GPUs tracked29Price history3 dayssince Sep 3, 2026Price points (24h)10,787 pointsBiggest 24h dropBiggest 24h riseData updated Sep 5, 2026

Instinct MI100: 32GB VRAM — What Local LLMs Can It Run?

Check prices
UsedView on eBayNewView on AmazonBuying links may carry affiliate tags; this never changes the price you pay

The Instinct MI100 has 32GB of HBM2 memory with 1,229 GB/s of bandwidth. No used-price data yet. At Q4 quantization it runs models up to 32B parameters at roughly 31 tokens/s.

Full Instinct MI100 specifications

Cores7,680
Tensor cores480
ArchitectureCDNA (Arcturus)
Memory32 GB · HBM2
Bus width4,096-bit
Bandwidth1,229 GB/s
BF16 / FP8 / INT892.3 / — / 92.3
EcosystemROCm
Also known asMI100 · AMD MI100 32GB
eBay (US used)No price data
TypeData center
Power300 W
Power connector2× 8-pin
Slots2
Length267 mm
InterfacePCIe 4.0 x16
Launch dateNov 16, 2020
MSRP

What LLMs can the Instinct MI100 run?

One reference model per family (its main version)
ModelQ4_K_MQ8_0FP16
FitsSpeed t/sContextFitsSpeed t/sContextFitsSpeed t/sContext
DeepSeekDeepSeek R1 Qwen3 8B107.4128K63.1128K35.232K
GemmaGemma 4 31B27.88K
GLMGLM 4.7 Flash140.5128KOffload22.1Offload12.2
gpt-ossgpt-oss 20B130.8128K95.1128KOffload9.2
KimiKimi Linear 48B A3B146.3128KOffload25.7Offload14.3
LlamaLlama 3.1 8B110.3128K64.7128K3632K
MistralMistral 7B v0.3119.932K70.832K39.732K
QwenQwen3 8B107.432K63.132K35.232K

Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.

Instinct MI100 used price history

Current Samples 0

Not enough history to draw a chart yet — price tracking just started.

Prices by platform

eBayUsed
No listings
AmazonNew
No listings

Buy links may carry affiliate tags; they do not change the price you pay.

FAQ

How much VRAM does the Instinct MI100 have?

The Instinct MI100 has 32GB of HBM2 memory with 1,229 GB/s of bandwidth.

Can the Instinct MI100 run a 70B model?

Not on a single card. At Q4_K_M with an 8K context a 70B dense model needs about 43GB, more than the Instinct MI100's 32GB.

How much does a used Instinct MI100 cost?

No used-price data yet; it will appear here once the feed is live.