🔥 Hot Chips
Ranked by importance, recency, and FP16/BF16 performance — higher score means higher rank.
- ChipVendorArchMemoryBandwidthFP16TDPCategory
- B300 UltraNVIDIABlackwell Ultra288 GB HBM3e
8 TB/s2,700 TFLOPS1000WTraining - MI355XAMDCDNA 3.5288 GB HBM3e
8 TB/s2,600 TFLOPS1000WTraining - TPU IronwoodGoogleTPU v7192 GB HBM
5 TB/s2,400 TFLOPS1000WTraining - B200NVIDIABlackwell192 GB HBM3e
8 TB/s2,250 TFLOPS1000WTraining - Ascend 920HuaweiDa Vinci v496 GB HBM3
4 TB/s1,800 TFLOPS400WTraining - Ascend 950DTHuaweiDa Vinci v5144 GB HiZQ 2.0
4 TB/s1,000 TFLOPS400WTraining - Ascend 950PRHuaweiDa Vinci v5128 GB HiBL 1.0
3 TB/s1,000 TFLOPS300WInference - MI300XAMDCDNA 3192 GB HBM3
5.3 TB/s1,300 TFLOPS750WTraining - H200 SXMNVIDIAHopper141 GB HBM3e
4.8 TB/s989 TFLOPS700WTraining - MTT S5000Moore ThreadsMUSA 4.080 GB GDDR6X
1.6 TB/s1,000 TFLOPS350WTraining - H100 SXMNVIDIAHopper80 GB HBM3
3.35 TB/s989 TFLOPS700WTraining - Ascend 910CHuaweiDa Vinci 3.064 GB HBM2e
~3.2 TB/s800 TFLOPS310WTraining - MLU690CambriconMLUarch 04196 GB HBM3
3.35 TB/s700+ TFLOPS~500WTraining - WSE-3CerebrasWSE-340 GB SRAM
21 PB/s on-chip125 TFLOPSWhole systemWafer
📰 Latest News
Industry news, product launches, technical deep-dives, and selection guides.View all articles →