NVIDIA RTX 5000 Ada Workstation GPU Might Be In The Works With These Killer Specs

NVIDIA RTX 6000 graphics accelerator.
NVIDIA kicked off this year's GPU Technology Conference (GTC) with a flurry of announcements and an interesting keynote about bringing AI everywhere, all at once. As part of the busy first day, NVIDIA unveiled a new workstation card based on Ada Lovelace, the RTX 4000 in small form factor (SFF) trim. Now a day later, the rumor mill is already brimming with chatter about yet another workstation addition, the RTX 5000.

To be clear, NVIDIA made no such announcement regarding a possible RTX 5000. The RTX 4000 SFF, on the other hand, is now official and packs 6,144 CUDA cores, 192 fourth-generation Tensor cores, 48 third-generation RT cores, and 20GB of GDDR6 memory (16Gbps) with error correction code (ECC) support linked to a 160-bit bus, giving it 320GB/s of memory bandwidth.

It also boasts four mini DisplayPort 1.4a outputs, AV1 encode and decode support, and various other features. The card sports a dual-slot blower-style cooling solution and boasts a power envelope of just 70W. Overall it looks like a promising option for compact workstation, with a balance between performance and power efficiency that's suitable for smaller systems.

What about the RTX 5000, though?

Twitter post on the rumored RTX 5000.

According to @kopite7kimi, a prominent leaker on Twitter, NVIDIA might be readying an RTX 5000 workstation card with 15,360 CUDA cores and 32GB of GDDR6 memory. If true, we'd be looking at a cut-down version of NVIDIA's AD102 GPU, as AD103 and AD104 both have fewer CUDA cores than what the leaker is claiming. That's also the same GPU found in NVIDIA's RTX 6000 workstation card, whereas the newly unveiled RTX 4000 SFF is built around the company's AD104 GPU.

While only a rumor at this point, an RTX 5000 would neatly slot in between the RTX 4000 and RTX 6000. Here's a high-level overview...
  • RTX 6000 Ada: 18,176 CUDA cores, 48GB GDDR6, 384-bit bus
  • RTX 5000 Ada: 15,360 CUDA cores, 32GB GDDR6, 256-bit bus
  • RTX 4000 Ada: 6,144 CUDA cores, 20GB GDDR6, 160-bit bus
Again, the RTX 5000 does not exist (yet), and perhaps never will. However, if it does come to fruition and the leaked specs turn out to be accurate, it could end up with up to 640GB/s of memory bandwidth (based on 20Gbps memory chips), which is twice as much as the RTX 4000 SFF. Or if it uses 16Gbps chips like the RTX 4000 SFF, memory bandwidth would end up at 512GB/s. That's still a respectable 60 percent gain.

As a side note, the RTX 6000 still leaves some performance untapped within NVIDIA's AD102 GPU. In full-fat form, the AD102 features 18,432 CUDA cores. Likewise, the RTX 6000 and RTX 4000 both use GDDR6 memory instead of GDDR6X.