The NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition carries the same 96GB of GDDR7 and the same 24,064 CUDA cores as the full 600W card, in a standard dual-slot full-height board that draws half the power. For rackmount, multi-GPU and flypack builds, that is the difference between a design that fits and one that needs a new power supply and a new chassis.
Verified against NVIDIA's own published specifications. Confirm current part numbers at quote time — board partner SKUs vary.
RTX PRO 6000 Max-Q
96GB of ECC GDDR7 on a 512-bit bus at 1792 GB/s, across 24,064 CUDA cores, 752 fifth-generation Tensor cores and 188 fourth-generation RT cores.
Video engines
Ninth-generation encoders add 4:2:2 H.264 and HEVC encode, and the sixth-generation decoders add 4:2:2 decode. That is the chroma format professional cameras actually record, which is why this matters more to a post house than the FLOPS number.
Displays & partitioning
Four DisplayPort 2.1b outputs, and Multi-Instance GPU splits one card into up to four isolated instances with their own memory and cores — one box, four users, guaranteed quality of service.
Identical memory. Identical core count. The difference is power and physical size, and in a rack that decides the build.
Max-Q: 300W total board power, 3511 AI TOPS, 110 TFLOPS single-precision, 333 TFLOPS RT, 4.4" x 10.5" dual-slot full-height.
Workstation Edition: 600W, 4000 AI TOPS, 125 TFLOPS single-precision, 380 TFLOPS RT, 5.4" x 12" dual-slot extended height.
So Max-Q gives up roughly twelve percent of the compute and keeps every gigabyte of the memory, at half the power draw and in a card that fits a normal full-height slot. If you are building a single tower with headroom to spare, take the 600W part. If you are putting GPUs in a 3U or 4U chassis, running more than one per machine, or working inside a power budget that already has a UPS sized for it, Max-Q is the card that ships.
Both are Blackwell, both are 96GB GDDR7 with ECC, both are PCIe 5.0 x16, both use a single PCIe CEM5 16-pin connector, and neither supports NVLink or vGPU.
Buying the card is easy. Getting it into a machine that has the PSU headroom, the physical clearance, the airflow and the CPU and memory to keep it fed is the part that goes wrong — and a 300W card in a chassis designed for 250W is a support call, not a bargain.
SHTS specs the whole machine: GPU, platform, memory, storage, capture, networking and the rack it lives in. Over a hundred custom production machines built, by a broadcast engineer who still works the jobs they get used on. You will get told which parts of your build are wrong before the purchase order, not after.
Related: the full AI infrastructure stack · custom production PCs & workstations · business IT equipment · government & education IT
Send the workload and the chassis constraints. You get a real line-item quote with lead times.