPU grunt over the last generation, and that’s not including any architectural magic, or transition across to GDDR6 or HBM memory. A 33 percent minimum increase in brute force performance, which is exciting enough, if current price points hold across the generational jump.
Nvidia’s co-founder and CEO Jen-Hsun hasn’t stipulated that we should expect an immediate response to Vega, so it’s unlikely we’ll see these cards before the first quarter of 2018. However, with SK Hynix’s recent announcement of its latest GDDR6 memory standard being produced for highspec cards already, and Volta enterprise-grade GPUs now being available to the public, we’d be very surprised if we didn’t see these launch before or at Computex 2018.