Cerebraswse2024

Cerebras CS-3 (WSE-3): Specs, VRAM, Throughput & Cost

Cerebras CS-3 (WSE-3) has 44 GB of VRAM and 20,000 GB/s of memory bandwidth. It runs 22 models from the catalog on a single card — up to Qwen 3 30B-A3B (MoE) — and serves Llama 3.3 70B at 45076 tok/s aggregate. All figures are computed by tokcalc's open-source capacity model.

VRAM
44 GB
Memory bandwidth
20,000
GB/s
FP16/BF16
1250 TF
Price
—
quote required
Hardware specifications
VendorCerebras
Categorywse
VRAM44 GB
Memory bandwidth20,000 GB/s
FP16/BF16 dense compute1250 TFLOPS
NVLink bandwidthnone
Release year2024
Typical cloud pricequote required
NoteWafer-scale; not comparable to per-GPU specs. Use cited benchmarks.
Models that fit on a single Cerebras CS-3 (WSE-3)
At each model's recommended quantization, 8K context and 8 concurrent users. Cost is per 1M output tokens.

Cerebras CS-3 (WSE-3) is a wafer-scale system, so “fits on one card” is not a meaningful comparison — the whole system is the unit of deployment. Use the cited benchmarks for this platform rather than the per-GPU numbers on this page.

Llama 3.3 70B across every accelerator
Same model (fp8), same workload, every GPU in the catalog. Cerebras CS-3 (WSE-3) ranks #1 of 29 by aggregate throughput.
GPUAggregate tok/sCost / 1M tok$/hr
Cerebras CS-3 (WSE-3)45076——
RTX 4080 16GB6464$0.1031$0.3
RTX A4000 16GB4039$0.1651$0.3
RTX 5090 32GB4039$0.1376$0.5
A100 40GB SXM43493$0.2863$0.9
RTX 4090 24GB2272$0.1956$0.4
V100 SXM2 32GB2028$0.2739$0.5
RTX 3090 24GB2017$0.1377$0.25
H100 SXM5 80GB1888$0.7358$2.5
TPU v6e Trillium (32GB)1846$0.4816$0.8
T4 16GB1836$0.2421$0.2
A10G 24GB1352$0.2958$0.36
Frequently asked questions

What is Cerebras CS-3 (WSE-3) used for?

Cerebras CS-3 (WSE-3) is a wse accelerator from Cerebras with 44 GB of VRAM and 20,000 GB/s of memory bandwidth. Wafer-scale; not comparable to per-GPU specs. Use cited benchmarks.. It is aimed at specialised inference workloads.

Which models fit on a single Cerebras CS-3 (WSE-3)?

22 models in the tokcalc catalog fit on one Cerebras CS-3 (WSE-3), up to Qwen 3 30B-A3B (MoE) (30B params) at fp8. Below that, throughput and cost per million tokens improve as the model gets smaller.

How fast is Llama 3.3 70B on Cerebras CS-3 (WSE-3)?

At fp8 with 8 concurrent users, 4× Cerebras CS-3 (WSE-3) serves 45076 tokens/sec aggregate (939.1 tok/s per stream), with a 494 ms time to first token. That works out to — per 1M output tokens.

How does Cerebras CS-3 (WSE-3) compare to other accelerators?

Memory bandwidth drives decode speed and VRAM drives capacity. Cerebras CS-3 (WSE-3) has 20,000 GB/s of bandwidth and 44 GB of VRAM, plus 1250 TFLOPS of dense FP16/BF16 compute. The comparison table on this page ranks every accelerator in the catalog on the same reference workload so you can see where it lands.

Size a real deployment on Cerebras CS-3 (WSE-3)

Pick your model, context length, and concurrency — the calculator computes the rest with the same formulas.

Open in calculator