1080 TI - new best bang for buck?

[Flomac]

I'm sorry, but whatever you predicted five years ago, you're ignoring what's going on today with AMD. I'm not a fanboy or anything like that. I'm just following the market since 25 years and actually, the signs are pretty much in favor for AMD - to not be swapped away from the market.

Everything started with the Polaris GPU, which is a solid middle class card for middle priced PCs. With Vega they have a very promising high end GPU in the pipeline. And Ryzen seems to be a big leap forward, at least it's very competitive, at a reasonably price.

Naples in the server market is an even bigger change. I don't know how much involved you are in this particular market, but the Server 2016 license conditions are in no way fitting the Intel CPU range (price, cores, clock rates). I't a nightmare for the IT advisor or system architect, because those CPUs making most sense in a system have ridiculous high price tags. And a server CPU with 32 cores? Not available from Intel. Companies don't care what kind of CPU their VMs are running on. As long as the big vendors (HP, Dell, Fujitsu etc.) deliver drivers and service they are TCO sensitive.
Microsoft already has announced they're gonna use Naples CPUs in their Cloud enviroment. At the end of this year, the server market should look completly different in terms of prices.

The AMD share today is almost six times higher than one year ago. The revenues are steady upwards. The signs are there.

They were in need of delivering and if wouldn't they'd be swapped away by the end of this year. But it looks like they do have a competitive line up on every market and that's good for everyone.

[r9290xocl]

Flomac: Till today, AMD still has issues with their drivers. Not sure what it would take for that to change! I have an R9 290x and am planning to switch to Nvidia for 2 reasons: Drivers and ULMB which comes with G-sync.

Vega sounds real good on paper.. There are 2 questions: how much will it be vs the 1080 (which dropped in price recently), and will its drivers be up to the task?

[Flomac]

That's what I've mentioned above, drivers/software are a big issue for the graphics market. Not sure how they'll come out of that.

In the server market drivers never where a problem, since most crucial components (RAID-controller, NIC) came from 3rd party vendors.

And I think the price of the Vega GPU in full spec will be around the 1080Ti, depending on RAM and performance. There is a smaller version in the pipeline that might cover the GTX 1080.

[r9290xocl]

imo a best bang for the buck should also consider other factors like temps, power consumption, time it takes to crack, reliability, and so on.

Temps of the GPU do not matter. If the system is stable and you don't care about the noise, turn the fans up and everything is fine. If you have a Founders Edition and 3 years warranty you simply cannot overstress it, since the power limit is setting the right limits.

Power consumption is always an important point and correlates with the amount of heat produced by one card. The latter may only matter in bigger environments like server rooms, but energy can be a serious factor in 24/7 hash cracking. In Germany a rig with 4x GTX1080 Ti and an efficient PSU easily consumes 5 to 6 bucks a day.

With NVidia, every new generation uses less power for the same performance as the old one. An old GTX980Ti e.g. is slower than a GTX 1080, but uses 40% more power. So new is always better in energy terms. And since the performance of a GPU generation scales very good under hashcat. 30% more performance result in 30% more energy used.
A big difference in systems can be the PSU, so if you plan to heavily crack hashes 24/7 you better invest in a Titanium or Platinum PSU instead of a Gold standard (and don't even think below that anyway). The higher price will be drawn in over time, even at only two percent difference in efficiency.

Reliability depends on many factors, but along with NVidia the reference cards respectively Founders Editions are pretty much in the same ballpark. One might have better experience with a special manufacturer, but from my experience the more important question is: how fast and uncomplicated do they change a broken card?

The time to crack can be crucial for some if you have limited time sets to be played. E.g. checking passwords in an ISO27001 audit usually defines a certain amount of time, like 24h, where the quota of cracked password specifies the quality of the staffs password creating ability. The more cracked the better.

You kind of reinforced my point: best bang for the buck is not just about hash/$