VRAM
44 GB
TDP
23000 W
MSRP
$2.50M
Perf/W
0.02 tok/s/W
Cost/1K tok
$0.0587/k
Tested
2025-03-29
Cerebras WSE-3 (CS-3) has a source-attributed result of 450.0 tok/s on Llama 3 70B Q4 (batch 1, 8192-token context, FP16; runtime not documented). This is a reference report, not an independently verified lab result. A 70B Q4_K_M artifact needs roughly 40GB for weights alone; smaller memory configurations require explicit offload or model splitting and do not establish full-GPU residency.
Cerebras WSE-3 — wafer-scale AI accelerator with 4 trillion transistors, 900,000 AI cores, 44GB on-chip SRAM at 21 PB/s bandwidth, 23kW system power. The largest single chip ever built for AI.
Memory bandwidth (21 PB/s) is orders of magnitude beyond any GPU — token generation at 1,420 tok/s on Gemma 2 9B demonstrates the architecture's strength. Optimized for sparse training and inference at unprecedented scale. Not a GPU replacement — requires the CS-3 system ($2M+) and is only accessible via Cerebras cloud. Best for: organizations with frontier-scale AI budgets needing maximum throughput on supported model architectures.
Cerebras WSE-3 (CS-3) with 44GB VRAM at 23000W TDP, scored across 4 workloads with 5 benchmark records.
Reference workload
Llama 3 8B FP16
1850 tok/s
Quantization
FP16
8K context · batch 1
| Workload | Score | Quant | Context | σ | Status | Tested |
|---|---|---|---|---|---|---|
| Llama 3 70B Q4 llm | 450.0tok/s | FP16 | 8K | , | Curated Aggregate | 2025-03-29 |
| Llama 3 8B FP16 llm | 1850.0tok/s | FP16 | 8K | , | Curated Aggregate | 2025-07-26 |
| Mistral 7B llm | 1620.0tok/s | FP16 | 8K | , | Curated Aggregate | 2024-09-12 |
| Gemma 2 9B llm | 1420.0tok/s | FP16 | 8K | , | Curated Aggregate | 2024-10-30 |
There is insufficient comparable evidence to score Cerebras WSE-3 (CS-3). A score needs results for multiple devices with matching workload, runtime version, quantization, context and batch size.
Planning estimates for 44GB: weights plus at least 2GB or 10% overhead. Actual KV cache depends on model, context and cache format; confirm the artifact before buying.
Trust score
6/10
MyAI rating
Not scored
Runs
Not documented
Freshness
Stale
Source-linked row with explicit verification status.
Record date: 2025-07-26; 410 days old.
Enterprise buying note
Retailer we'd check first
Amazon search plus reseller quotes
For datacenter and accelerator parts, Amazon is useful for spotting live listings, accessories, or used pulls, but serious procurement usually happens through integrators, brokers, or cloud partners.
Enterprise procurement, not retail
This silicon is typically acquired through an authorized OEM partner, system integrator, or hyperscaler reseller. For on-demand access, compare hourly rates at RunPod, Vast.ai, Lambda Labs, or your existing cloud provider before committing to capital expenditure.
There is insufficient comparable evidence for a purchase recommendation. Check model fit, software support and a current seller quote.
Affiliate note: we do not have a device-level ASIN yet, so this opens tagged Amazon search results for the exact product name.