DEPLOYDatabase

Chip comparison

Groq LPU (Language Processing Unit) vs SambaNova SN40L

Side-by-side specs

Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.

FieldGroq LPU (Language Processing Unit)SambaNova SN40L
Process nodeGlobalFoundries 14nmTSMC N5
TDP375 W700 W
Memory0.23 GB SRAM (230 MB on-die)1500 GB DDR5 (three-tier: 64 MB SRAM + 64 GB HBM3 + 1.5 TB DDR5)
Memory bandwidth80,000 GB/s
FP16 (dense)188 TFLOPS638 TFLOPS
FP8 (dense)1,250 TFLOPS
INT8 (dense)750 TOPS
Form factorPCIe (GroqCard) / rack-scale GroqRackPCIe (SN40L DataScale system)
Announced2020-01-012023-09-19
Released2021-06-012024-01-01

Robots running Groq LPU (Language Processing Unit)

No robots publicly running it yet.

Robots running SambaNova SN40L

No robots publicly running it yet.

Data centers with Groq LPU (Language Processing Unit)

No data centers publicly running it yet.

Data centers with SambaNova SN40L

No data centers publicly running it yet.

Common questions

Groq LPU (Language Processing Unit) vs SambaNova SN40L: which is faster for training?

SambaNova SN40L has 6.65x the dense FP16/BF16 throughput of the other (Groq LPU (Language Processing Unit): 188 TFLOPS; SambaNova SN40L: 1,250 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

Groq LPU (Language Processing Unit) vs SambaNova SN40L: which has more memory?

SambaNova SN40L carries more HBM (Groq LPU (Language Processing Unit): 0.23 GB SRAM (230 MB on-die); SambaNova SN40L: 1500 GB DDR5 (three-tier: 64 MB SRAM + 64 GB HBM3 + 1.5 TB DDR5)). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Groq LPU (Language Processing Unit) vs SambaNova SN40L: which is more power-efficient?

Dense FP16/BF16 TFLOPS per watt: Groq LPU (Language Processing Unit) 0.50 (188 ÷ 375 W); SambaNova SN40L 1.79 (1,250 ÷ 700 W). SambaNova SN40L wins at the die level; system-level efficiency also depends on cooling and interconnect.

See also: every chip comparison · Groq LPU (Language Processing Unit) full page · SambaNova SN40L full page.