AMD Instinct MI350X OAM vs NVIDIA H200 SXM 141GB

The Instinct MI350X delivers 2.3x the FP8 throughput of the H200 (4,614 vs 1,979 TFLOPS dense). The Instinct MI350X also carries 147 GB more memory (288 GB vs 141 GB).

Instinct MI350X: CDNA 4, 2025 H200: Hopper, 2024

FP8 dense lead

2.3x

Instinct MI350X: 4,614 vs 1,979 TFLOPS

Memory

288 vs 141 GB

147 GB more for the Instinct MI350X

Bandwidth

8.0 TB/s vs 4.8 TB/s

Instinct MI350X moves data faster

TDP

1.0 kW vs 700 W

H200 draws 300 W less

See rental pricing from $0.79/hr Download comparison sheet Printable version →

AMD Instinct MI350X OAM

Leads on FP8 Full specs →

What is the AMD Instinct MI350X?

The Instinct MI350X is a GPU from AMD on the CDNA 4 architecture, an accelerator for AI training, inference and high-performance computing. It is an OAM module, the open accelerator format used in multi-GPU servers. Instinct MI350X full specifications →

Can I rent the Instinct MI350X in the cloud?

Yes. Cloud providers tracked by Flopper rent the Instinct MI350X by the hour, and the pricing page compares their current rates and the price history side by side. Instinct MI350X rental prices →

How does the Instinct MI350X compare with other accelerators?

This page sets the Instinct MI350X against the H200. The Instinct MI350X is the newer generation of the two. This is a cross-vendor matchup, AMD against NVIDIA; the Instinct MI350X is an OAM module while the H200 is an SXM module. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the Instinct MI350X figures come from?

Every Instinct MI350X figure on this page traces to AMD's own datasheet or specification table. How Flopper verifies specifications →

NVIDIA H200 SXM 141GB

Full specs →

What is the NVIDIA H200?

The H200 is a GPU from NVIDIA on the Hopper architecture, an accelerator for AI training, inference and high-performance computing. It is an SXM module, the mezzanine format used in dense multi-GPU servers. H200 full specifications →

Can I rent the H200 in the cloud?

Yes. Cloud providers tracked by Flopper rent the H200 by the hour, and the pricing page compares their current rates and the price history side by side. H200 rental prices →

How does the H200 compare with other accelerators?

This page sets the H200 against the Instinct MI350X. The H200 is the older generation of the two. This is a cross-vendor matchup, NVIDIA against AMD; the H200 is an SXM module while the Instinct MI350X is an OAM module. Flopper also compares it head to head with the H100 SXM5 80GB, B300 SXM 262GB and A100 SXM4 80GB, across every published precision, memory and power. H200 vs H100 SXM5 80GB →

Where do the H200 figures come from?

Every H200 figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

Specifications

Instinct MI350X

H200

Architecture

CDNA 4Hopper

Launch Year

20252024

Form Factor

OAMSXM

Memory

288 GB141 GB

Memory Bandwidth

8.0 TB/s4.8 TB/s

TDP

1.0 kW700 W

Process Node

3nm/6nm4nm

Spec

Instinct MI350X

H200

Architecture CDNA 4Hopper

Launch Year 20252024

Form Factor OAMSXM

Memory 288 GB141 GB

Memory Bandwidth 8.0 TB/s4.8 TB/s

TDP 1.0 kW700 W

Process Node 3nm/6nm4nm

Performance (TFLOPS)

