Granite 4.1 8B

IBM · ibm/granite-4-1-8b

ProviderIBM
Familygranite
Typellm-chat
Statusactive
Released2026-04-06
Updated2026-05-04
Parameters8.8B
Open weightsyes

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sbf1617.58 GB~836026.0~3344103.9 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1617.58 GB~637.0~2547.9 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1617.58 GB~318.5~1273.9 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1617.58 GB~306.5~1226.2 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1617.58 GB~293.8~1175.2 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1617.58 GB~238.9~955.5 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1617.58 GB~211.0~844.0 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1617.58 GB~191.1~764.4 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sbf1617.58 GB~155.3~621.0 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sbf1617.58 GB~133.4~533.5 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1617.58 GB~127.4~509.6 q4
Google TPU v5p95.0 GB2765.0 GB/sbf1617.58 GB~110.1~440.3 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sbf1617.58 GB~81.2~324.7 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sbf1617.58 GB~79.6~318.5 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sbf1617.58 GB~71.3~285.4 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sbf1617.58 GB~65.2~260.8 q4
AMD Instinct MI21064.0 GB1600.0 GB/sbf1617.58 GB~63.7~254.8 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sbf1617.58 GB~61.9~247.6 q4
Google TPU v432.0 GB1200.0 GB/sbf1617.58 GB~47.8~191.1 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sbf1617.58 GB~40.1~160.5 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sbf1617.58 GB~38.2~152.9 q4
NVIDIA GeForce RTX 508016.0 GB960.0 GB/sfp88.79 GB~76.4~152.9 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sbf1617.58 GB~37.3~149.1 q4
NVIDIA L40S48.0 GB864.0 GB/sbf1617.58 GB~34.4~137.6 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1617.58 GB~32.6~130.4 q4
AMD Radeon RX 7900 XT20.0 GB800.0 GB/sfp88.79 GB~63.7~127.4 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1617.58 GB~31.8~127.4 q4
Google TPU v5e16.0 GB800.0 GB/sfp88.79 GB~63.7~127.4 q4
NVIDIA GeForce RTX 4080 SUPER16.0 GB736.0 GB/sfp88.79 GB~58.6~117.2 q4
NVIDIA GeForce RTX 4070 Ti SUPER16.0 GB672.0 GB/sfp88.79 GB~53.5~107.0 q4
AMD Radeon RX 9070 XT16.0 GB640.0 GB/sfp88.79 GB~51.0~101.9 q4
Apple M4 Max128.0 GB546.0 GB/sbf1617.58 GB~21.7~86.9 q4
Apple M1 Max64.0 GB400.0 GB/sbf1617.58 GB~15.9~63.7 q4
Apple M2 Max96.0 GB400.0 GB/sbf1617.58 GB~15.9~63.7 q4
Apple M3 Max128.0 GB400.0 GB/sbf1617.58 GB~15.9~63.7 q4
NVIDIA GeForce RTX 3060 12GB12.0 GB360.0 GB/sfp88.79 GB~28.7~57.3 q4
NVIDIA L424.0 GB300.0 GB/sbf1617.58 GB~11.9~47.8 q4
NVIDIA GeForce RTX 4060 Ti 16GB16.0 GB288.0 GB/sfp88.79 GB~22.9~45.9 q4
Apple M4 Pro64.0 GB273.0 GB/sbf1617.58 GB~10.9~43.5 q4
Apple M432.0 GB120.0 GB/sbf1617.58 GB~4.8~19.1 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging Faceibm-granite/granite-4.1-8b

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.
BenchmarkEvaluated asScoreEvidence dateSource kind
aa_lcrGranite 4.1 8B13.33%2026-09-10 evaluatedindependent evaluatorsource
critptGranite 4.1 8B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondGranite 4.1 8B43.33%2026-09-10 evaluatedindependent evaluatorsource

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub