GPU Testing/Reviewing techniques and Runt Frames with Crossfire

[CAT-THE-FIFTH]

Yeah true. Guess we will just have to wait and see!

I would quite like to see the affects of the APU with a discreet card crossfire. Mate of mine just build a lovely 5800K system and is thinking of picking up a £50(ish) 6000 series card to crossfire it.

Last time I check the scaling was a bit all over the place,but those are older reviews. Supposedly someone got an HD7750 to work in Crossfire too,although it could have photoshopped.

AFAIK,the latest Richland APUs have a more temperature dependent based Turbo too IIRC,so I really hope that review sites test the games for longer periods,ie,over a minute instead of 30 seconds,especially with the aluminium cooler of fail!!

[Biscuit]

'ere, that stock 7950 is pretty good it hits 300 FPS at 5760x1080 on Crysis 3!

http://www.pcper.com/reviews/Graphic...-7950/Crysis-3

But shouldnt the 7950 actually outperform the stock 660ti more generally?

[Willzzz]

No.

On the same game:
http://www.techspot.com/review/642-c...nce/page4.html

Over multiple games:
http://techreport.com/review/23981/r...i-revisited/11

They are generally pretty close over a large number of games.

But I am not that convinced by the PCPer data, much more convinced by the techreport stuff and looking forward to updates from Anand and Toms.

[kalniel]

Well the thing I'm taking from the pcper article (27th march) is just how the 7970 is basically a slightly better card than the gtx 680. Granted, the text is concentrating on the big performance problems of crossfire, but I'm not so interested in dual card setups.

PCPer themselves say that animation smoothness is the most important thing, so the bottom line on these graphs is the best card:


(ignore CF line for that one)

Crysis3 1080p = tie

(though too slow for meaningful use)





triple display = tie



(again, ignore cf line)

sleeping dogs 1080p = tie
sleeping dogs triple screen = tie

Only Far Cry 3 gives a worse result for the 7970 compared to the 680,.

edit: and to their credit, pcper echo this in their conclusion.

The overall picture comparing the two cards indicates that the AMD Radeon HD 7970 GHz Edition is a faster card for gaming at 1920x1080, 2560x1440 and 5760x1080 triple-monitor resolutions.

http://www.pcper.com/reviews/Graphic...ormance-Tes-12

[Willzzz]

I generally prefer the FPS percentiles, but there is little doubt that the 7970 is the higher performer in the single card stakes.

A lot of the 5760x1080 tests are pretty meaningless as even the best performing setups are sub 30 FPS for large sections. Neither 680 SLI or 7970 are playable.

[kalniel]

A lot of the 5760x1080 tests are pretty meaningless as even the best performing setups are sub 30 FPS for large sections. Neither 680 SLI or 7970 are playable.

Yeah I agree, just included them to show the range.

[jigger]

I would rather have the testers set a minimum frame rate and omit any results that don't make it. Having cards grouped by playable resolution would give everyone a much clearer idea of what performance is on offer.

Most of the review sites nowadays have 5-10 sets of results at various resolutions ranging from single figure frames a second to 200 plus. Everything under 40 and over 80 FPS is pointless IMO.

[kalniel]

I would rather have the testers set a minimum frame rate and omit any results that don't make it. Having cards grouped by playable resolution would give everyone a much clearer idea of what performance is on offer.

Most of the review sites nowadays have 5-10 sets of results at various resolutions ranging from single figure frames a second to 200 plus. Everything under 40 and over 80 FPS is pointless IMO.

Well the point of this discussion is saying that aiming for a minimum frame rate doesn't help determine gamer experience, because frame rate doesn't.

[jigger]

Frame rate is the single most important factor when it comes to a good experience though.

I agree FRAP's is far from ideal but bar actually watching the results of a review first hand you can't get a understanding of micro stutter or badly drawn frames. I would say if your happy gaming in the 30fps range you probably wouldn't even notice an issue.

[Willzzz]

I'm pretty happy with the way reviews are going. I'm pretty sure that by the time I'm ready to buy my next card I'll have all the information I need to make an informed purchase.

[kalniel]

Frame rate is the single most important factor when it comes to a good experience though.

Well the pcper article argues otherwise. A consistent frame rate is more important than outright higher one. Obviously there are extremes, but that the gist of what the the article is about.

[jigger]

I just don't see much value in micro measuring frame rates to the point of margins of error, or to the point of when your readers mouse, monitor or DVI could be a factor. It makes even less sense if the test is done within pointless margins to begin with.

If one card looks better at 25FPS than another then great, but I wouldn't buy either as they are both still useless.

[Willzzz]

Jigger, I don't think you get this. Those things are always a factor regardless of how you measure FPS.

Average FPS is useless, you might get 100 fps normally but if there is some spike and you die then it won't feel nice.

Yeah obviously, any card with an average 25fps is useless, but if you have two cards with 40 fps and one goes between 20 and 60 and the other between 30 and 50 then it's obvious which is the better card.

[jigger]

I too hate results based on average frame rate, and yes they are useless as they prompt issues as per the thread. For me in the average gaming rig set up (60hz LCD monitor at native resolution all eye candy on) a card needs to hit between a minimum of between 40 and 60FPS for most games to be enjoyable, with a constant 60FPS being the holly grail in that situation.

What I'm saying is. If a card can't run a game *smoothly* then simply don't list the results for that test regardless of the average frame rate. Review sites using a higher minimum frame rate cut off would be much more beneficial to improving PC gaming. As things are now we just see a lot of data and most of that is misleading to many peoples perception of a cards performance. I would rather just see meaningful results across various resolutions than see testers add anything more to reviews that already read as a pile of waffle.