Welcome again, buddies, to the present that by no means ends: a fast peek on the newest GPU leaks. As normal for this technology, this one comes courtesy of kopite7kimi who’s tweeting a few particular NVIDIA Ada GPU product with the SKU of both “PG136-SKU360” or “PG139-SKU360”.
That is a minimize of eight shader modules from the complete 84 on the AD103 die. No cuts to the reminiscence controllers, although; the anticipated specs for this product embody 16 GB of GDDR6X reminiscence working at 21 Gbps throughout a 256-bit bus, giving the GPU some 672 GB/sec of reminiscence bandwidth. It is not clear if the cuts to shader modules will embody any slashes to that GPU’s 64 MB of last-level cache.
The purported product would have a complete board energy of round 420 watts. That will put it in an analogous vary with the GeForce RTX 3090 Ti regardless of being on a a lot smaller course of node. As anticipated, the advantages of the smaller course of have apparently been utilized to clock fee; kopite7kimi says that this supposed GeForce RTX 4080 scores round 15,000 factors in 3DMark Time Spy Excessive.
For those who do not comply with 3DMark, let’s simply say that such a rating can be a #1 world file at the moment—albeit solely simply barely. The highest single-GPU rating is presently held by Korean overclocker biso biso who achived 14977 factors with a GeForce RTX 3090 Ti slotted right into a system with a hot-clocked 28-core Xeon W-3175X CPU.
It’s in no way stunning {that a} next-generation GPU may surpass the previous-generation merchandise however seeing a second-tier GPU demolish a closely overclocked halo product from the earlier line-up is fairly darn spectacular.
