The latest leak comes out regarding the configuration of the AMD’s MI300 GPU

More From Author

See more articles

Myntra Upcoming Sales 2025: Your Fashion Calendar for Maximum...

Myntra Upcoming Sales 2025 In the ever-evolving world of fashion e-commerce, Myntra continues to be India's go-to destination...

Dimensity 6020 vs Snapdragon 695: Mid-Range Chipset Battle

Dimensity 6020 vs Snapdragon 695: Qualcomm Snapdragon 695 5G (SD695) is a fast mid-range ARM-based SoC found...

My Jio Recharge Plans as of January 4,...

My Jio Recharge Plans: Since its establishment in 2016, Reliance Jio has made a remarkable impact on...

According to a recent leak, we now have a would-be confirmation of the configuration of the MI200 graphics processor. The latest leak comes from the ROCm software update which states that the accelerator offers 110 Compute Units.

However, something that isn’t yet made clear is whether this configuration refers to the whole card or just a single graphics chipset.

In the leaked news, the GPU ID string of GFX90A_110 suggests the configuration of the full MI200 accelerator or just one chipset, meaning that it is either the 220 CUs (2x 110) or 110 CUs (2x 55).

Many speculations are surrounding the MI200 accelerator, and according to the famous leaker @Kepler_L2, the MI300 will be featuring twice as many GPU dies as the MI200. All these rumours have one thing in common is that future AMD accelerators based on CDNA architectures will be going to continue to use the MCM design with an increasing number of GCD (Graphics Complex Die) or graphics tiles/chipsets.

There is likely not much information regarding the MI300 GPU and even its codename is not yet known, however, it is likely going to adopt a name after another giant star, such as Rigel or Antares.

AMD Instinct Accelerators
Accelerator NameAMD Radeon Instinct MI60AMD Instinct MI100AMD Instinct MI200AMD Instinct MI300
Architecture7nm GCN5 (GFX906)7nm CDNA1 (GFX908)CDNA2 (GFX90A)CDNA3 (?)
GPUVega 20ArcturusAldebaran (MCM)? (MCM)
Compute Tiles1124
Compute Units64 (64)1202x 110 or 2x 554x (?)
FP32 Cores (Full GPU)4096 (4096)7680 (8192)TBC4x (?)
GPU Clock Speed1800 MHz~1500 MHzTBCTBC
FP16 Compute29.5 TFLOPS185 TFLOPSTBCTBC
FP32 Compute14.7 TFLOPS23.1 TFLOPSTBCTBC
FP64 Compute7.4 TFLOPS11.5 TFLOPSTBCTBC
VRAM32 GB HBM232 GB HBM2128 GB HBM2ETBC
Memory Clock1000 MHz1200 MHzTBCTBC
Memory Bus4096-bit4096-bitTBCTBC
Memory Bandwidth1 TB/s1.23 TB/sTBCTBC
Form FactorDual Slot, Full LengthDual Slot, Full LengthOAMTBC
CoolingPassive CoolingPassive CoolingTBCTBC
TDP300W300WTBCTBC

source

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.

━ Related News

Featured

━ Latest News

Featured