What do you want in a GPU review?

[jeshikat]

I'm thinking of getting a GTX 1060 in the near future for a project but I'll probably review it for the site as well.

What do you all want to see in a review?

I'm thinking that exhaustive performance testing isn't really necessary for this audience since there a dozens and dozens of sites testing GPUs with various games and synthetic benchmarks. I figure you all would be more interested in detailed physical measurements, power consumption, etc.

 

[Kmpkt]

I'd personally like to see something along the lines of a performance per watt curve for under clocking to be honest. Since we're really pushing the limits on performance with limited constraints on power and size, I think finding the sweet spot on the hardware would be really cool. Would certainly add a very different flavours to reviews here and help separate this website from the crowd.

 

[jeshikat]

I'd personally like to see something along the lines of a performance per watt curve for under clocking to be honest.

That's a neat idea. What would you prefer for measuring the performance part? 3DMark? Game benchmark?

 

[Kmpkt]

I've always been a fan of real-world benchmarks over synthetics. Depending on your access to games, this can obviously be an issue. Tomb Raider 4, Ashes of the Singularity, Witcher 3, etc. would be the obvious choices. A very simple way to do this (and functional) would be with and without VSync at say 60, 75, 120 and 144Hz (depending on appropriateness obviously)

Along these same lines, I'd be really curious to see a similar review done on some of the more popular CPUs like the 6700K or 4790K etc if the opportunity ever presents.

 

[jeshikat]

I'll check into games then. It'll have to be games that have built-in benchmarks though.

And what do you all think of open air versus testing in a case? I lean to open-air for consistency and to establish a baseline of performance.

 

[Phuncz]

I'd also like to see atleast one test measuring performance or atleast clock frequency for atleast half an hour with 100% gaming load, like a benchmark on a loop without long loading times. AMD and Nvidia abuse boost clocks a lot these days, because reviewers often don't run tests for much longer than a minute, while gaming usually means an hour or more.

DX12 and Vulkan are relevant, but don't focus too much on 3DMark, as they basically made their DX12 test as a DX11 one.

 

[Kmpkt]

I think closed case would be the fairest way to benchmark for this crowd. Looking at cases like the Dan A4 and NFC S4 mini as well as potential up and comers like Hutzy, Freilite and MI-6 and none of them have anything more than passive airflow through a side panel (sorry if I am mistaken on any of these, this is the top of my head). An open test bench isn't going to fairly reflect real world performance in the SFF community.

 

[GuilleAcoustic]

I'd say:

• openCL GPU rendering
• openGL performances

 

[jeshikat]

I'd also like to see atleast one test measuring performance or atleast clock frequency for atleast half an hour with 100% gaming load, like a benchmark on a loop without long loading times.

Good idea.

An open test bench isn't going to fairly reflect real world performance in the SFF community.

True, but every case is different too so what I use would still only reflect those using cases with a similar layout. And it looks like isolated power measurement of the card itself won't happen anytime soon due to test equipment costs, but if I do get those kind of tools it'll be difficult with the card inside a case.

• openCL GPU rendering

• openGL performances

Any particular benchmarks or games you prefer for those?

 

[Kmpkt]

Probably an easy way to standardize the bench is use a GPU with a card only wattage test elsewhere and redline it to find out the rest of systems draw. From there you can reasonably infer card wattage reasonably well in the future.

 

[jeshikat]

I guess for that matter I could measure the at-the-wall draw of the motherboard and system while powering the GPU externally using a powered PCIe extender.

The case thing I'm torn on. On the one hand I test in a case that's at least representative of some users but makes testing more difficult. On the other hand I could test in open air and make testing easier, it would be unrepresentative but at least it'd be equally unrepresentative.

Anyone else have thoughts on that?

 

[Kmpkt]

Honestly as long as it's consistent from test to test we can draw fair comparison from card to card. If it's an open bench, I think the community will probably eventually come to a consensus that a closed case scenario will probably add X degrees to the operating temps of the GPU.

 

[iFreilicht]

Alongside what has been said: Dimensions, as described here:

 

[veryrarium]

Are you getting a 1060 with a blower-fan-equipped cooler such as the F.E. or with an axial-fan--equipped open-air cooler? I suppose you're getting the latter type as that's what would make you concerned about the open bench vs the closed bench choice more than the former type.

 

[jeshikat]

Alongside what has been said: Dimensions, as described here:

Any other dimensions or just those? I was thinking measuring both the shroud and PCB would be useful for those wanting to either water cool or replace the shroud.

Are you getting a 1060 with a blower-fan-equipped cooler such as the F.E. or with an axial-fan--equipped open-air cooler?

Axial. I want one of the EVGA 1060 Mini cards.

 

[Soul_Est]

--snip--

Any particular benchmarks or games you prefer for those?

Blender Cycles renderer with the test .blend file (https://blenderartists.org/forum/showthread.php?239480-2-7x-Cycles-benchmark-(Updated-BMW)) and LuxMark can be used to benchmark OpenCL performance. The Phoronix Test Suite can be used to test various OpenGL games under Linux. I am not sure of an analog on Windows I'm afraid.

 

[Kooki]

The performance per watt curve is quite interesting idea!

Measure the card dimensions is also a really good idea. I bet everyone hates the lack of standard on the measurement of the cards dimensions (recalling little article wich ranted about the lack of "itx gpu" standard).

Is there anyway to measure the efficiency/importance of a design like the mini itx semi-blower axial evga gtx 1060 to remove hot air from a small case in any significant way?

 

[jeshikat]

Is there anyway to measure the efficiency/importance of a design like the mini itx semi-blower axial evga gtx 1060 to remove hot air from a small case in any significant way?

Not easily. Measuring the effect on CPU/HDD/mobo/etc. temp would work but it would be highly dependent on the case and fan layout.

 

[iFreilicht]

Any other dimensions or just those? I was thinking measuring both the shroud and PCB would be useful for those wanting to either water cool or replace the shroud.

PCB, too. If the shroud can be removed without the heatsink, then heatsink on its own as well. Good that you reminded me about that

 

[K888D]

As other have mentioned, some of the things I would like to see:

• The real the dimensions of the card including the shroud
• Idle Fan Speed and the dB reading at this speed
• Max Fan Speed of the card and the dB at this max speed
• Under stock settings the Fan Speed and temp it sits at whilst stress testing (with dB also)
• Under stock settings the Fan Speed and temp it sits at whilst gaming (with dB also)

The above readings should create an overall picture of the cards cooling performance.