Nvidia released its Ada Lovelace architecture in October 2022, starting with the GeForce RTX 4090 and then following up with additional variants. Prices have increased compared to the previous generation, some more than others, but these are still some of the best graphics cards currently available, with the 4090 occupying the top spot on our GPU benchmarks hierarchy.
But right from the start, we knew that Nvidia was holding back, saving room for a potential RTX 4090 Ti or a new Titan RTX Ada. Here's what we know about the future uber-GPUs of the Ada generation.
Nvidia currently has three desktop Ada GPUs in production, the AD102, AD103, and AD104 that are used in the RTX 4090, RTX 4080, and RTX 4070 Ti, respectively. It also has mobile RTX 40-series GPUs now in production that range from the entry-level RTX 4050 mobile up to the RTX 4090, which add the smaller AD106 and AD107 chips into the mix — though note that the mobile 4090 uses the same chip as the desktop 4080. The RTX 4090 Ti and Titan RTX Ada, depending on which way Nvidia decides to go, will naturally land at the top of the stack, and as such will also use the AD102 chip.
Right now, the RTX 4090, for all its performance and power, only uses a portion of the full AD102 GPU. It's a large portion, to be sure, with 128 out of a possible 144 Streaming Multiprocessors (SMs) enabled — 89% of the GPU cores, in other words. Except Nvidia also disabled 1MB of L2 cache on each of the twelve 32-bit memory interfaces, giving it 75% of the maximum L2 cache. All it needs to do is to turn on those missing pieces, and perhaps even crank up the clock speeds and power draw a bit, and we end up with a new, even faster RTX 4090 Ti.
And make no mistake, Nvidia already has basically fully functional AD102 chips in the wild. The professional Nvidia RTX 6000 Ada Generation, for instance, has 142 of the 144 SMs turned on, plus the full 96MB of L2 cache. Except power draw is capped at a 'meager' 300W, which means the card generally won't run as fast for as long — not actually a bad approach, but we already know that with a bit more power and voltage, it's possible to get most Ada chips to run at closer to 3.0 GHz. That's effectively what we expect to see from a future RTX 4090 Ti and/or Titan RTX Ada.
Nvidia RTX 4090 Ti / Titan RTX Rumored Specifications and Other Ada GPUs
| Graphics Card | Titan RTX Ada | RTX 4090 Ti | RTX 4090 | RTX 4080 | RTX 4070 Ti |
|---|---|---|---|---|---|
| Architecture | AD102 | AD102 | AD102 | AD103 | AD104 |
| Process Technology | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N | TSMC 4N |
| Transistors (Billion) | 76.3 | 76.3 | 76.3 | 45.9 | 35.8 |
| Die size (mm^2) | 608.4 | 608.4 | 608.4 | 378.6 | 294.5 |
| SMs | 142 | 142 | 128 | 76 | 60 |
| GPU Cores (Shaders) | 18176 | 18176 | 16384 | 9728 | 7680 |
| Tensor Cores | 568 | 568 | 512 | 304 | 240 |
| RT Cores | 142 | 142 | 128 | 76 | 60 |
| Boost Clock (MHz) | 2520 | 2625 | 2520 | 2505 | 2610 |
| VRAM Speed (Gbps) | 24 | 24 | 21 | 22.4 | 21 |
| VRAM (GB) | 48 | 24 | 24 | 16 | 12 |
| VRAM Bus Width | 384 | 384 | 384 | 256 | 192 |
| L2 Cache | 96 | 96 | 72 | 64 | 48 |
| ROPs | 192 | 192 | 176 | 112 | 80 |
| TMUs | 568 | 568 | 512 | 304 | 240 |
| TFLOPS FP32 (Boost) | 91.6 | 95.4 | 82.6 | 48.7 | 40.1 |
| TFLOPS FP16 (FP8) | 733 (1466) | 763 (1527) | 661 (1321) | 390 (780) | 321 (641) |
| Bandwidth (GBps) | 1152 | 1152 | 1008 | 717 | 504 |
| TDP (watts) | 800 | 600 | 450 | 320 | 285 |
| Launch Date | ? | ? | Oct 2022 | Nov 2022 | Jan 2023 |
| Launch Price | Arm + Leg + Kidney | Arm + Leg | $1,599 | $1,199 | $799 |
The currently rumored specs are basically similar to what we see with the RTX 6000 Ada Generation, but now with double the TGP (Total Graphics Power) and a slightly higher boost clock on the 4090 Ti, or even a significantly higher 800W power limit for the Titan RTX Ada.
The higher TGPs of course means these cards would be far more likely to hit the rated boost clocks. And if they're like other RTX 40-series desktop GPUs, it's more than likely they'll far exceed the boost clocks. Our testing of RTX 4090, 4080, and 4070 Ti cards to date has shown that real-world gaming boost clocks often average closer to 200 MHz above the official boost clock.
