Kentino AI Node V100 — 8× NVIDIA V100 SXM2, ex-datacentre
Enterprise-grade platform, assembled and tested in the EU.
Kentino AI Node V100 - refurbished
NVLink Multi-GPU on a Grant Budget
8x V100 SXM2 | 256 GB HBM2 | 1,000 TFLOPS fp16
Ex-datacentre Volta, refurbished in-house. Availability is whatever the market has.
An eight-GPU Volta node sourced from the ex-datacentre market and refurbished by us: eight V100 SXM2 modules on an NVLink baseboard, 256 GB of HBM2 at 900 GB/s per GPU, 300 GB/s GPU-to-GPU. This is deliberately a budget machine. It is the cheapest way we know to get real NVLink-connected multi-GPU with 7.8 TFLOPS of FP64 per GPU into a lab, and for FP64 HPC and validated fp16 training codes it is still a perfectly good tool. For modern LLM serving it is not - see below, we are specific about why.
Hardware
| Component | Detail |
|---|---|
| GPUs | 8x NVIDIA V100 SXM2 32 GB HBM2 (Volta, 300 W per module) |
| VRAM pool | 256 GB HBM2. 16 GB module variants give 128 GB - we will tell you which we have |
| Memory bandwidth | 900 GB/s per GPU |
| GPU interconnect | NVLink 2 hybrid-cube mesh, 300 GB/s. No NVSwitch - not full any-to-any |
| Compute | 1,000 TFLOPS fp16 aggregate | 62.4 TFLOPS FP64 aggregate |
| CPUs | 2x Intel Xeon Scalable, dual socket |
| System memory | 1 TB DDR4-2666 ECC RDIMM (16x 64 GB). Other capacities on request |
| Storage | 2x 1.92 TB U.2 NVMe boot + 2x 3.84 TB U.2 NVMe data |
| Chassis | 4U rack-mount, datacentre-grade front-to-back airflow |
| Refurbishment | Fresh thermal pads and paste, fan replacement, 72 h burn-in with written report |
What it runs
Volta is fp16 and FP64 hardware. These are the workloads it is still genuinely good at.
| Component | Detail |
|---|---|
| Classic HPC | CUDA / OpenACC, molecular dynamics, CFD, FEA at 62.4 TFLOPS FP64 per node |
| fp16 training, validated codes | ResNet/BERT-era architectures and anything already tuned for Volta |
| 13B-34B inference | fp16. ~15-30 tok/s single stream per model instance |
| Teaching and research clusters | real multi-GPU NVLink behaviour for students on a fixed grant |
| V100 estate expansion | node parity with existing Volta hardware you already operate |
Throughput figures are estimates derived from published vendor specifications and community benchmarks for this GPU generation, not measurements taken on this exact configuration. We will give you measured numbers for your model and context length before you commit.
Power and cooling
Roughly 3.2 kW sustained: 2.4 kW of GPU plus host. Lowest wall draw of our node range in absolute terms, but by far the worst performance per watt - around an eighth of an H100 node's fp16 throughput for a bit over half the power. Volta SXM2 was built for datacentre-grade front-to-back airflow and is unforgiving in a poorly ventilated rack.
Be clear about what this is
No bf16, no fp8, no INT8 tensor cores. Volta has none of them. Most 2026 quantised inference stacks assume bf16 or better, so a large part of the current ecosystem either will not run or will run badly. Treat this as fp16 training and FP64 HPC hardware, not an LLM serving node.
No NVSwitch. NVLink 2 is a hybrid-cube mesh, not full any-to-any. Eight-way tensor parallelism is measurably worse here than on the A100, H100 and H200 nodes.
Availability is not guaranteed. There is no new-channel supply for this hardware. Every unit is individually sourced from the secondary market, so we quote on confirmed availability only and cannot promise a date before we have a unit identified.
Refurbished, not new. This is six to eight year old hardware. Warranty is Kentino refurb terms, not manufacturer channel warranty. Degraded thermal pads are the main failure mode on these modules and every unit we ship has had them replaced.
One genuine advantage: V100 predates the Advanced Computing Item thresholds, so there is no KYC form and no End User Certificate. If export paperwork is the thing blocking you, this is the node that does not need it.
What you get from Kentino
Kentino sources, refurbishes and supports this node: thermal pad and paste renewal, fan replacement, 72 hour burn-in with a written report, OS and driver stack pinned to a Volta-compatible CUDA version, rack and airflow guidance, EU-based logistics and our refurb warranty.
We run the same refurbishment process on ex-mining accelerator hardware, so this is an established bench workflow rather than a one-off.
Price shown is EUR excluding VAT, EXW Prague (Incoterms 2020). Collection, freight, insurance and any destination import duties are for the buyer’s account. Lead time on request.
Priced per sourced unit — these come from the secondary market, so availability drives the date. Tell us what you need it for - if your workload is modern LLM inference we will point you at the A100 node instead, which is better value than this once you account for what Volta cannot do.
The questions buyers ask us most often before ordering a server.
How long does it take?
Machines built from components we hold ship quickly; anything requiring a specific GPU generation depends on supply. We give you a date before you pay, and if it moves we tell you rather than letting you find out.
Can the configuration be changed before you build it?
Almost always. GPUs, memory, storage and cooling are chosen per order, and the listed configuration is a starting point rather than a fixed package. If you need more VRAM, faster storage or a different cooling approach, say so before you order and we will quote the change.
Can I collect the server in person?
You can. Our warehouse is in Prague, and collection in person is welcome — most people who come use the visit to go through the machine with our engineer and ask the questions that are awkward over email. For orders within the Czech Republic we also try to deliver personally and walk you through the setup on site.
Can I talk to someone who actually understands the workload?
Yes. We have an engineer who works on AI systems specifically, not a general sales desk. If your question is about batch sizes, quantisation, interconnect or where your bottleneck will be, ask it — that conversation usually changes the configuration for the better.
Which model can I run on this configuration?
Yes. Every machine is assembled, burn-in tested and benchmarked on real AI workloads before it ships, and it leaves us with an LLM already installed and running. You plug it in, connect it to your network and start work — the only decision left is which project it runs first.
How do you test a server before shipping?
We run it against actual AI workloads rather than synthetic scores: inference throughput, sustained load behaviour and thermals under continuous operation. You get the benchmark results with the machine, so the performance you were promised is the performance you can verify on day one.
Is the server ready to run when it arrives?
Yes. Every machine is assembled, burn-in tested and benchmarked on real AI workloads before it ships, and it leaves us with an LLM already installed and running. You plug it in, connect it to your network and start work — the only decision left is which project it runs first.
Ships from our EU warehouse. Heavy items may require freight arrangement — contact us for a shipping quote and lead time. 2-year limited warranty with advanced RMA support; extended warranty available.
Not exactly what you need?
Tell us your workload and we'll spec this platform around it — GPUs, memory, storage and cooling matched to what you actually run.