NVIDIA H200 vs NVIDIA H200 NVL: which is better for AI?

Pick the H200 for speed: 18% more FP16 throughput. Both have 141 GB, so the same models fit on each. On Runcrate today the H200 NVL is cheaper, from $4.63/hr.

What's the price difference between NVIDIA H200 and NVIDIA H200 NVL?

On Runcrate today, NVIDIA H200 starts at $5.13/hr and NVIDIA H200 NVL starts at $4.63/hr; NVIDIA H200 NVL is cheaper by 10%.

Can I deploy NVIDIA H200 or NVIDIA H200 NVL on Runcrate today?

NVIDIA H200 is on demand in 4 regions from $5.13/hr, billed by the second on most GPUs. NVIDIA H200 NVL is on demand in 2 regions from $4.63/hr, billed by the second on most GPUs.

How much VRAM does each GPU have?

NVIDIA H200 has 141GB HBM3e. NVIDIA H200 NVL has 141GB HBM3e.

H200 or H200 NVL: which holds a 70B model in 8-bit weights?

It needs 70 GB of 8-bit weights. On the H200: 1 GPU, 71 GB spare. On the H200 NVL: 1 GPU, 71 GB spare. Fits on one GPU of either.

GPU comparison

H200 vs H200 NVL.

Pick the H200 for speed: 18% more FP16 throughput. Both have 141 GB, so the same models fit on each.

Launch H200 NVL Reserve a cluster

Rent either on Runcrate

H200from $5.13/hr
$0.036 per GB/hr$0.52 per 100 TFLOPS/hr
Launch H200 H200 pricing
H200 NVLCheaperfrom $4.63/hr
$0.033 per GB/hr$0.55 per 100 TFLOPS/hr
Launch H200 NVL H200 NVL pricing

Live hourly price, then per GB of memory and per 100 FP16 TFLOPS.

Which is better for

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

JobH200H200 NVLPick

Hold a 70B model in 8-bit weights
70 GB of 8-bit weights
H200: 1 GPU, 71 GB spare
H200 NVL: 1 GPU, 71 GB spare
Either. Fits on one GPU of either.
Serve an 8B model in 16-bit weights
16 GB of 16-bit weights
H200: 1 GPU, 125 GB spare
H200 NVL: 1 GPU, 125 GB spare
Either. Fits on one GPU of either.
Fine-tune a 70B model with QLoRA
About 46 GB with 4-bit weights
H200: 1 GPU, 95 GB spare
H200 NVL: 1 GPU, 95 GB spare
H200. Same GPU count, 18% more FP16 throughput.
Train on a full node
FP16 tensor throughput per node
H200: 8x, 7,912 TFLOPS, NVLink
H200 NVL: 8x, 6,680 TFLOPS, NVLink in groups of 4
H200. 18% more FP16 throughput per GPU.
Throughput per watt
FP16 TFLOPS per 100 W of TDP
H200: 141 TFLOPS per 100 W
H200 NVL: 139 TFLOPS per 100 W
Either. Close per watt.

Compare another pair

Hopper

NVIDIA H200

from$5.13/hr

Available now · 4 regions

Memory: 141 GB HBM3e
FP16 tensor: 989 TFLOPS
Bandwidth: 4.8 TB/s

Launch H200 Reserve a cluster

Full pricing details

Hopper

NVIDIA H200 NVL

from$4.63/hrCheaper

Available now · 2 regions

Memory: 141 GB HBM3e
FP16 tensor: 835 TFLOPS
Bandwidth: 4.8 TB/s

Launch H200 NVL Reserve a cluster

Full pricing details

At a glance

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

Memory

NVIDIA H200141 GB

NVIDIA H200 NVL141 GB

Memory bandwidth

NVIDIA H2004.8 TB/s

NVIDIA H200 NVL4.8 TB/s

FP16 tensor

NVIDIA H200989 TFLOPS

NVIDIA H200 NVL835 TFLOPS

Hourly price, from (live)

NVIDIA H200$5.13/hr

NVIDIA H200 NVL$4.63/hr

Full specs

The complete spec sheet.

Spec	NVIDIA H200	NVIDIA H200 NVL
Architecture	Hopper	Hopper
Memory	141 GB	141 GB
Memory type	HBM3e	HBM3e
Memory bandwidth	4.8 TB/s	4.8 TB/s
FP32	67 TFLOPS	60 TFLOPS
FP16 tensor	989 TFLOPS	835 TFLOPS
INT8	1979 TOPS	1670 TOPS
TDP	700 W	600 W
Form factor	SXM	PCIe
NVLink	Yes	In groups of 4
GPUs per node	8	8
Hourly price (live)	from $5.13/hr	from $4.63/hr

Questions

H200 or H200 NVL. What buyers ask first.

Other comparisons