NVidia RTX 2080

[Nikos]

Oh man.
I really really don't understand what are you talking about.

Nvidia’s Director of Technical Marketing, Tom Peterson, joined HotHardware on its 2.5 Geeks podcast and when queried about the RTX 2080 outperforming the GTX 1080 Ti, he said that he thinks there would be cases that would happen but couldn’t say for sure.

So, what better proof than nVidia itself ?

The clocks and the number of CUDA cores have already
been published.
Just do the math.

2070 is slower than 2080
2080 is slower than 1080 Ti

2080 Ti has more than double the price of 1080 Ti and it's around 30% faster than 1080 Ti.

TDP is already published.

What further proof ?

Rumors regarding Vega20 are talking about a 20 TFlops beast.

Regarding Ray Tracing is not accurate what you are saying.

RTX cards do not have real Ray Tracing.

That's impossible for the current and many future generation cards because it needs PetaFlops not Teraflops.

nVidia provides hybrid Ray Tracing, be careful with the difference.

AMD doesn't need to provide support for such a feature that noone really supports right now.

It's another trick from nVidia like PhysX, HairWorks, GameWorks etc.

The usual tricks that nVidia does in order to kill the competition with "unique" features that eventually die.

AMD has the gaming consoles market so it's very difficult for nVidia to force such changes.

The killing feature of Turing cards is really non existent in games of today and you really don't know the future.

[Flomac]

boy, I don't know how did we live all these years without this almighty raytracing feature. medieval!

Yep, I know these comments for about 30 years now. "Who needs 3D-acceleration? 2D looks just fine! Where is my ET4000?"

Raytracing looks stunning in FPS. The RTX 20180Ti is already sold out at NVidia. If people want it, they buy it.

Another important fact: It makes the work of grame graphics designers easier.

A good counter-argument: computer games tend to be driven by console hardware and none of them has RTR as a feature. Yet.


No matter what: Turing is faster than Pascal and at the same price the choice would be simply clear. There is no real competition from AMD so NVidia can get a nice chunk back from their R&D work.

Maybe raytracing ends up like PhysX in a few games. But technically is was the only logical step forward for better game graphics.

For hashcat all that doen't matter. Let's wait up the benchmarks and then we'll see what GPU has the best p/p-ratio.

[Nikos]

Turing is a lot slower regarding price to performance ratio.

It will be faster in absolute performance, but with a huge premium in price not worthy to make the move from Pascal to Turing.

Maybe next generation 7nm nVidia cards that will be available in less than a year (my guess) due to the absolute disaster of Turing cards could do Ray Tracing better.

But 7nm Vega20 will be out at the end of the year.

OK, let's wait for real numbers of Turing in games and hashcat.

[Flomac]

The clocks and the number of CUDA cores have already
been published.
Just do the math.

2070 is slower than 2080
2080 is slower than 1080 Ti

2080 Ti has more than double the price of 1080 Ti and it's around 30% faster than 1080 Ti.

TDP is already published.

What further proof ?

Are we talking about games or hashcat? Because for hashcat, many more things are important than TFLOPS. Take a look at the GTX980. It had less cores, but higher clock rate than the GTX 780 Ti, ending up with the same TFLOPS. So, following your logic, same performance under hashcat and a big flop? I guess you know the answer.
A new instruction set, different core design and bigger caches did it for Maxwell.

The TDP includes hundreds of Tensor cores. When they are not being used, TDP will be lower. When they can be used for matrix operations, performance will be higher.

The price is always relative and can be lowered if needed. But: no competition, no price change.

About AMD: The actual Vega already has more TFLOPS than Pascal, and so did the Fury over Maxwell. Didn't help at all. Since generations, AMD has a big heat and power problem under hashcat. In gaming, their clever algorithms keep the GPU cool enough. But with hashcat running all 4096 cores at the same time and blowing >300W Vega doesn't stand a chance against the GTX 1080Ti with less cores.

