DEPLOY

Chip leaderboard

Highest memory bandwidth

Cerebras WSE-3 (Wafer-Scale Engine 3) leads at 21,000,000 GB/s. 43 chips ranked.

How fast each chip can read and write its own memory. This is the ceiling for LLM inference speed, where every generated token pulls the full model out of memory again. Higher is faster.

#ChipValueInputs
1Cerebras WSE-3 (Wafer-Scale Engine 3)
Cerebras Systems
21,000,000 GB/sType: on-die SRAM
2Groq LPU (Language Processing Unit)
Groq
80,000 GB/sType: SRAM (230 MB on-die)
3AMD Instinct MI400
AMD
19,600 GB/sType: HBM4
4NVIDIA GB200 Grace Blackwell Superchip
NVIDIA
16,000 GB/sType: HBM3e
5AMD Instinct MI350 series
AMD
8,000 GB/sType: HBM3e
6AMD Instinct MI355X
AMD
8,000 GB/sType: HBM3e
7NVIDIA B100 (Blackwell)
NVIDIA
8,000 GB/sType: HBM3e
8NVIDIA B200 (Blackwell)
NVIDIA
8,000 GB/sType: HBM3e
9NVIDIA B300 (Blackwell Ultra)
NVIDIA
8,000 GB/sType: HBM3e
10NVIDIA GB300 (Blackwell Ultra)
NVIDIA
8,000 GB/sType: HBM3e
11Google TPU v7 (Ironwood)
Google
7,400 GB/sType: HBM3e
12AMD Instinct MI325X
AMD
6,000 GB/sType: HBM3e
13AMD Instinct MI300X
AMD
5,300 GB/sType: HBM3
14NVIDIA GH200 Grace Hopper Superchip
NVIDIA
4,900 GB/sType: HBM3e
15NVIDIA H200 Tensor Core GPU
NVIDIA
4,800 GB/sType: HBM3e
16Huawei Ascend 910D
Huawei
4,000 GB/sType: HBM3
17NVIDIA H20
NVIDIA
4,000 GB/sType: HBM3
18Intel Gaudi 3
Intel Corporation
3,675 GB/sType: HBM2e
19NVIDIA H100 Tensor Core GPU
NVIDIA
3,350 GB/sType: HBM3
20NVIDIA H800
NVIDIA
3,350 GB/sType: HBM3
21AMD Instinct MI250X
AMD
3,200 GB/sType: HBM2e
22Huawei Ascend 910C
Huawei
3,200 GB/sType: HBM2e
23AWS Trainium 2
Amazon
2,900 GB/sType: HBM3
24Google TPU v5p
Google
2,765 GB/sType: HBM2e
25Intel Gaudi 2
Intel Corporation
2,450 GB/sType: HBM2e
26Biren BR100
Biren Technology
2,300 GB/sType: HBM2e
27NVIDIA A100 Tensor Core GPU
NVIDIA
2,039 GB/sType: HBM2e
28NVIDIA A800
NVIDIA
2,039 GB/sType: HBM2e
29Google TPU v6 (Trillium)
Google
1,600 GB/sType: HBM3
30Huawei Ascend 910B
Huawei
1,600 GB/sType: HBM2e
31Google TPU v4
Google
1,200 GB/sType: HBM2
32NVIDIA V100 Tensor Core GPU
NVIDIA
900 GB/sType: HBM2
33NVIDIA L40S
NVIDIA
864 GB/sType: GDDR6
34AWS Inferentia 2
Amazon
820 GB/sType: HBM3
35AWS Trainium (Trn1)
Amazon
820 GB/sType: HBM2e
36Google TPU v5e
Google
819 GB/sType: HBM2
37NVIDIA A10 Tensor Core GPU
NVIDIA
600 GB/sType: GDDR6
38Tenstorrent Blackhole
Tenstorrent
512 GB/sType: GDDR6
39NVIDIA T4 Tensor Core GPU
NVIDIA
320 GB/sType: GDDR6
40Tenstorrent Wormhole
Tenstorrent
288 GB/sType: GDDR6
41NVIDIA Jetson AGX Thor
NVIDIA
273 GB/sType: LPDDR5X
42Meta MTIA v2
Meta
205 GB/sType: LPDDR5
43NVIDIA Jetson AGX Orin
NVIDIA
205 GB/sType: LPDDR5

43 chips ranked. 17 more are in our catalog but don't publish the numbers needed for this ranking.

Other rankings: Best perf-per-watt (TFLOPS/W) · Most HBM memory per chip · Highest FP8 dense throughput · Highest FP16 dense throughput · Highest INT8 dense throughput · Cheapest per TFLOP at launch (list price) · Most transistors per die

See also: every chip on record · head-to-head comparisons.