HX Cooler Kit - 2 GPU - 120 mm PWM Fan Manifold for CMP 170HX / A100
Enterprise-grade platform, assembled and tested in the EU.
HX Cooler Kit — 2 GPU
One manifold, two passive accelerators. A single 120 mm PWM fan feeds a plenum spanning both cards, so a dual-GPU desktop build needs one fan and one duct rather than two of everything. Ships with two PCIe-to-EPS converters — one per card.
A datacenter GPU has no fan of its own
Every passive accelerator — the CMP 170HX, the A100, the whole Tesla line before them — ships as a bare finned heatsink with no fan attached. In a rack that is deliberate: a 2U chassis puts a wall of 15,000 RPM fans in front of the cards and forces air down their length. Drop the same card into a desktop tower and there is no such wall. Case airflow drifts past the heatsink instead of through it, the card heat-soaks within minutes, and you watch it clock down to nothing — or trip thermal shutdown.
This kit supplies the missing half of the design. A 120 mm fan is sealed into a manifold that mates to the intake face of the heatsink, so every litre the fan moves is forced through the fins and out the back of the card rather than around it. It is the same principle as the server shroud, sized for two cards side by side on a desktop board.
What you get
- The manifold — a sealed plenum sized for two cards side by side, mating to the heatsink intake face.
- 1 × 120 mm PWM fan — 1,000–6,000 RPM, 2.7 A, 4-pin PWM control.
- 2 × PCIe-to-EPS converters — one per card, so nothing else is needed to power the GPUs.
- Mounting hardware for the fan-to-manifold and manifold-to-card interfaces.
Why the pair shares a plenum
Spanning both cards from a single sealed plenum is not a cost compromise, it is the better arrangement. A shared plenum pressurises across the full width and pushes into whichever heatsink offers the least resistance, which is the hotter one. Two independent fans, by contrast, hold two independent pressure zones and cannot balance between them.
Why the card needs a converter, not just a cable
Datacenter cards take power through an 8-pin EPS inlet at the end of the card — the CPU-style connector, not the graphics-card one. It is the single most common way people destroy one of these on the first boot: an EPS 8-pin and a PCIe 8-pin look alike, and in the wrong hands one will go into the other.
They are not the same. A PCIe 8-pin carries three 12 V lines and five grounds; an EPS 8-pin carries four and four, in a different arrangement. Cross them and you feed 12 V into a ground pin. The converters in this kit remap the pins correctly and take their feed from your PSU's PCIe cables, which every desktop supply has spare.
6,000 RPM is the ceiling, not the setting
The fan in this kit is an industrial-grade unit with a very wide range: it will idle near 1,000 RPM and it will pull 2.7 A flat out at 6,000. At the top of that range it is genuinely loud — this is server-class air movement and we are not going to pretend otherwise.
The point of the 4-pin PWM control is that you almost never run it there. Put the fan on a curve tied to GPU temperature and a two-card build sits in the 2,000–3,000 RPM band during sustained inference. The headroom above that is there for the days you saturate both cards at once, or for a warm summer with no air conditioning.
Which cards this fits
The manifold is cut for the standard NVIDIA full-height, full-length, dual-slot passive body — the 267 mm card with a finned tunnel running its whole length and the 8-pin EPS inlet at the far end. That covers CMP 170HX, A100 PCIe, A30, A40, L40, V100, P100 and P40, and the AMD Instinct cards of the same generation share the geometry closely enough to work.
Technical data
| Cards supported | 2 × full-height, full-length dual-slot passive accelerators |
| Fan | 1 × 120 mm, 1,000–6,000 RPM |
| Fan current | Up to 2.7 A at 12 V, per fan |
| Fan control | 4-pin PWM |
| Power converters included | 2 × PCIe to 8-pin EPS |
| Card power inlet | 8-pin EPS at the end of the card |
| Verified cards | CMP 170HX, A100 PCIe, A30, A40, L40, V100, P100 and P40 |
| Price | €65 ex VAT |
FAQ
Do I need this, or will good case airflow do?
You need it. Case airflow moves air through the case; a passive accelerator needs air forced through a narrow finned tunnel against real static pressure. Those are different jobs. Builders who skip the duct report the card idling fine and then collapsing under the first sustained load — which is exactly the load you bought it for.
Is one fan really enough for two cards?
Yes, and it is the better arrangement — see the section above. A shared plenum balances pressure between the two heatsinks in a way two separate fans cannot. You will run it somewhat higher on the curve than a single-card build.
Is it loud?
At 6,000 RPM, yes. On a temperature-driven PWM curve during normal inference work it settles around 2,000–3,000 RPM. You control where on that range you live.
Can I run the fan from a motherboard header?
Take the PWM signal from the header, but not the power. At 2.7 A the fan exceeds what a standard chassis header is rated to deliver. Feed it from the PSU through a powered hub.
What PSU do I need for two cards?
Budget around 250 W per accelerator on top of the rest of the system, and make sure the supply has enough spare PCIe 8-pin outputs to feed both converters. A 1,000 W-class supply covers a typical two-card desktop build.
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.