NVIDIA192 GB VRAM8,000 GB/s$6/hr

DeepSeek V3 671B (MoE) on B200 SXM 192GB: Does It Fit, How Fast, What It Costs

DeepSeek V3 671B (MoE) at AWQ 4-bit needs 369.1 GB for its weights and 631.0 GB in total at 8K context with 8 concurrent users. That means it needs 4× B200 SXM 192GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 4×
631.0 GB needed / 768 GB available
Decode speed
764.5
tok/s per stream
Aggregate
36698
tok/s @ 8 concurrent
Cost / 1M tokens
$0.1817
4× B200 SXM 192GB
What changes as you add B200 SXM 192GBs
Tensor-parallel degree vs VRAM, throughput, and cost. Cost scales linearly with GPU count; decode throughput scales near-linearly too, so cost per token stays roughly flat.
GPUsTotal VRAMVRAM neededDecode tok/sAggregate tok/sTTFTCost / 1M tokFits?
1× B200192 GB631.0 GB0.0000.000——✗
2× B200384 GB631.0 GB0.0000.000——✗
4× B200768 GB631.0 GB764.536698233.4 ms$0.1817✓
8× B2001536 GB631.0 GB1529146792131.7 ms$0.0908✓
Concurrency vs context length
On 4× B200 SXM 192GB at AWQ 4-bit. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K2416.37 GB✓ fits
8K1232.75 GB✓ fits
33K3131.00 GB✓ fits
66K1261.99 GB✓ fits
131K0523.99 GB✗ does not fit
Other GPUs that run DeepSeek V3 671B (MoE)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
A100 80GB SXM48×36698$0.0727
H100 SXM5 80GB8×61469$0.0904
H200 SXM5 141GB8×88075$0.1009
Frequently asked questions

Can DeepSeek V3 671B (MoE) run on B200 SXM 192GB?

Yes, but it needs 4× B200 SXM 192GB. At AWQ 4-bit the total footprint is 631.0 GB (369.1 GB of weights plus 262.0 GB of KV cache at 8K × 8 concurrent users), against 768 GB of available VRAM across 4 GPUs.

How fast is DeepSeek V3 671B (MoE) on B200 SXM 192GB?

764.5 tokens/sec per stream and 36698 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 233.4 ms, and inter-token latency is 1.3 ms.

What does DeepSeek V3 671B (MoE) on B200 SXM 192GB cost per million tokens?

$0.1817 per 1M output tokens at 4× B200 SXM 192GB ($6.00/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does DeepSeek V3 671B (MoE) need on B200 SXM 192GB?

369.1 GB for the weights at AWQ 4-bit, plus 262.0 GB of KV cache for 8 concurrent requests at 8K context — 631.0 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is B200 SXM 192GB the right choice for DeepSeek V3 671B (MoE)?

It works. Compare $0.1817 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× A100 80GB SXM4 at $0.0727, and the fastest is 8× H200 SXM5 141GB at 88075 tok/s.

Adjust the workload yourself

Same formulas, your context length, concurrency, and quantisation.

Open in calculator