Asking Ryan Smith of AT if a special examination could be done?

[Keysplayr]

Hey Ryan,

Big wars are a brewing over this Nvidia is smoother than AMD at similar framerates thingy.

Would you be interested in doing an examination and report on this?
So far, HardOCP and TechReport have reported on this phenomena and I'd think it wouldn't hurt if Anandtech did some digging.

Thanks!
Please let us know if this is a problem.

 

[blastingcap]

Nothing short of using high speed cameras aimed at computer monitors will work at capturing end-user experience in an objective manner. FRAPS records frametimes at a point in the graphics pipeline prior to the image reaching the monitor. Use multiple monitors if you're worried about that skewing results. We've been over this already. TechReport itself mentioned it:

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking/11

"The slide above shows the frame production pipeline, from the game engine through to the display, and it's a useful refresher in the context of this discussion. Things begin with the game engine, which has its own internal timing and tracks a host of variables, from its internal physics simulation to graphics and user input. When a frame is ready for rendering, the graphics engine hands it off to the DirectX API. According to Petersen, it's at this point that Fraps records a timestamp for each frame. Next, DirectX translates high-level API calls and shader programs into lower-level DirectX instructions and sends those to the GPU driver. The graphics driver then compiles DirectX instructions into machine-level instructions for the GPU, and the GPU renders the frame. Finally, the completed frame is displayed onscreen."

 

[krumme]

Hey Keys
For a moment looking at the headline i thought you would ask them not to test nv the same time as amd. But i can see you found the exception. Good digging.

 

[OVerLoRDI]

I would definitely be interested in this. I hope someone writes up an article about it.

 

[Jacky60]

I'd be interested to see this too and have to agree that high speed video of the event is the only thing short of my eyes I will trust. Not 'ooh the blonde haired girl somehow seemed kinda sexier' sort of impression. Also vsync needs to be applied, if you're spending $800-2000 on gpu's you really should be able to use and be using vsync I now know.

 

[thilanliyan]

I'm for this...but with a high speed camera as someone suggested. Only thing is, with a high speed camera, how do you know you are measuring the same frames on both systems since the camera is decoupled from the computer?

 

[HurleyBird]

Indeed, this would be a good test. Could you also test latency, to see what kind of lag AFR frame smoothing introduces for both companies?

 

[The Alias]

I'd be interested to see this too and have to agree that high speed video of the event is the only thing short of my eyes I will trust. Not 'ooh the blonde haired girl somehow seemed kinda sexier' sort of impression. Also vsync needs to be applied, if you're spending $800-2000 on gpu's you really should be able to use and be using vsync I now know.

not really v-sync introduces input lag in a major way .

also I suggest that they bench the cards with and without radeon pro as well

 

[cmdrdredd]

not really v-sync introduces input lag in a major way .

also I suggest that they bench the cards with and without radeon pro as well

I totally agree. In some games input lag is nearly unplayable to some people. I also think there should be tests using frame limiters(radeonpro) and without it. The difference could be considerable.

 

[Ferzerp]

Why use a camera and not just monitor the signal to the monitor itself if you aren't going to trust software readings?

(Or at least check to see if fraps is accurate, and then use that if found to be so).

 

[ICDP]

I totally agree. In some games input lag is nearly unplayable to some people. I also think there should be tests using frame limiters(radeonpro) and without it. The difference could be considerable.

I have yet to find a game that has input lag so bad with vsync enabled that it is unplayable. I suppose if someone games professionally then it might be a factor.

Of course you need to add triple buffering and play around with flip queue sizes, but it was always fixable for the majority. Now that adaptive/dynamic vsync is available it doesn't have to be a problem.

 

[Annisman*]

I have yet to find a game that has input lag so bad with vsync enabled that it is unplayable. I suppose if someone games professionally then it might be a factor.

Of course you need to add triple buffering and play around with flip queue sizes, but it was always fixable for the majority. Now that adaptive/dynamic vsync is available it doesn't have to be a problem.

Input lag is the only reason I never use Vsync is any game, ever.

Some people may not notice it, great, but it can be unbearable for others.

 

[Keysplayr]

Hey Keys
For a moment looking at the headline i thought you would ask them not to test nv the same time as amd. But i can see you found the exception. Good digging.

No idea what you're getting at.

 

[HurleyBird]

Why use a camera and not just monitor the signal to the monitor itself if you aren't going to trust software readings?

To test latency. The longer the driver holds frames back to rearrange them in a more even manner the more more input lag is added. The default assumption is that SLI incurs more lag than CF because Nvidia puts more emphasis on reducing micro-stutter, but it could be the opposite and CF is just less efficient at smoothing out frames. By now anyone with half a brain realizes that SLI has less stutter on average, what we don't know is what trade-off -- if any -- SLI users pay for the added frame smoothing versus CF users. If SLI doesn't have more input lag, then it's flat out superior. If it does then it's just different and which one works best will come down to individual preference. We can't say anything for sure without proper testing, but until we have actual data the smart money is on SLI introducing added latency over CF.

