Micron has shared some efficiency figures for its next-gen graphics reminiscence, and if these numbers develop into true, it may imply nice issues for upcoming graphics playing cards.
The corporate claims that its GDDR7 VRAM will supply as much as 30% efficiency enchancment in gaming situations, and this reportedly applies each to video games that rely closely on ray tracing and on pure rasterization. Will Nvidia’s RTX 50-series, which is alleged to be utilizing GDDR7 reminiscence, develop into an even bigger improve than anticipated?
Within the knowledge, which was first shared by harukaze5719 on X (previously Twitter), Micron pits its new GDDR7 32Gb/s modules in opposition to its GDDR6 VRAM that max out at 20GB/s. For comparability, GDDR6X reminiscence can go as much as 24GB/s per pin, so it’s an enormous improve it doesn’t matter what, but it surely’s value noting that GDDR7 reminiscence will begin at 28Gb/s.
Get your weekly teardown of the tech behind PC gaming
Micron additionally lists just a few of the important thing advantages of its GDDR7 tech. For starters, the bandwidth is elevated by as much as 60% should you examine the highest GDDR6 and GDDR7 modules, with GDDR7 with the ability to obtain over 1.5TB/s in system bandwidth.
Regardless of these spectacular positive aspects, Micron additionally studies that GDDR7 is as much as 50% extra environment friendly by way of energy consumption, and its new sleep modes can massively reduce on the quantity of energy consumed whereas on standby — as much as 70%. Micron additionally managed to scale back response instances by as much as 20%, which can be useful for generative AI workloads.
micron, 30% FPS enhance
28Gbps: https://t.co/hj9RtqyraN
32Gbps: https://t.co/sWJAe1oDnG pic.twitter.com/lsToW4nwr9— Posiposi (@harukaze5719) June 29, 2024
With first-party benchmarks like this, it’s necessary to at all times take them with a bit of little bit of skepticism. Nonetheless, some numbers are exhausting to dispute. Whereas we will’t ensure how common that 30% enchancment in frames per second (fps) goes to be, we do know for a proven fact that the quickest GDDR7 reminiscence will supply a large improve in reminiscence bandwidth.
Micron’s comparisons match up a setup with 12 GDDR7 built-in circuits (ICs) and a 384-bit reminiscence interface in opposition to a precise match from the earlier era, and that brings the bandwidth over the 1.5TB/s line. In the meantime, the RTX 4090, geared up with quicker GDDR6X modules that hit as much as 21Gbps, maxes out at a bit of over 1TB/s. The positive aspects are undoubtedly there.
This feels like excellent news for Nvidia. Current rumors concerning the specs for the RTX 50-series depart so much to be desired, with the same enormous enchancment within the RTX 5090 and considerably smaller boosts throughout the decrease finish of the product stack. However with this a lot quicker reminiscence, we’d see GPUs that share comparable CUDA core counts nonetheless shoot forward in fps simply due to the quicker VRAM.
In fact, that 30% determine isn’t going to use to each GPU. A slender reminiscence bus can stifle even the very best VRAM, as we’ve seen in GPUs just like the RTX 4060 Ti 16GB. Within the RTX 40-series, Nvidia stored the quantity of VRAM pretty conservative for many of the graphics playing cards, and the reminiscence interface was the identical method. With that very same strategy within the RTX 50-series, GDDR7 reminiscence will nonetheless be an enchancment, however that 30% could also be reserved for GPUs just like the RTX 5090.
Micron’s new reminiscence modules sound fairly superb, and hopefully, we’ll be seeing them this manner in Nvidia’s upcoming graphics playing cards. AMD is alleged to be sticking to GDDR6 in RDNA 4. This might find yourself giving Nvidia an edge over its competitor in additional methods than one.