NVIDIA’s Next-Gen Ada Lovelace Flagship GPU to Hit 100 TFLOPs of FP32 Compute Power

The GPU market is heating up as both the main competitors, NVIDIA and AMD, prepare to release their next-generation graphics cards before the end of this year. With the Arc A-Series, Intel, a firm mostly focused on producing CPUs, enters the realm of discrete graphics for the first time, too. With three companies set to compete for the people’s pockets this time, competition is at an all-time high, and leaks are inevitable.

Leaks and rumors from multiple sources have been making the rounds in a time when official news has been lacking. New claims pertaining to NVIDIA and AMD’s top-end flagship SKUs are surfacing every day and today is no exception as we have two new ones covering the computer power of both rivals. Interestingly, the current numbers are surprisingly similar for Team Green and Team Red, providing another preview of how hot the market is set to get.

While I’ve already done an article on AMD’s flagship GPU and the kind of performance we can expect, this article is dedicated to NVIDIA’s next BFGPU. Ada Lovelace, NVIDIA’s next-gen GPU architecture behind the upcoming RTX 4000 series (or whatever it may be called) is supposedly bringing a substantial upgrade over the current-gen RTX 3000 series. This strategy is nearly identical to that of AMD’s who is also said to bring forth massive improvements over RDNA 2 with RDNA 3.

NVIDIA Ada Lovelace GPU | NVIDIA

Ada Lovelace hitting 100 TFLOPs 

NVIDIA’s next-gen GPU is the AD102, and it will be used to power the RTX 4090 (and 4090 Ti), and maybe even a TITAN-class card. According to two new rumors from popular leakers kopite7kimi and Greymon55, AD102 will break the 100 TFLOPs barrier and offer the highest FP32 compute performance of any mainstream consumer graphics card, ever. 

For context, the current-gen RTX 3090 Ti offers 40-45 TFLOPs of FP32 compute power, the most in any GPU out right now. So we’re looking at, at least a 2x improvement over the current BFGPU with AD102. Last year, when the RTX 3090 (non-Ti) was still the most performant consumer gaming GPU NVIDIA made, a leak from Greymon themselves claimed that AD102 would pack more than double the power of the RTX 3090, a report that has largely remained true to this day.

Both leakers resist from putting an exact number on AD102’s FP32 capabilities, but the general sentiment is that it will most likely hit 100 TFLOPs and cross it. AMD’s next-gen Navi 31 flagship GPU, on the other hand, is rumored to pack 92 TFLOPs of FP32 computer power. That’s a 4x improvement from RX 6900XT‘s 23 TFLOPs. This kind of an unprecedented boost in performance further explains the massive jump in power requirements, too.

Today, 350-400W GPUs have become the norm and the RTX 3090 Ti easily breaks 500W in some extreme variants of the GPU. That already is a sizeable change from the 275W norm of previous-gen. But come next-gen, we are expecting graphics cards to hit up to 900W of TGP. You don’t need to be a hardware engineer to know that number is insane. GPUs were already power-hungry but the next generation is about to take that precedent to a whole different level.

It’s important to note that TFLOPs are not all that matters at the end of the day. More teraflops does not directly translate into better gaming performance. There are a myriad of other factors at play here and even the leakers themselves suggest that things like ray tracing capabilities, proprietary upscaling tech, and more will all contribute into making the next-gen GPU successful.

Further breakdown

As for the specs, AD102’s 100 TFLOPs number is achieved through housing 18,432 CUDA Cores clocked at 2.7Ghz. However, that is the maximum amount of CUDA Cores present on the full-fat die that the RTX 4090 Ti (and possibly a new RTX TITAN) will use. The RTX 4090 will most likely have a cut-down version of AD102 which will affect its TFLOP count. But, as I said, TFLOPs are not that important and fewer cores means higher clock speeds which actually does impact the gaming performance significantly.

Case in point, Greymon55 has said that AMD’s next-gen RDNA 3 flagship GPU will hit speeds of up to 3Ghz. Even though AD102’s 2.7Ghz is already very close to that, the cut-down RTX 4090 could actually match AMD’s boost clocks. Therefore, expect ~3Ghz to become the standard boost clock for next-gen flagship graphics cards. And you can thank TSMC‘s 4nm process node for partially being reponsible for that as it will be used to manufacture both RDNA 3 and Ada Lovelace GPUs.

Again, all of this is theoretical performance, you would never even need to know the compute power of your GPU in TFLOPs to judge its gaming performance. Teraflops are calculated by taking the total amount of cores and multiplying it by the clock speed of the GPU, then multiply that answer by 2 to get the TFLOP count. Notice how this is a very superficial level of judging how many (floating-point) operations the GPU could calculate in a second as it does not take any other spec into account, proving how teraflops are best preserved for flexing only.

Still, it’s fun to look at numbers to over-analyze everything months before the product comes out, I mean that’s the crux of gaming hardware journalism after all. But we won’t have to wait that long before news from official channels starts to pour in. AMD and NVIDIA are expected to launch their next-gen GPUs around the same time in the third quarter of this year, with preliminary announcements as early as Computex 2022 in May.

Huzaifa Haroon
Born and raised around computers, Huzaifa is an avid gamer and a keyboard enthusiast. When he's not solving the mysteries of technology, you can find him scrutinizing writers, striving to inform the curious.