Everything else follows from the core counts and boost clock. In terms of raw specs, if the rumors are correct, the RTX 4090 Ti will hit peak throughput of 95.4 teraflops for FP32 work, and up to 1.5 petaflops for FP8 computations on the Tensor cores. That's 15% faster than the RTX 4090 in both cases.
To help the card sustain maximum throughput, the GDDR6X memory will also get bumped up to 24 Gbps, 14% faster than the 21 Gbps memory used with the RTX 4090. Coupled with 33% more L2 cache, the memory subsystem should be well-equipped to keep the rest of the GPU fed with data.
The Titan RTX Ada rumors are less solid, and currently show a 2520 MHz boost clock — the same as the RTX 4090. Given it's also supposed to have dual 16-pin power connectors and an 800W TGP, however, we would expect such a card (if it actually comes out) to push for even higher clocks. What we've seen of the latest GDDR6X 24 Gbps modules from Micron suggests power and heat from the memory won't be nearly as problematic as it was on the earlier RTX 3090, even with chips on both sides of the PCB.
There's still the power to contend with, however, and even if the cards don't always hit 600W/800W, they're likely to push well above 500W when running demanding games at 4K and maxed out settings. We've already seen that happen with the RTX 4090 cards we've reviewed once we manually overclocked them, but the 4090 Ti / Titan Ada will make that the norm rather than an OC.
Coping with that much power will require an exceptional cooling subsystem. We generally found the RTX 4090's triple-slot cooling to be adequate, but there are images and leaks floating around the web showing a quad-slot gold cooler. Perhaps they're fakes, but the extra slot width plus the gold coloring suggests we're more likely than not looking at a new Titan RTX Ada rather than an RTX 4090 Ti — or maybe Nvidia will do both, although the 4090 Ti would likely have a silver cooler. It's too early to say for certain how everything will play out.
The 4-slot Nvidia RTX 4090 Ti / Titan RTX Ada card images look comically large, and considering even the existing RTX 4090 already starts hitting CPU limits in many games at anything below 4K (at least if the game doesn't have ray tracing), this definitely feels like Nvidia is getting ready to jump the shark. Hopefully Nvidia provides an extra sturdy IO bracket (secured to the top and bottom of the cooler) that covers all four slots to help prevent sag. Or maybe Nvidia will just include a kickstand of some form like many of its AIB partners have elected to do with the RTX 4090.
That just leaves the pricing and release date, neither of which are known yet. Given the leaks of the cooler, we could see the new halo card within the next month or two, or Nvidia may hold out for a summer or even fall launch window, letting the card serve as the mid-cycle refresh at the top of the RTX 40-series stack. Launching at either GDC (in March) or the next GTC (also in March) is certainly a possibility, particularly if Nvidia opts for the "prosumer and creator" target market with a new Titan RTX Ada, and it could hold the RTX 4090 Ti for a later date.
Pricing will naturally be ridiculous. Any hope of a future RTX 4090 Ti or Titan RTX Ada card carrying a reasonable price tag basically went out the window when the RTX 4090 cards were routinely sold out or priced above $2,000 for the first three months after launch. We still suspect a lot of that came from professional users who were willing to pay more for top performance in AI and other computational workloads, and a 4090 Ti or Titan RTX Ada would cater to those same users.
If Nvidia goes with GeForce RTX 4090 Ti branding, it will probably set the price at $1,999. For a new Titan RTX Ada, $2,499 or even $2,999 seems likely. A Titan would offer most of what the RTX 6000 Ada already provides (depending on the VRAM capacity), plus higher clocks thanks to the increased TGP, and three grand would still be less than half the price of the true professional GPU. Those who are only interested in bragging rights at any price also appear to have sufficient funds to pay whatever Nvidia asks.
The rest of us can stick with more reasonable GPUs, like a replacement for the RTX 3060 Ti — or just wait for the inevitable RTX 50-series, which will probably arrive in late 2024. Because if there's one thing we know for certain, it's that no matter how fast the future RTX 4090 Ti and/or Titan RTX Ada end up being, at some point Nvidia will release an RTX 5080 and RTX 5090 (or some other name, perhaps) that will provide even higher performance than what the fastest Ada Lovelace graphics card can hope to deliver.
If we're lucky, maybe the current downturn in GPU and PC sales will even spur more aggressive pricing for a change. That happened, in theory at least, with the RTX 3080 in 2020. It's too bad cryptocurrency mining and the pandemic ended up ruining prices for the next two years and influencing the current round of higher prices.
We do at least have a codename for Ada's successor: Blackwell, presumably after David Harold Blackwell, a mathematician and statistician who specialized in game theory and information theory. Except we don't know for certain if Blackwell is solely a future Hopper replacement (for the data center market), or if it will span both the data center and consumer markets, similar to Ampere. Considering Ada Lovelace is only a few months old, pinning our sights on Blackwell at this stage seems a bit premature.
