New leaks are coming around the AMD Radeon RX 7900XTthe company’s future high-end GPU based on the graphics chip Navi 31. This information comes from @Greymon55a fairly reliable source of leaks, which indicates that this GPU with a multichip design would offer 28% improvement in FP32 performance compared to the initial rumors of a 75 TFLOP performance, so we’re basically talking about quadruples the performance of the AMD Radeon RX 6900 XTthe company’s current high-end GPU based on the Navi 21 (23.04 TFLOP) graphics chip.
By digging into the information, this GPU with a multichip design, will combine two manufacturing processes of TSMC: lbone 6nm + 5nm. This will bring to life a graphics chip that will offer up to 15360 stream processors at a frequency that could touch 3000MHz In order to provide a performance of 92 TFLOPs versus 75 TFLOPs on today’s high-end GPU.
This graphics chip could accompany 256 or 512 MB of cache memory Infinity Cache, could offer 24 GB of 18 Gbps GDDR6 memory accompanied by a 256-bit interface, and for now the energy consumption it will reach is unknown, recalling that its direct rival, the Nvidia GeForce RTX 4090it is assumed that it will have a consumption of 600W of power achieve a performance of 100 TFLOPs.
|Comparison of the next generation of high-end GPUs (as rumored)|
|Model||GeForce RTX 4090||RadeonRX7900|
|Lithography||TSMC 5nm or 4nm||TSMC 5nm + 6nm|
|GPU design||Monolithic||Multi-Chip Module (MCM)|
|Ten||a||2 GCDs + 4 MCDs + 1 IOD|
|FP32 Cores||18432 CUDA cores||15360 stream processors|
|Memory||24 GB of GDDR6X memory||24 GB of GDDR6 memory|
|Freq. memory + interfaces||21 Gbps with 384 bit||NA / 256 bit|
|L2 cache||96 MB||256MB/512MB|
|Energy consumption||600W (TDP)||N / A|
|Launch||Q3/Q4 2022||Q3/Q4 2022|
|Performance||~100 TFLOPs FP32||~92 TFLOPs FP32|
via: video cardz