MiniMax M3.1 Flash: Self-Hosting Estimate
MiniMax M3.1 Flash (MiniMax) on different GPUs: configuration, estimated performance and use cases.
Data updated: 2026-09-16B300 288GB
288 GB · 8,000 GB/s · Top-tier Blackwell card (no official channels to China)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 2priced as 1 × 8-GPU HGX node
- Est. decode speed
- 348 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 591 tok/sin 12,717 tok/s · many concurrent requests
- Hardware list price
- $497,650Cloud rental (live) ×2: $11.80–$19.88/hr (2026-10-23)
- Hardware market price
- $1,047,669
- Typical use
- Enterprise-grade ultra-large-scale training and high-concurrency inference clusters
GB200 192GB (single card equivalent)
192 GB · 8,000 GB/s · Top-tier Blackwell card (no official channels to China)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 4
- Est. decode speed
- 696 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 1,183 tok/sin 28,261 tok/s · many concurrent requests
- Hardware list price
- $119,829–$179,744
- Hardware market price
- $359,201–$718,402
- Typical use
- Ultra-large-scale trillion-parameter model training and inference clusters (networked with Grace CPU)
B200 180GB
180 GB · 8,000 GB/s · Top-tier Blackwell card (no official channels to China)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 4priced as 1 × 8-GPU HGX node
- Est. decode speed
- 696 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 1,183 tok/sin 25,435 tok/s · many concurrent requests
- Hardware list price
- $240,679–$563,744Cloud rental (live) ×4: $25.01–$34.44/hr (2026-10-23)
- Hardware market price
- $449,001–$523,834
- Typical use
- Enterprise-grade large-scale online inference and training
H200 141GB
141 GB · 4,800 GB/s · Top-tier export-controlled card (channels to China are volatile)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 4priced as 1 × 8-GPU HGX node
- Est. decode speed
- 417 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 710 tok/sin 11,180 tok/s · many concurrent requests
- Hardware list price
- $248,505–$450,916Cloud rental (live) ×4: $18.43–$18.43/hr (2026-10-23)
- Hardware market price
- $224,500–$233,481
- Typical use
- Enterprise-grade production environment high-concurrency inference
H100 SXM5 80GB
80 GB · 3,350 GB/s · Top-tier export-controlled card (no official channels to China)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8priced as 1 × 8-GPU HGX node
- Est. decode speed
- 583 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 990 tok/sin 22,360 tok/s · many concurrent requests
- Hardware list price
- $216,440–$833,693Cloud rental (live) ×8: $19.21–$49.41/hr (2026-10-23)
- Hardware market price
- $404,101–$419,068
- Typical use
- Enterprise-grade production environment high-concurrency inference/training
H800 SXM 80GB
80 GB · 3,350 GB/s · Top-tier export-controlled card (discontinued)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8priced as 1 × 8-GPU HGX node
- Est. decode speed
- 583 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 990 tok/sin 22,360 tok/s · many concurrent requests
- Hardware list price
- $293,347–$311,307
- Hardware market price
- $404,101–$419,068
- Typical use
- Enterprise-grade production deployment (original China-compliant model, now discontinued)
H20 96GB
96 GB · 4,000 GB/s · Export-compliant special supply card (significantly reduced compute power)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 696 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 1,183 tok/sin 3,346 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- $95,787–$131,707
- Typical use
- Suboptimal choice in export-compliant environments, medium-to-large scale inference deployment (FP16 compute power is only about 15% of H100, but higher memory bandwidth offers an advantage in decode-intensive scenarios)
A800 PCIe 80GB
80 GB · 1,935 GB/s · Compliant special supply card (export compliance status adjusted multiple times)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 337 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 343 tok/sin 4,232 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $119,734–$124,523
- Hardware market price
- $173,614
- Typical use
- Medium-scale inference/training deployment in export-compliant environments
Ascend 910C 128GB
128 GB · 3,200 GB/s · Huawei dual-chip data center card (domestically produced)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 557 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 946 tok/sin 17,002 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- n/a
- Typical use
- Domestically produced large model training and inference clusters, CloudMatrix and other super-node solutions
Ascend 910B 64GB
64 GB · 400 GB/s · Domestic alternative (not subject to US export controls)- Suggested precision
- INT8
- GPUs needed
- 8
- Est. decode speed
- 139 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 237 tok/sin 7,235 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- $131,707–$143,680
