NVidia RTX 2080

[Flomac]

The new generation of NVidia GPUs will be presented on the 20th of August at Gamescon. Preorder starts the same day, delivery is expected in the following 7-10 days.

Codename of the graphic chips is Turing, named after the famous computer scientist Alan Turing, who developed the "Bombe" that cracked the notorious Enigma-code. He also invented the Turing, a mathematical model of computation that defines an abstract machine, which manipulates symbols on a strip of tape according to a table of rules

Its name change from 'GTX' to 'RTX' is reasoned in the additional _R_aytracing cores the new generation is housing, which make raytracing available in real time. The price will start somewhere below the GTX 1080Ti and well above the GTX 1080.

It also looks like that they will launch the Ti version right after or maybe even at the same time. There are definitly already samples out there, so maybe NVidia wants to bring this card well before christmas season.


Safe tech facts:
RTX 2080: 8GB GDDR6 RAM with 3500MHz, core clock ~1515-1525 MHz
RTX 2080 Ti: 11GB GDDR6 RAM with 3500MHz, core clock ~1350MHz


Speculated facts:
RTX 2080: 2944 cores, 184 TMUs und 64 ROPs; also 3072 cores possible; TDP 210W
RTX 2080 Ti: 4352 cores, 272 TMUs und 88 ROPs; TDP >250W


It's rumored there is a huge overclocking headroom and custom cards with more than 2GH should be  available, while manual OC might reach up to 2,5GHz. That would be quite impressive.

[MattSR]

Can't wait to see the benchmarks on this one (for hashcat of course!)

[atom]

I need to do add them to tuning database, but this requires me to have an clinfo output first. If anyone has it, ping me.

[Flomac]

The new cards are out.

1st Surprise: There is also a RTX 2070

2nd surprise: The card design changend completely. Two axial fans are supposed to lower the noise level to 1/5th of a GTX 1080Ti. Bad for multi GPU builds.

3rd surprise: realse date will be September 20th. Preorder starting today.

The cards make real time raytracing in games available and it looks absolutely stunning. No question this is the biggest step in computer game graphics of the last years, thanks to the new tensor cores which deliver brutal performance

The biggest question for Hashcat will be: Can atom & Co. make these migthy tensor cores usable for hashcracking?


Final specs:
RTX 2070: 2304 cores, 8GB RAM, 1410MHz base, 1620 MHz boost, 1710 MHz (OC)*, TDP 175W
RTX 2080: 2944 cores, 8GB RAM, 1515MHz base, 1710 MHz boost, 1800 MHz (OC)*, TDP 225W
RTX 2080 Ti: 4352 cores, 11GB RAM, 1350MHz base, 1545 MHz boost, 1635 MHz (OC)*, TDP 260W

*=NVidias FE (Founders Edition) comes with onboard overclocking

[Flomac]

Bye bye radial fans...*




* there seem to be at least radial designs available by EVGA and others

[wf2]

So what to expect from those ray tracing capable graphic cards? gtx 2080ti costs more than 1000$, in Germany it would be more than 1100€. Isnt it better to get 2x 1080ti for this money? CEO says rtx 2070 is 5x faster than titan xp in raytracing. But we dont care about raytracing, but it may be attractive on those new family is the overclocking. I think new rtx 2080 would be  5-15% faster than 1080ti, not more. otherwise price policy doesnt make any sense.

[Flomac]

Hard to say. On paper, the RTX 2080 should be ~10% slower than a GTX 1080 Ti.

But the tensor cores have a 10x higher output in FLOPS than the traditional GPU cores. That would be a HUGE step forward, although the question is: How good can you make use of these cores under hashcracking? There are way less tensor cores on each card than traditional ones, but they are also much more powerful. Everything else is speculation so far.

[soxrok2212]

In regards to gaming, great. In regards to everything else, im pretty disappointed. Nvidia basically screwed over everyone else with the physical design of the cards. No more blower means hot cases and hot parallel cards. This design won’t be great for hashcat.

[undeath]

But the tensor cores have a 10x higher output in FLOPS than the traditional GPU cores.

hashcat doesn't use any floating point operations.

[Flomac]

But the tensor cores have a 10x higher output in FLOPS than the traditional GPU cores.

hashcat doesn't use any floating point operations.

FLOPS meant as a universal measering method to communicate the comparable strength of a processor unit.

The tensor cores are also used for deep learning or voice analyzing and are capable of operating 4x4 matrix operations at once. Should be useful in some way.