DEPLOY

Chip leaderboard

Most HBM memory per chip

SambaNova SN40L leads at 1,500 GB. 48 chips ranked.

How much high-bandwidth memory each chip has. This is the ceiling on the largest model that fits on one chip without splitting. Cerebras WSE-3 lists on-die SRAM, not HBM, but it's included since the question is capacity.

#ChipValueInputs
1SambaNova SN40L
SambaNova Systems
1,500 GBType: DDR5 (three-tier: 64 MB SRAM + 64 GB HBM3 + 1.5 TB DDR5)
2AMD Instinct MI400
AMD
432 GBType: HBM4
3NVIDIA GB200 Grace Blackwell Superchip
NVIDIA
384 GBType: HBM3e
4AMD Instinct MI350 series
AMD
288 GBType: HBM3e
5AMD Instinct MI355X
AMD
288 GBType: HBM3e
6NVIDIA B300 (Blackwell Ultra)
NVIDIA
288 GBType: HBM3e
7NVIDIA GB300 (Blackwell Ultra)
NVIDIA
288 GBType: HBM3e
8NVIDIA Rubin (R100)
NVIDIA
288 GBType: HBM4
9AMD Instinct MI325X
AMD
256 GBType: HBM3e
10AMD Instinct MI300X
AMD
192 GBType: HBM3
11Google TPU v7 (Ironwood)
Google
192 GBType: HBM3e
12Huawei Ascend 910D
Huawei
192 GBType: HBM3
13NVIDIA B100 (Blackwell)
NVIDIA
192 GBType: HBM3e
14NVIDIA B200 (Blackwell)
NVIDIA
192 GBType: HBM3e
15NVIDIA GH200 Grace Hopper Superchip
NVIDIA
144 GBType: HBM3e
16NVIDIA H200 Tensor Core GPU
NVIDIA
141 GBType: HBM3e
17AMD Instinct MI250X
AMD
128 GBType: HBM2e
18Huawei Ascend 910C
Huawei
128 GBType: HBM2e
19Intel Gaudi 3
Intel Corporation
128 GBType: HBM2e
20Meta MTIA v2
Meta
128 GBType: LPDDR5
21NVIDIA DRIVE Thor
NVIDIA
128 GBType: LPDDR5X
22NVIDIA Jetson AGX Thor
NVIDIA
128 GBType: LPDDR5X
23AWS Trainium 2
Amazon
96 GBType: HBM3
24Intel Gaudi 2
Intel Corporation
96 GBType: HBM2e
25NVIDIA H20
NVIDIA
96 GBType: HBM3
26Google TPU v5p
Google
95 GBType: HBM2e
27NVIDIA A100 Tensor Core GPU
NVIDIA
80 GBType: HBM2e
28NVIDIA A800
NVIDIA
80 GBType: HBM2e
29NVIDIA H100 Tensor Core GPU
NVIDIA
80 GBType: HBM3
30NVIDIA H800
NVIDIA
80 GBType: HBM3
31Biren BR100
Biren Technology
64 GBType: HBM2e
32Huawei Ascend 910B
Huawei
64 GBType: HBM2e
33NVIDIA Jetson AGX Orin
NVIDIA
64 GBType: LPDDR5
34Cambricon MLU370-X8
Cambricon Technologies
48 GBType: LPDDR5
35NVIDIA L40S
NVIDIA
48 GBType: GDDR6
36Cerebras WSE-3 (Wafer-Scale Engine 3)
Cerebras Systems
44 GBType: on-die SRAM
37AWS Inferentia 2
Amazon
32 GBType: HBM3
38AWS Trainium (Trn1)
Amazon
32 GBType: HBM2e
39Google TPU v4
Google
32 GBType: HBM2
40Google TPU v6 (Trillium)
Google
32 GBType: HBM3
41NVIDIA V100 Tensor Core GPU
NVIDIA
32 GBType: HBM2
42Tenstorrent Blackhole
Tenstorrent
32 GBType: GDDR6
43NVIDIA A10 Tensor Core GPU
NVIDIA
24 GBType: GDDR6
44Google TPU v5e
Google
16 GBType: HBM2
45NVIDIA T4 Tensor Core GPU
NVIDIA
16 GBType: GDDR6
46Tesla HW4 (FSD Computer 2)
Tesla
16 GBType: GDDR6
47Tenstorrent Wormhole
Tenstorrent
12 GBType: GDDR6
48Groq LPU (Language Processing Unit)
Groq
0.23 GBType: SRAM (230 MB on-die)

48 chips ranked. 12 more are in our catalog but don't publish the numbers needed for this ranking.

Other rankings: Best perf-per-watt (TFLOPS/W) · Highest memory bandwidth · 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.