buyaigear
Prices checked 11 min ago

AMD · Vega 20 · 32 GB · Qwen3.8 27B · Q4_K_M: ~31 tok/s

AMD Radeon Instinct MI50 32GB

32 GB of HBM2 at a terabyte per second. The 32 GB version has become scarce and sells for several times the 16 GB card; passive, 300 W and retired from official ROCm support, so expect some driver archaeology, but a 32B model at Q4 runs.

Lowest price right now

$600

eBay · free shipping · $18 per GB · observed 2026-10-15 11:51 UTC

Buy the cheapest on eBay, $600

free shipping · Used · detsin 18 · 100.0%

All 3 listings on eBay

Where to buy

VendorPriceConditionObserved
eBay
cheapest active listing		 $600
free shipping		 new and used 2026-10-15 11:51 UTC Go to eBay

On eBay now

3 listings passed our filters at the last check. Cheapest delivered first. Seller reputation is feedback count and positive rate.

What it runs

For each model, grouped by vendor with the newest first: the largest quant that fits with 8K of context and the estimated speed for one user. The model you picked on the ladder is highlighted. Estimates from the spec sheet, not measurements: see how we estimate.

ModelQuantNeeds at 8KFitMax contextDecode tok/sPrefill tok/s
DeepSeek
DeepSeek V4.1 Flash 552B-A16B Sep 2026Q4_K_M338 GBdoes not fit
DeepSeek V4 Flash 284B-A13B Jul 2026Q4_K_M175 GBdoes not fit
DeepSeek V3.1 671B-A37B Aug 2025Q4_K_M411 GBdoes not fit
Gemma
Gemma 4 12B May 2026Q8_014 GBfits128K+~40~280
Gemma 4 31B Mar 2026Q6_K27 GBfits63K~20~110
Gemma 4 26B-A4B Mar 2026Q8_028 GBfits128K+~130~870
Gemma 3 27B Mar 2025Q8_031 GBfits17K~18~120
GLM
GLM-5.3 753B-A40B Aug 2026Q4_K_M462 GBdoes not fit
GLM-5.3 Flash 320B-A18B Aug 2026Q4_K_M197 GBdoes not fit
GLM-4.6 355B-A32B Sep 2025Q4_K_M221 GBdoes not fit
GLM-4.5 Air 106B-A12B Jul 2025Q4_K_M68 GBdoes not fit
gpt-oss
gpt-oss-120b Aug 2025Q4_K_M73 GBdoes not fit
gpt-oss-20b Aug 2025Q8_024 GBfits128K+~130~920
Llama
Llama 4 Scout 109B-A17B Apr 2025Q4_K_M70 GBdoes not fit
Llama 3.3 70B Dec 2024IQ2_M28 GBfits19K~21~47
Llama 3.1 8B Jul 2024Q8_011 GBfits128K+~60~410
MiMo
MiMo V2.6 Pro 1.02T-A42B Sep 2026Q4_K_M624 GBdoes not fit
MiMo V2.6 Flash 309B-A15B Sep 2026Q4_K_M190 GBdoes not fit
Mistral
Mistral Medium 3.5 128B Mar 2026Q4_K_M83 GBdoes not fit
Mistral Small 4 119B-A6.5B Mar 2026Q4_K_M79 GBdoes not fit
Mistral Small 3.2 24B Jun 2025Q8_028 GBfits30K~20~140
Qwen
Qwen3.8 Flash-Next 125B-A6B Aug 2026Q4_K_M78 GBdoes not fit
Qwen3.8 27B Aug 2026 (your pick)Q8_031 GBfits28K~18~120
Qwen3.6 27B Apr 2026Q8_031 GBfits28K~18~120
Qwen3.6 35B-A3B Apr 2026Q6_K30 GBfits85K~210~1,100
Qwen3 Coder 480B-A35B Jul 2025Q4_K_M296 GBdoes not fit
Qwen3 235B-A22B Jul 2025Q4_K_M146 GBdoes not fit
Qwen3 32B Apr 2025Q6_K31 GBfits13K~19~100

Specifications

Memory
32 GB HBM2
Memory bandwidth
1,024 GB/s
Bus width
4096-bit
Shader cores
3,840
Tensor cores
0
FP32
13.3 TFLOPS
FP16 (dense)
26 TFLOPS
Power
300 W
Form factor
Dual slot, passive, needs a fan
Interface
PCIe 4.0 x16
Architecture
Vega 20
Launched
2018

Source: manufacturer specifications.