One RTX PRO 6000 or Four RTX PRO 6000 GPUs: Why OEM Workstation Design Matters

One RTX PRO 6000 Blackwell fits in many workstations. Four need a chassis designed for them.

• Edition and power: four 300 W Max-Q cards draw 1,200 W before the CPU, so the system needs a 2–2.7 kW power design.
• Slots and spacing: it needs enough Gen5 x16 slots wired to the CPU, spaced so the inner cards can breathe.
• Cooling: airflow must move the heat of four blowers out of the case.

That is why the OEMs publish hard limits. HP, Lenovo and Dell each list four Max-Q cards for their top towers, and most allow only one 600 W Workstation Edition. Their slot maps also show that not every four-card layout runs every card at full Gen5 speed.

Which towers are designed for four cards?

Source: HP Z8 Fury spec page and StorageReview reviews (Z8 Fury G5, 9 May 2023; G6i, 23 Jun 2026); Lenovo PSREF for PX (9 Oct 2026), P8 and P7 (Sep 2026) and PX Power Configurator v1.9; Dell Precision 7960 technical guidebook (Aug 2024) and Dell configurator; StorageReview, 16 Mar 2026 (Dell Pro Precision). Slot and speed assignments for four-card builds are our reading of the published slot maps, not OEM statements.

Why does PCIe topology matter with four cards?

Each card talks to the CPU, and to the others, over PCIe. A Gen5 x16 slot carries about 63 GB/s each way, and Gen4 x16 about half that. With one card, any good Gen5 x16 slot will do. With four, where each card is attached starts to matter:

• HP Z8 Fury G6i: one Xeon 600 with four Gen5 x16 slots, so all four cards can sit on the same CPU at full Gen5 speed.
• HP Z8 Fury G5 and Dell Precision 7960: each publishes two Gen5 x16 and two Gen4 x16 slots for graphics. Our reading is that a four-card build runs two cards at Gen4.
• Lenovo ThinkStation PX: a dual-socket design. Its population order puts two cards on each CPU, the fourth in a Gen4 slot, so traffic between a card on one socket and a card on the other crosses the inter-CPU link.

For workloads that keep each GPU busy on its own job, such as rendering separate frames or running one model per card, this barely matters. For a single model split across four cards with tensor parallelism, the slowest link sets the pace. Our post on multi-GPU AI without NVLink explains why. Ask the OEM to confirm slot speeds for your exact configuration before you rely on our inference.

How much space do the cards need?

More than the slot pitch suggests. Users on the Level1Techs forum who built four-card Max-Q systems on a standard workstation motherboard reported that the inner two cards throttled, and recommended a slot of space between cards. HP designs the Z8 Fury G5 around GPUs set "a nonstandard 2.5 slots apart". Lenovo's P8 and P7 place their three double-width x16 slots in alternate positions. Integrators report the same limit. Exxact said in March 2026 that its four-Max-Q workstation throttled "without the custom cooling" it added, and VRLA Tech reported that four 600 W Workstation Editions failed its validation while four Max-Q passed.

At one card, the workstation hosts the GPU. At four, the workstation is the product, and the GPUs are only as fast as its slot map and airflow allow.

Why Max-Q, and how much power does it take?

The Max-Q's blower sends heat out of the back of the case, while the Workstation Edition's flow-through cooler blows it into the chassis. NVIDIA's own Max-Q datasheet says "up to four" per system, and our edition comparison covers the trade-off. Puget measured the Max-Q at 5–14% slower than the Workstation Edition in content-creation applications.

Four Max-Q cards draw 1,200 W, against 2,400 W for four Workstation Editions. Add a 350 W workstation CPU, memory and storage, and the OEM four-card towers above all use 2.2–2.7 kW supply designs. Several details matter at that level:

• Lenovo PX: four cards in redundant power mode require a CPU of 165 W TDP or less. Team mode delivers 2,350 W at 115–127 V.
• HP Z8 Fury G5: in aggregate mode, HP advises a separate circuit for each supply.
• Current draw: by our arithmetic at 230 V, a 2,350 W system draws about 11.1 A and a 2,700 W system about 12.5 A, close to a 13 A socket's rating. Have an electrician confirm the circuit.

What counts as a certified multi-GPU configuration?

Each OEM's configurator and spec sheet is the practical certification: it only lets you order counts and editions the OEM has tested and will support. Certification by software vendors is per system and GPU. Dassault's 2026 lists include RTX PRO 6000 Max-Q entries for the HP Z8 Fury G5 and Dell Precision 7960, among others, but don't state a GPU count. Independent four-card testing is thin. StorageReview has reviewed two Max-Q cards in an HP Z8 Fury G6i and in a Dell Precision 7875, and we found no published thermal test of a four-Max-Q HP, Lenovo or Dell tower. That is another reason to buy the configuration the OEM sells rather than adding cards later.

Whether four cards are worth it for your models is a separate question. Our look at four RTX PRO 6000s and 384 GB of VRAM covers which workloads benefit and why the memory doesn't become one pool. For brand-by-brand detail, see HP vs Lenovo vs Dell vs ASUS. If you're weighing a self-built four-card system, read what can go wrong in a DIY build.

What do one and four cards cost?

Four Max-Q cards at SourceIT's listed price come to S$116,400 before a host. Dell Singapore's Precision 7960 configurator listed four Max-Q cards as a S$135,098.96 add-on on 6 Nov 2026, or about S$144,325 with the featured base system, including GST. A single Max-Q added S$33,453.19. The difference between buying four cards and a validated four-card Dell system is therefore mostly the chassis, power design, warranty and support. Business quotations can differ from these online figures.

Buying in Singapore

SourceIT lists the RTX PRO 6000 Blackwell Max-Q Workstation Edition and the 600 W Workstation Edition at S$29,100 each (listed 6 Nov 2026). The four-card towers above aren't in stock. We can quote HP Z8 Fury G6i, Lenovo ThinkStation PX or P8, and Dell Precision or Pro Precision 9 systems in OEM-supported multi-GPU configurations, and advise on power and circuits for your site. For a UPS, our APC Smart-UPS SRT range includes 5–10 kVA models. Prices change, so ask for a current quotation. Use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.

Frequently asked questions

How many RTX PRO 6000 cards can an OEM workstation take? HP's Z8 Fury G5 and G6i, Lenovo's ThinkStation PX and Dell's Precision 7960 list four Max-Q cards. Lenovo's P8 and P7 list three. Dell's Pro Precision 9 T6 lists five 300 W GPUs.

Can I use four Workstation Editions instead of Max-Q? Not in the OEM towers we checked; most allow only one 600 W card, and Dell's Pro Precision 9 T6 allows two. The Workstation Edition's cooler exhausts into the case, and VRLA Tech reported four of them failed its validation.

Do all four cards run at PCIe 5.0 x16? Not always. Our reading of HP's Z8 Fury G5 and Dell's Precision 7960 slot maps is that two of four cards would run at Gen4. The HP Z8 Fury G6i lists four Gen5 x16 slots. Confirm with the OEM for your configuration.

How much power does a four-card workstation need? The four Max-Q cards draw 1,200 W, and OEM four-card towers use about 2.2–2.7 kW supply designs. At Singapore's 230 V that can mean 11–12.5 A, so check the circuit with an electrician.

Is a four-card OEM workstation much more expensive than four cards? On Dell Singapore's online configurator, four Max-Q cards in a Precision 7960 came to about S$144,325 including GST, against S$116,400 for four cards alone at SourceIT's listed price. The difference pays for the validated chassis, power, warranty and support.