Last year, Nvidia revealed Turing, and the Turing architecture forms the basis of several Nvidia products, despite gamer’s largely being focused on only the GeForce cards.
The first range of products is of course, GeForce and there are GTX and RTX cards, with GTX being lower-end SKUs such as the GTX 1660 Ti and while has the architectural enhancements of Turing over Pascal such as cache tweaks, simultaneously Integer and Floating point execution, it packs far fewer CUDA cores compared to its RTX brothers, and also doesn’t have Ray Tracing or Tensor Cores. This means that the GTX 16 cards lack the Ray Tracing, DLSS and other advanced Turing features.
Then there’s the RTX 20 series, which starts at RTX 2060 and currently the highest-end SKU is the RTX 2080 Ti. Nvidia did a ‘refresh’ of sorts of the RTX 20 lineup by adding the RTX 2060 Super, the RTX 2070 Super (which we’ll soon have an analysis and review of up), and finally the RTX 2080 Super.
The latter two cards replace the vanilla cards outright, and the RTX 2060 Super acts as a halfway house in performance between the RTX 2060 and the RTX 2070.
Nvidia also has the Titan RTX, which is a ‘prosumer’ card, aimed at content professionals or those who run say physics simulations. Then there is are the Quadro GPUs. There’s often confusion between the difference of the RTX Titan and Quadro, as in the case of the Quadro RTX 6000 and the Titan RTX both have 24 GB VRAM.
The difference is that the Quadro card has ECC (Error Correction memory) and also slightly different drivers, the Quadro RTX 8000 though has 48 GB VRAM.
Either way, the fact is that the RTX 2080 Ti, the RTX Titan and the RTX Quadro all use variants of the same Turing core, the TU-102.
The Titan uses TU102-400-A1, the RTX 2080 Ti uses TU102-300A-K1-A1 and both the RTX Quadro cards use TU102-875-A1.
The GeForce RTX 2080 Ti though does have fewer enabled CUDA cores, 4352, vs 4608 CUDA cores.
Okay, now I’ve used a good portion of the article discussing the differences between GeForce, Quadro and finally Titan, so what’s the actual point of the article?
A new TU102 core has been spotted on AIDA64, the core is (and yes, the small ‘n’ is how it is written) ‘nVIDIA GeForce RTX T10-8 (TU102)’ though details of the specifications of the GPU are scarce. To my knowledge, this has been the first time that we’ve heard of a GPU core called ‘RTX T10-8’ (TU102).
Clearly, with the name TU102, it is definitely Turing and based on the same core as found in the Quadro, RTX 2080 Ti and RTX Titan.
It’s possible Nvidia are creating an RTX 2080 Ti ‘Super’ card, though they have gone on record and said there weren’t any current plans to do so (TweakTown asked Nvidia and that was their response). But plans can change. An RTX 2080 Ti Super could have more CUDA cores (upping the number to 4608 instead of 4352), or potentially a higher clock frequency too.
The new Super cards have all sported higher speeds compared to their vanilla cousins, the RTX 2060 Super had a 105MHz base frequency increase (though only 30MHz boost), while the RTX 2070 Super runs at 1605 base and 1770 boost. That’s compared to 1410 and 1620 of the base and boost speeds of the ‘vanilla’ RTX 2070, despite more CUDA cores being present.
| RTX 2080 Super | RTX 2080 Ti | RTX 2080 | RTX 2070 Super | RTX 2070 | RTX 2060 Super | RTX 2060 | |
| CUDA Cores | 3072 | 4352 | 2944 | 2560 | 2304 | 2176 | 1920 |
| ROPs | 64 | 88 | 64 | 64 | 64 | 64 | 48 |
| Base Core Clock | 1650MHz | 1350MHz | 1515MHz | 1605MHz | 1410MHz | 1470MHz | 1365MHz |
| Boost Core Clock | 1815MHz | 1545MHz | 1710MHz | 1770MHz | 1620MHz | 1650MHz | 1680MHz |
| Memory Clock | 15.5Gbps GDDR6 | 14Gbps GDDR6 | 14Gbps GDDR6 | 14Gbps GDDR6 | 14Gbps GDDR6 | 14Gbps GDDR6 | 14Gbps GDDR6 |
| VRAM Bus Width | 256-bit | 352-bit | 256-bit | 256-bit | 256-bit | 256-bit | 192-bit |
| Total RAM | 8GB | 11GB | 8GB | 8GB | 8GB | 8GB | 6GB |
| Single-Precision | 11.2 TFLOPS | 13.4 TFLOPs | 10.1 TFLOPS | 9.1 TFLOPS | 7.5 TFLOPS | 7.2 TFLOPS | 6.5 TFLOPS |
| Total TDP | 250W | 250W | 215W | 215W | 175W | 175W | 160W |
| GPU Core Name | TU104 | TU102 | TU104 | TU104 | TU106 | TU106 | TU106 |
| MSRP (Launch) | $699 | $999 | $699 | $499 | $499 | $399 | $349 |
Indeed, the RTX 2080 vanilla and the RTX 2070 Super both use the TU102 core, compared to the TU104 of the RTX 2070 (that same core is no longer used for RTX 2070s of course, but was also present in the RTX 2060 and now also used in the RTX 2060 Super.
Despite the RTX 2070 Super using the same TU102 core, core clocks were increased to 1770 MHZ from 1710. Nvidia did the same for the RTX 2080 Super too. It still uses TU104, but RTX 2080 Super runs at 1815MHZ over the 1710MHz of the older RTX 2080.
So one possibility here is Nvidia will simp0ly crank the clocks up and possibly enable the full 384-bit bus (rather than 352-bit of the RTX 2080 Ti). We also saw the RTX 2080 Super outfitted with 15.5gbps GDDR6 memory too, while all of the other cards run with 14gbps. It’s possible Nvidia could raise bandwidth by simply using faster memory.
Many of these possibilities also make sense for the Titan and Quadro cards too – Nvidia releases GPUs with higher clock speeds (now that Turing’s yields have improved) and/or outfitting the cards with a different memory configuration too.
There are also several other things of note in the AIDA64 Beta – here’s a list of everything that’s listed.
Release notes:
- CPU TDP limit detection for 2nd Generation Epyc
- Intel Processor Number detection for Core i3-1000G1, 1000G4
- Intel Processor Number detection for Core i5-1030G7, 1030G4
- Intel Processor Number detection for Core i7-1060G7, 1068G7
- sensor support for Dell SMI of Vostro 5581
- motherboard specific sensor info for Asus TUF B360-Plus Gaming S
- improved motherboard specific sensor info for ASRock boards
- GPU information for nVIDIA GeForce RTX T10-8 (TU102)
- fixed: RGB LED / row of keys background color
- fixed: motherboard specific sensor info for MSI X299 Series
Unfortunately nothing for the 3rd generation ThreadRipper that we know is coming, but there is lots of info for Intel’s 10th generation processors and hey, if you have an AMD Rome system you can now detect the CPU TDP limit!
It will also be fascinating to see how Nvidia responds next year with their GeForce 30 series GPUs, which is said to be produced on the 7nm process from Samsung. Given the company will be facing off against a potential ‘Nvidia Killer’ with Navi 20 series and also the mighty Intel getting in on the graphics market with Xe, 2020 will be a very cool year.
Thanks to viewer Jason for emailing the AIDA64 link and the tip!
