1080 TI - new best bang for buck?
#1
https://www.cnet.com/news/nvidias-new-gp...0-cheaper/

10th March release date. I'm personally looking forward to the release :-)
#2
12B Transistors
1.6 GHz Boost, 2 GHz OC
28 SMs, 128 cores each
3584 CUDA cores
28 Geometry units
224 Texture units
6 GPCs
88 ROP units
352 bit GDDR5x
11 GB GDDR5x

as a side effect the msrp of the 1080 drops by 100$ to 499$
sch0.org
#3
Im not sure but I think amd r9 furyx still best.
#4
(03-01-2017, 01:56 PM)kamsulla Wrote: Im not sure but I think amd r9 furyx still best.

it never were
#5
R9 Fury X was never the best.

GTX 1080 Ti and GTX Titan XP should have clock-for-clock identical performance. But it won't have the best bang for the buck. If we look at MD5 perf per dollar:

GTX 1070 - 53 MH/$
GTX 1080 - 50 MH/$
GTX 1080 Ti - 44.29 MH/$

So at $349, the GTX 1070 is the best bang for the buck.
#6
For a single gpu system the Fury x isn't bad, but if you intend to run the gpus in a datacenter then water cooling is out, nor could you fit more than a couple at most.

I also take the system cost into consideration, a 4U system without gpus has a certain base price and that could be a non trivial cost if you are limited by rack space.
#7
(03-01-2017, 05:11 PM)epixoip Wrote: So at $349, the GTX 1070 is the best bang for the buck.

Yes and no. If you compare single cards, sure. But these cards work in a whole system, so we have to compare system prices instead of single card prices.

E.g.: add $600 for a double slot system and calculate again with 2 GPUs:

GTX 1070:  28.54 MH/$
GTX 1080:  31.25 MH/$
GTX 1080 Ti: 31.0 MH/$

So a standard 1080 is the best and the 1070 suddenly the worst bang for the buck.

For more or less GPUs it goes the same.
#8
Well... Yes and no Wink

You're correct: you should take the total build cost into consideration. There's a reason we do not offer the GTX 1070 in the Brutalis in a clustered configuration, it throws the Perf/$ and Perf/Watt ratios for the cluster way off.

On the other hand, it's not that simple unless we have a standard baseline build to compare against. Like, the Brutalis!

Code:
N4X48-GTX1070-2620-128-2X500       20389.00 USD      147.1 GH/s     7.21/MH/$
N4X48-GTX1080-2620-128-2X500       21589.00 USD      200.3 GH/s     9.28/MH/$
N4X48-GTX1080TI-2620-128-2X500     22789.00 USD      243.7 GH/s    10.69/MH/$

So in this case, the GTX 1080 Ti would be the best bang for the buck, not the GTX 1080 Wink
#9
You're absolutly right and I just wanted to point out the importance of the system costs. We could also bring power consumption and manpower to the TCO-level, but this far we can say just comparing pure GPU prices does not paint the picture.

And by the way I made the mistake and used a $500 price tag for the GTX 1080 (like you in your first calculation?). The all recommended F.E.-edition comes for 50 bucks more, which changes the calculation. Same for the GTX 1070 ($400 instead of $350).



To point out the calculation:

System:        $600
GTX 1070FE: $400
GTX 1080FE: $550
GTX 1080 Ti: $700

System with 2 GPUs each:
GTX 1070FE:  26.43 MH/$
GTX 1080FE:  29.41 MH/$
GTX 1080 Ti:  31.0 MH/$



So I guess the Ti will be the best choice in most cases, except when your budget is tight or you decide to choose really cheap hardware for the system. But you can always do the maths and make a choice on your own.
#10
Yeah I didn't realize until later that the price points Nvidia advertised were for OEM cards, not FE. We used FE prices when calculating the new Brutalis price points.

However, if you're simply looking to upgrade your existing GPUs, then Perf/$ of the bare card should be taken into consideration.