RTX PRO 5500 Shares the RTX 5090's GB202 Config at a Much Higher Price

The RTX PRO 5500 isn't a straight rebadge of the RTX 5090, but it draws on the same silicon. Both cards use GB202 with 170 of its 192 streaming multiprocessors enabled, which works out to 21,760 CUDA cores, 680 Tensor cores and 96 MB of L2 cache. That's the detail that makes the 5500 worth a closer look, because it means the workstation card is competing for the same dies that used to go almost entirely into gaming flagships.

Same die, different card

The differences sit in the board around that chip. The RTX PRO 5500 carries 84GB of ECC GDDR7 across 28 3GB modules, with memory on both sides of the PCB. The RTX 5090 has 32GB. Nvidia gets to that capacity by running the memory over a narrower, slower interface, so bandwidth drops from 1,792GB/s on the 5090 to 1,398GB/s, a cut of about 22%. For inference workloads that depend on fitting large models into one card, the capacity matters more than the bandwidth loss. For gaming, the 5500 isn't in the conversation.

The rest of the spec sheet points the same way. The board is rated for 600W instead of 575W. The card can be partitioned into two isolated 42GB instances, ships with professional drivers, and is designed for rack-mounted workstations with blower or liquid cooling. Nobody is pulling a finished 5090 off a shelf and relabeling it. What's shared is the die, and the die is the scarce part.

Tom's Hardware's comparison of GB202 products makes the point clearly. Among the cards built on this chip, only the RTX 5090 and the RTX PRO 5500 use the 170-SM configuration. The RTX PRO 6000 uses a higher bin at 188 SMs, the 6000D uses 156, and the RTX PRO 5000 uses 110. The 5500 is the first pro card to take the same cut of silicon that the 5090 used to get almost exclusively.

Why the die split matters

The price gap is the hard part to explain away. The RTX 5090 launched at $1,999. Reuters reported in October 2026 that ByteDance, Alibaba and Tencent had ordered more than three million RTX PRO 5500s at about $11,400 each. A US reseller, SHI, lists the card's MSRP at $17,999. Against the 5090's launch price, the Chinese bulk figure is roughly 5.7 times higher, and the US list price is about nine times higher.

Memory cost explains part of that. Adding 52GB of GDDR7 isn't cheap in the current memory market. It doesn't come close to accounting for a gap of that size, though. Per die, Nvidia appears to earn far more from a 5500 than from a 5090, and that's a strong commercial reason to direct GB202 output toward the workstation product.

Demand makes the calculation sharper. Reuters' sources said total Chinese orders for the export-compliant card had passed six million, while Nvidia could produce no more than three million units of it in 2027. If those numbers are anywhere near accurate, Nvidia has more buyers for the 5500 than it can supply. Sending a 170-SM die to a gaming card at a fraction of the margin would be hard to justify, and that's a plausible explanation for the 5090 shortages showing up across markets.

Not discontinued, just not available?

Rumors about GB202 being reassigned started before the 5500's price made headlines. On October 10, leaker MEGAsizeGPU claimed that "all future GB202 supplies are RTX Pro exclusive." A second account, hongxing2020, shared what it described as a factory notice saying production of every RTX 5090 model would stop, including China's D variants. TweakTown covered the claims, which remain unverified. TechRadar's Darren Allan argued that any end-of-life would likely come without an announcement, with allocation simply shrinking until stock "will effectively dwindle to next to nothing."

Nvidia's response was narrow. Speaking in Taipei on October 16, SVP Jeff Fisher said that "although the RTX 5090 has other variants, it has definitely not been discontinued." He acknowledged constrained supply. Neither he nor Nvidia has said how GB202 output is split between the 5090 and the other products built on the chip. A denial of discontinuation doesn't rule out reallocation.

Secondhand reports point the same way. German overclocker der8auer reported that an unnamed Nvidia board partner, which said it had never been told the 5090 was end-of-life, could no longer order the GB202 and 32GB memory kit it needs to build one. It also couldn't place backorders the way it had during earlier shortages, according to Wccftech. A cooler supplier told der8auer that two partners had cut 5090 cooler orders for the first quarter of 2027 and placed none for the second. Those accounts are anonymous, and they're thin on verification. They do fit a pattern where a product is never formally discontinued and simply starved of supply.

Gray-market activity shows where some of those chips may be going. HKEPC shared video that appears to show hundreds of retail RTX 5090s in a Chinese facility, where technicians strip GPUs and memory and move them onto custom boards with dual-slot blower coolers for AI servers. HKEPC also supplied September photos of 5090s on pallets, which TechRadar Pro reported may have been AI-generated, so the new footage deserves caution too. Separately, Suqiao listed a modified 96GB RTX 5090 on Alibaba for $3,888. Tom's Hardware noted that the listing's memory specifications didn't fully add up.

Nvidia doesn't need to rebadge the 5090 to get what it wants. The RTX PRO 5500 lets the company sell the same GPU configuration at several times the price into a market with more demand than supply, without announcing that the gaming flagship is gone. Its product page still lists the 5500 as coming soon, with no price.

Until Nvidia says how it's dividing GB202 between gamers and workstation buyers, "not discontinued" and "not available" can both be true. For anyone deciding between a 5090 and a 5500 for compute, the die is the same, and the board, memory and price are what differ. For gamers, that's probably the bigger problem.