So, again, we have to sit and wait.

[Nikos]

But I have already told you that Vega20 is on 7nm architecture with very low TDP and thermals and high clocks.

TSMC rules here.

The prices are not going to go down for Turing cards because there is a lot of Pascal stock to be sold.

Also Turing cards are very expensive dies.

Tensor cores and Ray Tracing cores can't be used by Hashcat.

Anyway, we are repeating ourselves.

Let's wait for the results.

[Flomac]

But I have already told you that Vega20 is on 7nm architecture with very low TDP and thermals and high clocks.

And you know that because you already saw one? Or because AMD told so like they did the last five years?
Rumors say it stays at 4096 cores and there will be no changes in core design, just Vega 10 in 7nm (a classic AMD move).

TSMC rules here.

No actual 7nm product available. No yield known. Looks promising though.

The prices are not going to go down for Turing cards because there is a lot of Pascal stock to be sold.

OEM products were sold out for months. But yes, prices should be stable at least till next year.

Also Turing cards are very expensive dies.

True indeed.

Tensor cores and Ray Tracing cores can't be used by Hashcat.

Speculative.

Anyway, we are repeating ourselves.

Let's wait for the results.

D'accord

[GrepItAll]

I have been running 4 x 1080 Ti Asus Turbo cards since September nearly 24/7 (brief downtimes due to hardware relocation) as it was very difficult to purchase multiple new FE cards at the time. Been very happy with their performance so far, both in terms of cooling and H/s.

I'll be interested to see how the 2080 Ti Turbo holds up.

I think statements like this must be at least accompanied with the details of the cooling solution. Are you really using a server case where one GPU is just few millimeters away from the next GPU? Does the fan of one GPU almost touches the next GPU like for most server/retail mainboards ? .... or are you instead using risers and the GPUs are "miles" away from each others etc
These are very different ways to setup rigs and of course you can't really say that the fan/cooling of a GPU is very effective/good (for server/retail mainboard setups without risers) if you use risers for your setup.

I'm not saying that I'm sure that you are using risers and this is the only way in which that specific model (I don't own this model therefore I have literally "no clue" ) can be cooled within a multi-GPU setup... but I think without the full details about your cooling strategy (which type of motherboards/case/fans/risers, how many inches/mm the GPUs are apart from each other etc) the statement could be a little bit incomplete/dangerous/misleading etc. I would like to know the details, because it could be also a very good fan designed by ASUS (again, I have no clue, because I don't own this specific model).
Please let us know about the details of your specific cooling strategy (and/or how you setuped your GPUs)
Thanks

Sorry for the late reply! I don't blame you for being sceptical re: risers etc, but I am using these in a server chassis. It's for business use, therefore an open-air rig was inappropriate (plus we wanted a system that we could get a warranty and service agreement on).

This is the system we have them installed in: https://www.supermicro.com/products/syst...8gr-tr.cfm

The system lives in a fairly well cooled server room. Mild overclocks applied to all 4 GPUs as well. I genuinely think that ASUS have done a decent job with their blower fan design.

The cards don't have the aluminium backing plates of the FE, which is good as improves clearance between the cards by a couple of mm.

I am aware of people in various organisations using the ASUS Turbo cards successfully in https://www.supermicro.com/products/syst...GR-TR2.cfm with good feedback from them.

[john]

1x2080ti or 2x1080ti? for hashcat

[GrepItAll]

1x2080ti or 2x1080ti? for hashcat

On release, 2x1080 Ti will certainly be better than 1x2080 Ti. 

In fact several other users have pointed out that even 1x1080 Ti is likely to be faster than 1x2080 Ti initially.

[Flomac]

1x2080ti or 2x1080ti? for hashcat

On release, 2x1080 Ti will certainly be better than 1x2080 Ti. 

In fact several other users have pointed out that even 1x1080 Ti is likely to be faster than 1x2080 Ti initially.

You mean 2x1080 Ti will be faster than 1x 2080 Ti.