RTX PRO 5000 Blackwell: 48 GB or 72 GB, and what the extra 24 GB holds

One card, two memory sizes

NVIDIA’s datasheet gives both versions of the RTX PRO 5000 Blackwell the same 14,080 CUDA cores, 300 W and 1,344 GB/s of memory bandwidth, so a model that fits in 48 GB runs at the same speed on both. The 72 GB card helps when a model, its cache or a scene needs the extra 24 GB, and only it is on NVIDIA’s vGPU list. The RTX PRO 5000 Blackwell arrived with 48 GB in 2025. The 72 GB version appeared on NVIDIA’s site in October 2025 and NVIDIA announced its general availability on 18 December 2025; the current datasheet covers both in one table.

NVIDIA RTX PRO 5000 Blackwell datasheet (July 2026), MIG user guide (4 November 2026) and vGPU documentation (September 2026). NVIDIA’s December 2025 launch blog quotes 2,142 TOPS for the 72 GB card; every NVIDIA specification table gives 2,064.

NVIDIA’s MIG guide lists the RTX PRO 5000, in its 48 GB version, on the GB202 chip, the same die as the RTX PRO 6000, and the datasheet gives both versions the same 14,080 CUDA cores. Cores, bandwidth and power are identical; only the memory differs, and it decides what fits.

The same speed when the model fits

Token generation reads the weights once per token, so with equal bandwidth the two versions generate at the same rate on any model that fits both. Puget Systems tested both in Topaz Video’s Starlight models and, in Starlight Mini, found them “less than 1% apart, showing that VRAM capacities this size did not affect performance in our testing”. For language models we found no published figure for either version that states the engine and its version, the precision, the concurrency and the test system in text: a vLLM benchmark of the 48 GB card names the engine, the precision and the concurrency, but not the vLLM version or the test system. By our arithmetic, 1,344 GB/s limits generation of Llama 3.3 70B in NVFP4, which reads about 40.6 GB per token, to about 33 tokens per second for a single user, on both cards alike; that is a ceiling from the specifications, not a measurement.

The same holds for rendering. Puget measured the RTX PRO 5000, tested with 48 GB, at 50 per cent above the RTX 5000 Ada in Blender and 35 per cent above it in V-Ray GPU, where it also beat the older RTX 6000 Ada by 11 per cent. A scene that fits in 48 GB should render no faster on 72, although we found no published rendering test of the 72 GB card.

What the extra 24 GB holds

The difference shows up when the weights and the cache no longer fit. The table uses the rule of our users-per-card guide: weights from the published checkpoints, 90 per cent of the memory the driver reports, less about 3 GiB for activations and the runtime, and FP8 cache at 160 KiB per token for Llama 3.3 70B, which is 1.25 GiB per 8,192-token conversation.

Our arithmetic from checkpoint sizes; 32B figures for Qwen3-32B (32.8 billion parameters; Qwen’s FP8 checkpoint is 32.0 GiB). An FP16 cache instead of FP8 halves the conversation counts. Driver-reported memory is taken as 47.8 and 71.7 GiB, in proportion to the 95.6 GiB of the 96 GB card. ECC is always on with GDDR7 and, by NVIDIA’s vPC sizing guide (19 August 2026), does not reduce usable memory, so these figures apply with ECC on. We size with the memory the driver reports (nvidia-smi, FB Memory Total); check it on your card. Production servers keep more in reserve.

For language models, the 70B row shows the main difference. On 48 GB a 4-bit 70B model loads but leaves almost no room for conversations; on 72 GB it serves a small team. gpt-oss-120b also separates the two: at 60.8 GiB it loads on the 72 GB card with about 0.7 GiB to spare, if the driver reports the full 71.7 GiB, and it does not load on the 48 GB card at all. Both versions are available to order from us. Name your model and how many people will use it at once, and we reply with the memory size that fits.