 

[cmdrdredd]

I have yet to find a game that has input lag so bad with vsync enabled that it is unplayable. I suppose if someone games professionally then it might be a factor.

Of course you need to add triple buffering and play around with flip queue sizes, but it was always fixable for the majority. Now that adaptive/dynamic vsync is available it doesn't have to be a problem.

When vsync is enabled when running adaptive vsync you have input lag. So no it's not gone or fixed. Triple buffering helps but doesn't eliminate it. There was an article right here on AT that tested exactly that. If I play BF3 with vsync on it's a totally different game to me and to be honest it sucks.

Off topic but I think all settings should be tested. Adaptive vsync, vsync, no vsync, frame limiters etc. Determine what difference each makes if any.

 

[blastingcap]

I'm for this...but with a high speed camera as someone suggested. Only thing is, with a high speed camera, how do you know you are measuring the same frames on both systems since the camera is decoupled from the computer?

Partial screens are okay too, especially if you are testing w/o vsync anyway, as you can infer the exact frame time to the closest 1/1200th second (if using a 1200fps camera). A fast enough camera can simply take shots far faster than the monitor can refresh, so you can get a decent idea of the true frametime even if you don't know it exactly.

Btw, they make cameras that can take shots at a million frames/second; granted they cost a fortune but you can probably rent them and not buy them, plus you don't even really need a million fps, even something like 1200 fps can probably do well enough... they do not need to 100% sync up to the picture drawn on the screen. Maybe even a Nikon 1 camera could do it, as they go to 1200fps in low-res mode with the latest firmware. Casio also has a high-speed camera that is affordable, the Casio EX-F1, also at 1200 fps.

Someone said to measure the signal to the monitor which I suppose is possible, but I'm not aware of hardware than can capture that data in a way that Ryan could use. If you are, let us know.

Edit to add: I think I misinterpreted your question; you seem to be thinking about taking one camera and filming two systems next to each other, but I thought you meant one system at a time, which I believe is how most reviewers review video cards (to make sure that all else is equal). I think it's better to do it sequentially but either way could work, in theory.

 

[Will Robinson]

Big wars are a brewing over this Nvidia is smoother than AMD at similar framerates thingy.

Oh really?...In whose mind?
Yours?
Perhaps its a focus group thing....

 

[Will Robinson]

Gamegpu found HDAO works better on AMD hardware and presented higher IQ.

They should probably look into the growing scandal about NVDA's image quality sacrifices in Far Cry 3 as well....

 

[lavaheadache]

I've got frametimes for ya'll. 7970 Lightning, 2560x1600 ultra settings no AA.

What is the best way to post them without making a super loooong post?



here are the preliminary results

2012-12-05 19:12:12 - farcry3_d3d11
Frames: 4423 - Time: 91167ms - Avg: 48.515 - Min: 38 - Max: 69

and a small sample of frametimes, I waited till the game was done loading and ready to be benched

Frame, Time (ms)
1, 0.000
2, 19.708
3, 41.276
4, 63.143
5, 84.565
6, 106.033
7, 127.538
8, 149.467
9, 171.014
10, 192.565
11, 213.882
12, 235.968
13, 257.911
14, 279.568
15, 300.969
16, 322.830
17, 344.023
18, 365.832
19, 388.203
20, 409.610
21, 431.489
22, 452.704
23, 474.168
24, 496.092
25, 517.309
26, 538.921
27, 560.617
28, 582.449
29, 603.422
30, 625.412
31, 646.514
32, 668.073
33, 689.479
34, 711.063
35, 732.735
36, 754.576
37, 776.043
38, 798.072
39, 818.969
40, 843.018
41, 861.868
42, 883.825
43, 905.018
44, 926.542
45, 947.706
46, 969.302
47, 989.867
48, 1011.659
49, 1033.039
50, 1053.341
51, 1075.391
52, 1096.599
53, 1117.756
54, 1138.566
55, 1160.179
56, 1180.727
57, 1202.329
58, 1222.778
59, 1245.034
60, 1265.673
61, 1286.865
62, 1307.871
63, 1329.400
64, 1349.830
65, 1372.211
66, 1392.262
67, 1414.470
68, 1434.775
69, 1455.693
70, 1476.529
71, 1497.943
72, 1518.081
73, 1538.942
74, 1559.724
75, 1581.000
76, 1601.486
77, 1622.026
78, 1642.736
79, 1663.846
80, 1684.045
81, 1705.802
82, 1725.423
83, 1747.014
84, 1767.163
85, 1788.602
86, 1808.565
87, 1830.200
88, 1850.135
89, 1872.024
90, 1892.543
91, 1914.437
92, 1935.186
93, 1955.942
94, 1976.769
95, 1998.325
96, 2018.890
97, 2040.277
98, 2061.358
99, 2082.912
100, 2104.020

 

[cmdrdredd]

They should probably look into the growing scandal about NVDA's image quality sacrifices in Far Cry 3 as well....

