Hunyuan A13B: Self-Hosting Estimate
Hunyuan A13B (Tencent) 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
- 1priced as 1 × 8-GPU HGX node
- Est. decode speed
- 308 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 523 tok/sin 11,250 tok/s · many concurrent requests
- Hardware list price
- $299,058–$498,429Cloud rental (live): $7.88–$7.88/hr (2026-11-04)
- Hardware market price
- $1,047,167
- 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
- 1
- Est. decode speed
- 308 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 523 tok/sin 12,500 tok/s · many concurrent requests
- Hardware list price
- $29,943–$44,915
- Hardware market price
- $89,757–$179,514
- 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
- 2priced as 1 × 8-GPU HGX node
- Est. decode speed
- 615 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 1,046 tok/sin 22,500 tok/s · many concurrent requests
- Hardware list price
- $299,058–$498,429Cloud rental (live) ×2: $14.90–$25.02/hr (2026-11-05)
- Hardware market price
- $448,786–$523,584
- 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
- 2priced as 1 × 8-GPU HGX node
- Est. decode speed
- 369 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 628 tok/sin 9,890 tok/s · many concurrent requests
- Hardware list price
- $279,837–$499,708Cloud rental (live) ×2: $5.98–$9.21/hr (2026-11-05)
- Hardware market price
- $224,393–$233,369
- 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
- 4priced as 1 × 8-GPU HGX node
- Est. decode speed
- 515 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 876 tok/sin 19,780 tok/s · many concurrent requests
- Hardware list price
- $239,860–$831,514Cloud rental (live) ×4: $7.74–$16.45/hr (2026-11-05)
- Hardware market price
- $320,732
- 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
- 4priced as 1 × 8-GPU HGX node
- Est. decode speed
- 515 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 876 tok/sin 19,780 tok/s · many concurrent requests
- Hardware list price
- $293,207–$311,158
- Hardware market price
- $403,907–$418,867
- 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
- 2
- Est. decode speed
- 308 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 523 tok/sin 1,480 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- $23,935–$32,911
- 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
- 4
- Est. decode speed
- 298 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 304 tok/sin 3,744 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $59,838–$62,232
- Hardware market price
- $64,625–$71,806
- 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
- 2
- Est. decode speed
- 246 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 418 tok/sin 7,520 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- $23,935–$59,838
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 4
- Est. decode speed
- 62 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 105 tok/sin 6,400 tok/s · many concurrent requests
- Hardware list price
- n/a
- Hardware market price
- $65,822–$71,806
- 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
- 2
- Est. decode speed
- 138 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 141 tok/sin 3,000 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $32,050Cloud rental (live) ×2: $2.41–$4.54/hr (2026-11-05)
- Hardware market price
- $35,173
- 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
- 4
- Est. decode speed
- 215 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 219 tok/sin 5,424 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $45,573
- Hardware market price
- $50,862–$53,854
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 4
- Est. decode speed
- 207 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 211 tok/sin 3,096 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- n/aCloud rental (live) ×4: $2.24–$4.28/hr (2026-11-05)
- Hardware market price
- $32,313
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 4
- Est. decode speed
- 133 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 136 tok/sin 4,344 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $26,927–$38,895Cloud rental (live) ×4: $2.40–$3.65/hr (2026-11-05)
- Hardware market price
- $26,927–$38,895
- Typical use
- SME teams building inference services, medium-low concurrency
L20 48GB
48 GB · 864 GB/s · Compliant inference card- Suggested precision
- FP16/BF16 (full precision)
- GPUs needed
- 4
- Est. decode speed
- 133 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 136 tok/sin 1,434 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $13,970–$14,051
- Hardware market price
- $20,943
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 551 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 562 tok/sin 10,056 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- n/aCloud rental (live) ×8: $3.54–$4.26/hr (2026-11-05)
- Hardware market price
- $53,853
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 414 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 422 tok/sin 7,128 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $19,745Cloud rental (live) ×8: $3.54–$4.24/hr (2026-11-05)
- Hardware market price
- $33,509–$47,871
- 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
- FP16/BF16 (full precision)
- GPUs needed
- 8
- Est. decode speed
- 310 tok/ssingle-stream theoretical ceiling
- Est. batched throughput
- out 316 tok/sin 7,056 tok/s · many concurrent requests · PCIe multi-GPU penalty
- Hardware list price
- $15,557–$22,737Cloud rental (live) ×8: $2.70–$3.49/hr (2026-11-05)
- Hardware market price
- $31,116–$35,903
- 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.