Cerebras WSE-3
WAFER-SCALE In production Released 2024 cerebras-wse-gen3
BF16
TFLOP/s
62500 厂商声称
FP8
TFLOP/s
125000 厂商声称
FP4
TFLOP/s
unsupported
Memory
GB
44 厂商声称
Mem BW
GB/s
21000000 厂商声称
TDP
W
23000 厂商声称
Full specs
Compute
FP4 TFLOPS
unsupported
FP8 TFLOPS
125000
BF16 TFLOPS
62500
FP16 TFLOPS
62500
INT8 TOPS
125000
Memory
Capacity
44 GB
Bandwidth
21000000 GB/s
Type
on-die-sram
Die architecture 🟢 vendor floorplan
Tile count
900000
Process
5 nm
Die area
46225 mm²
Transistors
4000 B
Scale-Up (intra-node)
Protocol
SwarmX
Per-link BW
16000 GB/s
World size
16
Topology
full-mesh
Switch
—
Scale-Out (inter-node)
Per-card NIC
1200 Gbps
Protocol
SwarmX-IB
NIC
—
Topology
拓扑结构 · Topology
16 卡 scale-up domain
芯片内部 / Die-level architecture
🟢 vendor floorplan 900,000 Tiles · 44,000 MB on-die SRAM · 5 nm · 4000 B transistors · 46225 mm²
🌐 wafer-scale (no host-device transfer) 📦 on-die SRAM only (capacity ≪ HBM, bandwidth ≫ HBM)
集群拓扑 / Cluster topology · SwarmX @ 16000 GB/s
Scale-Up · 域内
SwarmX
16000 GB/s · 拓扑: full-mesh
world_size = 16
Scale-Out · 跨域
SwarmX-IB
1200 Gbps/卡 NIC
Which models can it run?
Quick estimates · decode tok/s/card 上界
TP=8 · FP8 · batch=16 · prefill=1024 · decode=256 · 已应用 efficiency 校准
| 模型 | 参数 (active) | Decode tok/s/card | 瓶颈 |
|---|---|---|---|
| DeepSeek V4 Pro deepseek | 49B | — | 显存不足 |
| DeepSeek V4 Flash deepseek | 13B | 597,166 | 内存带宽 |
| Mistral Small 4 mistral | 22B | 272,102 | 内存带宽 |
| GLM-5 Reasoning zhipu | 32B | 225,221 | 内存带宽 |
| GLM-5.1 zhipu | 32B | — | 显存不足 |
| Qwen3.6 Plus alibaba | 35B | — | 显存不足 |
| Kimi K2.6 moonshot | 32B | — | 显存不足 |
| MiniMax M2.7 minimax | 46B | — | 显存不足 |
Operator-level fit · per-model bottleneck + upper bound
算子级 fit · operator-level fit (per-token roofline)
基于每个模型 operator_decomposition + 本卡 BF16 62,500 TFLOPS / 21,000,000 GB/s 计算 · ridge point ≈ 3 FLOPs/byte
| 模型 | domain | 主导算子 | AI · F/B | 瓶颈 | tok/s 上界 |
|---|---|---|---|---|---|
| GraphCast | scientific | graph-message-passing | 0.9 | 💾 内存带宽 | 38745k |
| AlphaFold 3 | scientific | pair-bias-attention | 2.3 | 💾 内存带宽 | 11641k |
| DeepSeek V4 Pro | llm | matmul | 245.5 | 🔥 计算 | 10391k |
| GPT-OSS | llm | matmul | 0.7 | 💾 内存带宽 | 1698k |
| Gemma 4 26B | llm | matmul | 0.7 | 💾 内存带宽 | 1262k |
| DeepSeek V4 Flash | llm | matmul | 0.8 | 💾 内存带宽 | 1194k |
| Mistral Small 4 | llm | matmul | 0.6 | 💾 内存带宽 | 544k |
| Llama 4 Maverick | llm | matmul | 0.8 | 💾 内存带宽 | 538k |
需要 efficiency 校准 + concurrency 扫描 + TCO 估算 → 在计算器中评估 →
Operator support & optimization headroom
算子支持 & 优化空间 / Operator support & headroom
Per-operator support derived from software_support.engines + scale-up topology. Optimization headroom from measured efficiency factor.
Optimization headroom
+50 pp
moderate
No cases yet — using default 0.5 efficiency. Real headroom unknown until first measurement lands.
Communication (collective)
All-to-All 🟢 mature
all-to-all via SwarmX world_size=16
AllReduce 🟢 mature
SwarmX ring all-reduce
Attention
Multi-Head Attention 🟢 mature
paged-attention via vLLM/SGLang/MindIE
FlashAttention-3 🟢 mature
FA-3 on modern engine + tensor cores
Matrix multiply (GEMM)
Matrix Multiplication 🟢 mature
GEMM supported on all inference engines
MoE routing
MoE Routing 🟢 mature
MoE gating supported via vLLM ≥0.4 / SGLang
Normalization
RMSNorm 🟢 mature
fused into engine kernels
Embedding
Rotary Position Embedding 🟢 mature
fused into engine kernels
Activation
SiLU / Swish 🟢 mature
fused into engine kernels
Softmax 🟢 mature
fused into engine kernels
最接近的替代卡 (按规格相似度)
基于 BF16 算力 / 显存 / 显存带宽 / FP8 加权欧氏距离。供选型决策参考。
Software-stack support
| Engine | Status | BF16 | FP16 | FP4 | FP8 E4M3 | FP8 E5M2 | INT4 AWQ |
|---|---|---|---|---|---|---|---|
| HanGuangAI | unconfirmed | — | — | — | — | — | — |
| LMDeploy | unconfirmed | — | — | — | — | — | — |
| MindIE | unconfirmed | — | — | — | — | — | — |
| MoRI | unconfirmed | — | — | — | — | — | — |
| SGLang | unconfirmed | — | — | — | — | — | — |
| TensorRT-LLM (Dynamo) | unconfirmed | — | — | — | — | — | — |
| vLLM | community | ✓ | ✓ | — | ✓ | — | — |
Existing deployment cases (0)
No measured cases yet for this card.
Be the first contributor?
Citations
- [1] Cerebras WSE-3 product page: 4 trillion transistors, 900,000 cores, 44 GB on-chip SRAM, 21 PB/s memory bandwidth, 5nm TSMC, 46,225 mm² die — https://www.cerebras.ai/product-chip/ · accessed 2026-04-29 厂商声称
- [2] FP64 estimate from CS-3 SC23 paper showing ~7.8 TFLOPS sustained on HPL-AI workloads (derived from system-level measurements) — https://www.cerebras.ai/blog/wse3-architecture · accessed 2026-04-29 社区估算
⚠ Wafer-scale architecture: throughput scales nonlinearly with model size due to no host-device transfer.
⚠ Memory bandwidth quoted is aggregate on-die SRAM bandwidth, not HBM-equivalent.
⚠ World-size for scale-up is at the CS-3 system level (1 system = 1 wafer); SwarmX clusters connect multiple systems.