Or investigate how in their screenshots it's clear that the wood paneling has text that is washed out and nearly invisible on the AMD shot while the Nvidia shot shows it clear as day. How about that stone like arch on top of the shed, it loses detail on the AMD card.

That's what their screenshots show. So...image quality sacrifices? I see more details presented in the image. Oh and the hand on the character is more reddish on the Nvidia shot. Is that more or less correct vs what the artist was trying to show?

 

[blastingcap]

Is that with FRAPS or with a high-speed camera? Because if it's FRAPS then that doesn't solve the problem that Techreport talked about. See post #2 of this thread.

But to answer your question, you could dump it into Excel and make a graph if you want.

I've got frametimes for ya'll. 7970 Lightning, 2560x1600 ultra settings no AA.

What is the best way to post them without making a super loooong post?

 

[Zanovar]

I've got frametimes for ya'll. 7970 Lightning, 2560x1600 ultra settings no AA.

What is the best way to post them without making a super loooong post?



here are the preliminary results

2012-12-05 19:12:12 - farcry3_d3d11
Frames: 4423 - Time: 91167ms - Avg: 48.515 - Min: 38 - Max: 69

and a small sample of frametimes, I waited till the game was done loading and ready to be benched

Frame, Time (ms)
1, 0.000
2, 19.708
3, 41.276
4, 63.143
5, 84.565
6, 106.033
7, 127.538
8, 149.467
9, 171.014
10, 192.565
11, 213.882
12, 235.968
13, 257.911
14, 279.568
15, 300.969
16, 322.830
17, 344.023
18, 365.832
19, 388.203
20, 409.610
21, 431.489
22, 452.704
23, 474.168
24, 496.092
25, 517.309
26, 538.921
27, 560.617
28, 582.449
29, 603.422
30, 625.412
31, 646.514
32, 668.073
33, 689.479
34, 711.063
35, 732.735
36, 754.576
37, 776.043
38, 798.072
39, 818.969
40, 843.018
41, 861.868
42, 883.825
43, 905.018
44, 926.542
45, 947.706
46, 969.302
47, 989.867
48, 1011.659
49, 1033.039
50, 1053.341
51, 1075.391
52, 1096.599
53, 1117.756
54, 1138.566
55, 1160.179
56, 1180.727
57, 1202.329
58, 1222.778
59, 1245.034
60, 1265.673
61, 1286.865
62, 1307.871
63, 1329.400
64, 1349.830
65, 1372.211
66, 1392.262
67, 1414.470
68, 1434.775
69, 1455.693
70, 1476.529
71, 1497.943
72, 1518.081
73, 1538.942
74, 1559.724
75, 1581.000
76, 1601.486
77, 1622.026
78, 1642.736
79, 1663.846
80, 1684.045
81, 1705.802
82, 1725.423
83, 1747.014
84, 1767.163
85, 1788.602
86, 1808.565
87, 1830.200
88, 1850.135
89, 1872.024
90, 1892.543
91, 1914.437
92, 1935.186
93, 1955.942
94, 1976.769
95, 1998.325
96, 2018.890
97, 2040.277
98, 2061.358
99, 2082.912
100, 2104.020

Christ,that looks like my beer receipt from tesco haha

 

[Keysplayr]

Oh really?...In whose mind?
Yours?
Perhaps its a focus group thing....

Yeah, it's all just me Willie. I'm at war with myself. I was always a self-conflicted individual longing for a sense of smoothness to my gaming experiences. It isn't a focus group thing. It's my thing. :thumbsup:

 

[RampantAndroid]

I'm for this...but with a high speed camera as someone suggested. Only thing is, with a high speed camera, how do you know you are measuring the same frames on both systems since the camera is decoupled from the computer?

That's why you'd use a 1000 FPS camera, where you should easily capture it all, but lighting is an issue at 1000 FPS.

 

[BD231]

I'd be interested to see this too and have to agree that high speed video of the event is the only thing short of my eyes I will trust. Not 'ooh the blonde haired girl somehow seemed kinda sexier' sort of impression. Also vsync needs to be applied, if you're spending $800-2000 on gpu's you really should be able to use and be using vsync I now know.

If you spen 800 to 2000 on video cards and run them on a 60hz panel you shouldnt be using a computer.