Instinct MI350X

H200

72 TFLOPS 34 TFLOPS

144 TFLOPS 67 TFLOPS

494 TFLOPS

989 TFLOPS sparse

2,310 TFLOPS

4,619 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

2,310 TFLOPS

4,619 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

4,614 TFLOPS

9,227 TFLOPS sparse

1,979 TFLOPS

3,958 TFLOPS sparse

FP6

9,228 TFLOPS

sparse not published

9,228 TFLOPS

sparse not published

4,614 TOPS

9,227 TOPS sparse

1,979 TOPS

3,958 TOPS sparse

Precision

Instinct MI350X

H200

72 TFLOPS 34 TFLOPS

144 TFLOPS 67 TFLOPS

494 TFLOPS

989 TFLOPS sparse

2,310 TFLOPS

4,619 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

2,310 TFLOPS

4,619 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

4,614 TFLOPS

9,227 TFLOPS sparse

1,979 TFLOPS

3,958 TFLOPS sparse

FP6 9,228 TFLOPS

sparse not published

9,228 TFLOPS

sparse not published

4,614 TOPS

9,227 TOPS sparse

1,979 TOPS

3,958 TOPS sparse

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

The MI350X wins the spec sheet; the H200 wins the market. AMD rates its CDNA 4 part at 288 GB of HBM3e and 8.0 TB/s against NVIDIA’s 141 GB and 4.8 TB/s, so a model split across two H200s can often sit on one MI350X. It also runs FP4 and FP6, formats Hopper lacks entirely.

Moving a serving fleet costs software, not silicon. A switch to ROCm means requalifying the inference stack, custom kernels and quantisation recipes, and the FP4 headroom only pays once your runtime supports it on CDNA 4. The H200 keeps CUDA, NVLink and MIG, so existing builds deploy unchanged.

Supply is the sharper gap. Flopper tracks one MI350X provider, DigitalOcean at $6.16 per GPU-hour, against 57 H200 listings from $2.00. Pilot on the MI350X pricing page before committing, and see the previous step in MI325X vs H200.

New to the terms? The glossary explains FP64, FP32, TF32, BF16, FP16, FP8, FP4, INT8, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

Is the MI350X faster than the H200?

On paper, yes. Comparing dense to dense, AMD rates the MI350X at 2,309.6 TFLOPS BF16 and 4,614 TFLOPS FP8, against NVIDIA's 990 TFLOPS BF16 and 1,979 TFLOPS FP8 for the H200. AMD also lists 8.0 TB/s of memory bandwidth to the H200's 4.8 TB/s, which matters most for LLM decode.

Does the H200 support FP4 or FP6?

No. Hopper bottoms out at FP8, so the H200 has no FP4 or FP6 path. The MI350X runs both, which AMD rates at 9,227.5 TFLOPS dense. Those gains only reach production once your inference runtime and quantised weights support the formats on ROCm.

How much more memory does the MI350X have than the H200?

The MI350X carries 288 GB of HBM3e against 141 GB on the H200, a little over twice as much. That headroom fits larger models or longer KV caches on one accelerator, so a deployment split across two H200s can often run on a single MI350X.

Can you rent an MI350X or an H200?

Flopper's listings show one provider for the MI350X: DigitalOcean, at $6.16 per GPU-hour on-demand in 1x and 8x configurations. The H200 is widely rented, with 57 listings across 31 providers from $2.00 to $10.60 per GPU-hour, the lowest on spot capacity.

Where to Rent

AMD Instinct MI350X OAM

Rent from

$6.16 / GPU-hr

Lowest of 2 live listings across 1 provider

See all Instinct MI350X rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

NVIDIA H200 SXM 141GB

Rent from

$0.79 / GPU-hr

Lowest of 54 live listings across 30 providers · median $4.38

See all H200 rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → Instinct MI350X full specs H200 full specs Estimate cluster cost in the calculator →

AMD Instinct MI350X OAM vs NVIDIA H200 SXM 141GB

AMD Instinct MI350X OAM

What is the AMD Instinct MI350X?

Can I rent the Instinct MI350X in the cloud?

How does the Instinct MI350X compare with other accelerators?

Where do the Instinct MI350X figures come from?

NVIDIA H200 SXM 141GB

What is the NVIDIA H200?

Can I rent the H200 in the cloud?

How does the H200 compare with other accelerators?

Where do the H200 figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

Is the MI350X faster than the H200?

Does the H200 support FP4 or FP6?

How much more memory does the MI350X have than the H200?

Can you rent an MI350X or an H200?

Where to Rent

AMD Instinct MI350X OAM

NVIDIA H200 SXM 141GB

Related comparisons

Get Comparison Updates

Stay Updated on GPU Specs