- Typical use
- Domestic alternative, not subject to US export controls, medium-to-large scale inference/training deployment, memory bandwidth significantly lower than NVIDIA cards of similar VRAM capacity
RTX PRO 6000 Blackwell 96GB
96 GB · 1,792 GB/s · Professional workstation/server card (China-specific version is 6000D 84GB)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 312 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 318 tok/sin 6,783 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $128,261Cloud rental (live) ×8: $10.26–$20.02/hr (2026-10-23)
- Hardware market price
- n/a
- Typical use
- Single-card large VRAM inference, local deployment of medium MoE models, private all-in-one machine
RTX PRO 5500 Blackwell 84GB
84 GB · 1,398 GB/s · Professional workstation card (specs are NVIDIA preliminary data)- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 243 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 248 tok/sin 6,131 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- n/a
- Hardware market price
- $101,774–$113,747
- Typical use
- Workstation large VRAM inference, departmental private deployment
RTX PRO 5000 Blackwell 48GB
48 GB · 1,344 GB/s · Professional workstation card- Suggested precision
- INT8
- GPUs needed
- 8
- Est. decode speed
- 467 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 477 tok/sin 3,500 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- n/aCloud rental (live) ×8: $5.35–$9.62/hr (2026-10-23)
- Hardware market price
- $64,656
- Typical use
- Small to medium model inference, development testing, single-machine multi-card private deployment
L40S 48GB
48 GB · 864 GB/s · Compliant inference card- Suggested precision
- INT8
- GPUs needed
- 8
- Est. decode speed
- 301 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 307 tok/sin 4,911 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $53,880–$77,827Cloud rental (live) ×8: $3.75–$6.47/hr (2026-10-23)
- Hardware market price
- $53,880–$77,827
- Typical use
- SME teams building inference services, medium-low concurrency
L20 48GB
48 GB · 864 GB/s · Compliant inference card- Suggested precision
- INT8
- GPUs needed
- 8
- Est. decode speed
- 301 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 307 tok/sin 1,621 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $27,954–$28,115
- Hardware market price
- $29,933–$41,907
- Typical use
- SME teams building inference services, medium-low concurrency
RTX 5090 32GB
32 GB · 1,792 GB/s · Consumer flagship (not sold in China, 5090D for domestic market)- Suggested precision
- INT4 (AWQ/GPTQ)
- GPUs needed
- 8
- Est. decode speed
- 1,247 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 1,272 tok/sin 5,684 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- n/aCloud rental (live) ×8: $4.07–$6.75/hr (2026-10-23)
- Hardware market price
- $41,907–$64,656
- Typical use
- Local inference and development for individuals/small teams, multi-card VRAM pooling for quantized MoE
RTX 5090D 24GB
24 GB · 1,344 GB/s · Consumer-grade/small-scale self-built- Suggested precision
- INT4 (AWQ/GPTQ)
- GPUs needed
- 8
- Est. decode speed
- 935 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 954 tok/sin 4,029 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $19,755Cloud rental (live) ×8: $4.18–$6.42/hr (2026-10-23)
- Hardware market price
- $33,525–$47,893
- Typical use
- Personal development/small-scale verification/POC, not recommended for production-level high concurrency
RTX 4090D 24GB
24 GB · 1,008 GB/s · Consumer-grade/small-scale self-built- Suggested precision
- INT4 (AWQ/GPTQ)
- GPUs needed
- 8
- Est. decode speed
- 701 tok/ssingle-stream theoretical ceiling · vendor API 92 tok/s
- Est. batched throughput
- out 715 tok/sin 3,988 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $15,564–$22,748Cloud rental (live) ×8: $3.29–$4.28/hr (2026-10-23)
- Hardware market price
- $31,131–$35,920
- Typical use
- Personal development/small-scale verification/POC, not recommended for production-level high concurrency
FAQ
FAQ
How is the estimated decode speed calculated?
Estimated speed = (memory bandwidth × GPU count) ÷ (active parameters × bytes per parameter): the memory-bound upper limit for one request (roofline). It scales linearly with GPU count and ignores interconnect overhead, batching and framework losses, so real results are usually well below it.
Can I trust the speed numbers for consumer GPUs?
The linear assumption is reasonable for NVLink data-center cards (H/B series) but too optimistic for consumer PCIe cards without NVLink (RTX 5090D/4090D); multi-GPU consumer setups lose far more to communication, so equal-looking numbers do not mean equal performance.
How is “Suggested precision” chosen?
It is the highest precision that fits within 8 GPUs; beyond that it is flagged as multi-node.
Where do the prices come from?
USD converted at ¥6.68 per USD. Cloud rental is the international vast.ai market rate, not a China purchase price.