MIG: two halves of 24 or 36 GB

Both versions split into two isolated instances. NVIDIA’s MIG user guide documents the 48 GB card with the profiles 1g.24gb and 2g.48gb, plus variants with graphics support; the 72 GB card splits into two instances of 36 GB according to NVIDIA’s datasheet, but its profiles are not yet in the MIG guide. NVIDIA’s MIG guide, updated on 4 November 2026, sets the requirements for the 48 GB card: Linux, driver 575.57.08 or later, a vBIOS of 98.02.73.00.00 or later, and the display mode switched from graphics to compute with DisplayModeSelector 1.72 or later, which switches off the card’s display outputs. The 72 GB card is not yet in the guide, and NVIDIA names no separate minimum for it; NVIDIA’s DisplayModeSelector page lists the RTX PRO 5000 Blackwell among compatible GPUs without naming a tool version.

A 24 GB half takes gpt-oss-20b or a 14B model in FP8. A 36 GB half takes a 32B model in 4-bit with cache for about nine 8K conversations, while on a 24 GB half the same model does not fit, which is a good reason to pick the larger card for a two-person setup. In FP8, a 32B model does not fit a 36 GB half.

vGPU: supported only on the 72 GB card

Only the 72 GB card is on NVIDIA’s vGPU supported-GPU list. Support starts with vGPU 20.2, released in August 2026 and supported until March 2027, and the release notes limit it to Red Hat Enterprise Linux with KVM 9.6; the long-term vGPU 19 does not cover the card. NVIDIA’s support matrix of 29 September 2026 still lists no other hypervisor for it, so VMware vSphere, Nutanix and Windows Server are not covered. The vGPU 20 user guide of 30 September 2026 lists time-sliced and MIG-backed graphics vGPU types (Q- and B-series) for the card; we found no compute (C-series) types for it in NVIDIA AI Enterprise’s documentation. Its MIG profiles are not yet in NVIDIA’s MIG guide, so check the profile list for your vGPU release before planning MIG-backed virtual machines. The 48 GB version has no vGPU support.

For virtual machines on other hypervisors, the cards to look at are the RTX PRO 6000 Server Edition, supported since vGPU 19.0, and the RTX PRO 4500 Server Edition, supported since 20.0. Our MIG and vGPU guide compares them.

We supply both server cards, with the vGPU licences in the same quote. Tell us the hypervisor and how many users will work in the virtual machines at once, and we size cards and licences together.

Two 72 GB cards or one bigger card

Two 72 GB cards give 144 GB at 600 W of board power, the same power as one RTX PRO 6000 Workstation Edition with 96 GB. The pair suits two separate models or two teams better than one model split across both, because NVIDIA lists no NVLink for the RTX PRO 5000 and a split model talks over PCIe. When one model needs more than 72 GB on one card, the RTX PRO 6000 with 96 GB and 1,792 GB/s is the next step, and the RTX PRO 5500 with 84 GB at up to 600 W, offered through NVIDIA’s OEM and distribution partners as of 5 November 2026, sits in between.

Choosing between 48 GB and 72 GB

Choose 48 GB for models up to 32B in FP8, rendering scenes that fit, and CAD, where Puget found the professional tiers perform much alike.

Choose 72 GB for a 4-bit 70B model serving a team, for gpt-oss-120b on one card or 32B models in BF16 (both with under 1 GiB to spare), for two MIG instances of 36 GB, and for graphics vGPU on Red Hat KVM.

Choose the 72 GB card only if the workload will use the extra 24 GB. Our comparison of the four 48 GB cards sets the 48 GB version beside the RTX 6000 Ada, the RTX 5880 Ada and the L40S.

What we supply

We supply both versions of the RTX PRO 5000 Blackwell across the EU with manufacturer warranty, on one EU contract and invoice, as a single card or in a workstation built to order. For models that need more memory, the other cards we supply from the RTX PRO family include the 84 GB RTX PRO 5500 and the 96 GB RTX PRO 6000 in all three editions.

FAQ