8× V100 32GB SXM2 server: what it runs, what it costs

Eight Tesla V100 32GB SXM2 modules on one NVLink board give 256GB of VRAM, enough to hold today's large mixture-of-experts models whole. The servers that carry them (Inspur NF5288M5, Supermicro SYS-4029GP-TVRT, Gigabyte G481-S80) are coming off lease, and 1Cat-vLLM, a vLLM fork for V100s, runs current models on them.

As of October 3, 2026, one in ten of the 30 complete 8× V100 32GB SXM2 server listings on eBay priced at 50% or more of the median asks $7,127 or less shipped. The median ask is $11,490 across all 31. 1Cat-vLLM's developer reports DeepSeek-V4-Flash decoding at 65.1 tok/s (batch 1, all eight GPUs) and GLM-5.3-Flash at 53.0 tok/s.

What it runs: published benchmarks

Reported: published by the engine's developer; not reproduced by rackbites. rackbites hasn't measured this configuration yet. Every figure is the source's own, on its stated setup; a different engine version, driver or workload gives a different number.

Before you buy

• Buy on VRAM. This configuration is 32GB per card; eight 16GB cards are 128GB, half as much, and the models above aren't shown to fit. A listing that doesn't state the card size is a near miss below, not a match.
• Power: eight 300 W SXM2 modules draw up to 2.4 kW before the CPUs. Server power supplies in this class reach full output only on 200–240 V, so plan for that circuit; a 120 V household outlet can't feed one at full load.
• Noise: datacenter fans. It belongs in a rack, a garage or a basement, not an office.
• Drivers: NVIDIA's R580 branch is the last to support Volta, and CUDA 13 dropped it. Stay on CUDA 12.x.
• Software: every figure here comes from 1Cat-vLLM, a vLLM fork for V100s (SM70) whose recent work comes from one developer, who also ran these benchmarks. Pin the commit you test; each figure names its commit.

Complete 8× V100 32GB SXM2 server listings as of October 3, 2026

eBay affiliate disclosure: The listing links below lead to eBay. rackbites may earn a commission if you make a purchase.

Listings with the same configuration and price share one row, and the numbered links are the other copies. Each copy still counts toward the listing total, the low end and the median.

Not shown: 1 of the 31 listings ask under 50% of the median. Prices that far below the market are often scams or mislabeled listings. They count toward the listing total and the median, not the low end.

The second figure, “with 256GB RAM”, adds the missing memory at the low-end price of DDR4-2666 RDIMMs ($5.62/GB, from DIMMsum), so servers that ship with different amounts compare fairly. A dash means the listing doesn't say how much RAM ships.

Near misses: 8× Tesla V100 servers that differ on one detail

46 near misses, showing 12 rows: listings that leave a detail out come first, since they may still be the right machine, then listings that differ. Cheapest first within each. The reason comes from what the title states.

Listings with the same configuration and price share one row, and the numbered links are the other copies. Each copy still counts toward the listing total, the low end and the median.

Or build it from parts

SXM2 barebones usually include the GPU board, but not always the GPU heatsinks, and loose SXM2 modules are often sold without them. Confirm both before you count on this total.

The total leaves out drives and rails. Each line is that part's low-end ask with shipping, which one in ten of its listings match or beat.

A line needs 3 live listings before it gets a price. See all proven builds.