Can I deploy NVIDIA GH200 or AMD MI210 on Runcrate today?

NVIDIA GH200 is not in stock on Runcrate right now; leave your email on its pricing page to register interest. AMD MI210 is not in stock on Runcrate right now; leave your email on its pricing page to register interest.

How much VRAM does each GPU have?

NVIDIA GH200 has 96GB HBM3. AMD MI210 has 64GB HBM2e.

GH200 or MI210: which holds a 70B model in 8-bit weights?

It needs 70 GB of 8-bit weights. On the GH200: 1 GPU, 26 GB spare. On the MI210: 2 GPUs (128 GB). Pick the GH200: fits on one GPU.

GPU comparison

GH200 vs MI210.

The GH200 is ahead on memory (96 GB vs 64 GB) and speed (5.5x the FP16 throughput and up to 2.4x the memory bandwidth). The MI210 makes sense when your model fits in 64 GB and its hourly price is lower.

Rent either on Runcrate

GH200Not in stock right now
Register interest GH200 pricing
MI210Not in stock right now
Register interest MI210 pricing

Which is better for

Which to pick for each job. Memory fit and throughput, from the datasheets.

JobGH200MI210Pick

Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
GH200: 1 GPU, 26 GB spare
MI210: 2 GPUs (128 GB)
GH200. Fits on one GPU.
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
GH200: 1 GPU, 80 GB spare
MI210: 1 GPU, 48 GB spare
GH200. Same GPU count, up to 2.4x the memory bandwidth.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
GH200: 1 GPU, 50 GB spare
MI210: 1 GPU, 18 GB spare
GH200. Same GPU count, 5.5x the FP16 throughput.
Train on a full node
FP16 tensor throughput per node
GH200: 1x, 990 TFLOPS
MI210: 4x, 724 TFLOPS
GH200. 5.5x the FP16 throughput per GPU.
Throughput per watt
FP16 TFLOPS per 100 W of TDP
GH200: Superchip power, not a GPU TDP
MI210: 60 TFLOPS per 100 W
Either. Not comparable.

Compare another pair

Hopper

NVIDIA GH200

Not in stock right now

Memory: 96 GB HBM3
FP16 tensor: 990 TFLOPS
Bandwidth: up to 4.0 TB/s

Full pricing details

CDNA 2

AMD MI210

Not in stock right now

Memory: 64 GB HBM2e
FP16 tensor: 181 TFLOPS
Bandwidth: 1.6 TB/s

Full pricing details

At a glance

Spec by spec. The longer bar is the better number.

Memory

NVIDIA GH20096 GB

AMD MI21064 GB

Memory bandwidth

NVIDIA GH200up to 4.0 TB/s

AMD MI2101.6 TB/s

FP16 tensor

NVIDIA GH200990 TFLOPS

AMD MI210181 TFLOPS

FP32

NVIDIA GH20067 TFLOPS

AMD MI21023.9 TFLOPS

Full specs

The complete spec sheet.

Spec	NVIDIA GH200	AMD MI210
Architecture	Hopper	CDNA 2
Memory	96 GB	64 GB
Memory type	HBM3	HBM2e
Memory bandwidth	up to 4.0 TB/s	1.6 TB/s
FP32	67 TFLOPS	23.9 TFLOPS
FP16 tensor	990 TFLOPS	181 TFLOPS
INT8	1979 TOPS	192 TOPS
Power	up to 1000 W per superchip (CPU, GPU and memory)	300 W per GPU (TDP)
Form factor	Superchip	PCIe
NVLink	One GPU	No
GPUs per node	1	4
Hourly price (live)	Not in stock right now	Not in stock right now

Questions

GH200 or MI210. What buyers ask first.

Other comparisons