Memory bandwidth is likely one of the most essential specs that determines a graphics card’s efficiency potential. It displays the utmost information throughput fee between the GPU chip itself and its reminiscence packages. It is calculated by taking the width of the reminiscence bus and multiplying it by the per-pin switch fee of mentioned reminiscence packages and is often expressed by way of gigabytes-per-second. Obviously, the potential switch fee of the DRAM ICs surrounding the GPU bundle is a essential component of this equation. That’s why {hardware} nerds like us are at all times fairly eager to listen to what the ultimate transport reminiscence pace might be for brand spanking new GPUs. For AMD’s upcoming RDNA 3-based Radeon RX 7000 household, we have heard conflicting rumors, however Greymon55 is right here to set us straight: it will prime out at 20 Gbps.

Replying to VideoCardz, Greymon55 confirms that the Radeon RX 7900 XT, anticipated to be the top-end SKU in AMD’s subsequent graphics card household, will use 20 Gbps reminiscence. Assuming different leaks are correct, it’ll seemingly use twelve 2GB GDDR6 ICs for a complete of 24GB of video RAM operating at 20 Gbps throughout a 384-bit bus.

That would ship a last reminiscence bandwidth spec of 960 GB/sec, which is not the highest we have ever seen on a Radeon—that honor goes to the Radeon VII and its 1 TB/sec—however it’s nonetheless extraordinarily quick. We are a bit of shocked AMD did not go for broke with Samsung’s screaming 24-Gbps reminiscence, however maybe the value was too pricey.
Note: order doesn’t essentially point out efficiency or worth. Click to enlarge.

AMD’s extant RDNA 2 graphics playing cards are aggressive with NVIDIA’s Ampere household regardless of drastically decrease video reminiscence bandwidth. That’s principally because of their use of huge on-die caches, which AMD calls “Infinity Cache.” However, when purposes’ framebuffers exceed the capability of the Infinity Cache, as can occur regularly in excessive resolutions, efficiency falls off considerably.

The large leap in reminiscence bandwidth on AMD’s next-generation elements could also be an try and alleviate these results considerably. It may additionally be an try to enhance efficiency in sure sorts of GPU compute workloads. While the RDNA structure is targeted on gaming, the playing cards are nonetheless large compute accelerators, and it’s attainable that some vendor or one other was disenchanted within the comparatively restricted reminiscence bandwidth of AMD’s current-gen playing cards.
Unofficial Navi 31 render by @Wild_C on Twitter.

According to the leaks we have seen, the Radeon RX 7900 XT must be based mostly on the Navi 31 GPU. It is predicted to make use of a single massive Graphics Compute Die (GCD) together with six Memory and Cache Dice, or MCDs. Each MCD has its personal 32MB of cache, which is able to give the RX 7900 XT as much as 192 MB of Infinity Cache, or probably even 384 MB if the rumors a few 3D V-Cache-equipped mannequin are correct.

It could be very seemingly that the MCDs should run on the identical clock fee because the reminiscence, and it could possibly be that using 3D V-Cache is one other limiting issue on reminiscence switch fee. Whatever the case, Navi 31 will definitely be the largest Radeon we have ever seen, and we’re completely right here for it—particularly contemplating AMD’s current successes cleansing up its graphics driver.