Turing as architecture was launched on September 20, 2018 and after a year of sales success in the market and despite its higher price, NVIDIA launched 10 months later a new range that replaced this original called SUPER. Today we are going to face two high-end GPUs and compare their architectures, because is it worth going from an RTX 2080 vs RTX 2080 SUPER?
Curiously, the RTX 2080 SUPER is the only serial GPU from NVIDIA that has a peculiarity that no other has and that makes the comparison very interesting from the point of view of architecture.
The problem that NVIDIA has found is precisely between this RTX 2080 and RTX 2080 Ti between this RTX 2080 SUPER, since the performance difference is not too high. So, let’s look at the changes made to both to understand NVIDIA’s movement.
RTX 2080 vs RTX 2080 SUPER, architecture comparison
We start from a common base such as the TU104 chip, but even within this we find differences that NVIDIA has highlighted. In the case of the RTX 2080 we will have the TU104-400A-A1 version , while its sister SUPER carries the newer TU104-450-A1 . This means that NVIDIA no longer distinguishes between A and non-A versions, so all SUPER GPUs are offered with the best of the brand.
As for their lithographic processes, they both share the 12 nm TSMC and curiously they have the same millions of transistors: 13.6 billion with a 545 mm2 matrix . Therefore we are facing the same physical chip, where the differences will be in adjacent sections as the frequencies.
And is that here are the main differences in these Turing chips, since the veteran RTX 2080 gets a base frequency of 1515 MHz for the 1650 MHz of its sister SUPER. In Boost there are also differences logically, since the first one gets 1710 MHz for the 1815 MHz of its improved version.
But this is not the only thing that distinguishes them and marks the final performance, since the chips despite having the same millions of transistors are not the same in their rendering configuration.
NVIDIA saved the best TU104 for a year later
As for Shaders, NVIDIA did not put all the meat in the starter grill and surprised with the SUPER range and its best features. Thus, the RTX 2080 had 2944 Shaders, 184 TMUs, 64 ROPs and all in 46 SM.
Her sister SUPER gets 3072 Shaders, 192 TMUs, 64 ROPs and 48 SM. This also logically affects its sections for Ray Tracing and DLSS, since the first gets 368 Tensor Cores and 46 RT Cores for the 384 and 48 of the second.
Interestingly, NVIDIA has respected the L2 cache with 4 MB for both cards as well as the purely unmovable arrangement of the 64 KB of L1 for each SM, but this obviously gives us two different values: 2944 KB vs 3072 KB respectively.
These data result in a fairly tight FP32 performance, however, as we always say, it is not everything, but it is an approximation. Thus, the RTX 2080 achieves 10.07 TFLOPS for the 11.15 TFLOPS of the RTX 2080 SUPER, which leaves the percentage difference at 10.72%.
New memories, new consumption, same price
The memory section has also been improved, since it started from the 14 Gbps of the RTX 2080 at 1750 MHz actual frequency and has gone to 15.5 Gbps and 1937 MHz from the RTX 2080 SUPER, this being the only one of The entire range of NVIDIA that carries this type of memory, being by the way the fastest that the manufacturer implements.
Although they remain as GDDR6, its 8 GB of VRAM and a 256-bit bus in both models, the bandwidth has gone from 448 GB / s to more interesting 495.9 GB / s , an increase that balances the increase chip performance.
The cards have hardly changed their heatsink as such, beyond a slight aesthetic section with new screen printing, but what has changed is the consumption. We went from 215 watts of TDP to dissipate for the RTX 2080 to an amazing 250 watts for the RTX 2080 SUPER, yes, keeping the same 6 + 8 power pins in both cases.
This change supposes 25% more heat to dissipate by the heatsink, thereby equalizing the heat generated by an RTX 2080 Ti, but without the performance of this, which indicates that the chip has now been pushed to a limit point.
In terms of performance, the RTX 2080 remains at 6% of the performance of this new RTX 2080 SUPER, while the latter in turn is set at 10% of the RTX 2080 Ti. The good thing is that NVIDIA has respected the same price for the two rivals, so they have been offered at $ 699. If we come from an RTX 2080, the jump to the SUPER range does not compensate us for the difference in benefits, since we will have to put at least between 150 to 200 euros for the jump, which does not compensate.
But if we are going to acquire one of the two rivals, the option of the RTX 2080 SUPER should be considered, since if the prices are very close between the models that we like, the extra performance is a plus to consider.
