Can you explain why it's wrong? 1 minute of a game is measured, roughly about 5-6k frames are rendered, and to make the graphical views a little clear, only 1500 frames off those are taken. Off those 1500, the real spikes get eliminated, just in case it's an OS hick-up, that's the 99 percentile. All the other spikes are the frames that take long to be rendered, due to whatever reason, mostly GFX or driver related.
Last edited by rsnubje; 03-07-2013 at 02:24 PM.
It is wrong because using a 99 percentile removes the information that you are specifically looking for. Removing 1% of spike frames does not take into consideration in any way, shape, or form what actually generated those spikes. It is literally just dumping them into the trash for no good reason.Originally Posted by rsnubje
All along the watchtower the watchmen watch the eternal return.
let me explain in other way why this frame time reporting is wrong beside the 99p thing as STEvil explained it. what they give us as frame time is the worst frame render time after ignoring %1 of worst frame times. n now for x vga lets say it is 30ms for other lets say y it is 25ms and both have the same average fps number. can we be %100 sure y is better then x? no because maybe for x has a steady 30ms all the time but y's frame time is fluctuating from 10 ms to 25ms very often.
When i'm being paid i always do my job through.
That's kinda why they show you the graphs and see the spikes with your own eyes. It's not as wrong as you say it is. You can clearly see in the graphs what the average is and what the spikes are.
yes but the graphs are only for titan can't compare with other cards
When i'm being paid i always do my job through.
I have try understand the point of show them, but cant find one, you only see 2-3-4 titan graphs.. even on the 99percentile graphs, you dont allways got all cards shown...
If they want do frametimes comparaison, at least they could do it correctly.
CPU: - I7 4930K (EK Supremacy )
GPU: - 2x AMD HD7970 flashed GHZ bios ( EK Acetal Nickel Waterblock H2o)
Motherboard: Asus x79 Deluxe
RAM: G-skill Ares C9 2133mhz 16GB
Main Storage: Samsung 840EVO 500GB / 2x Crucial RealSSD C300 Raid0
yes maybe not all the cards but i really wanted to see titan, titan sli 2 3 4 way graphs combined with 680 680 sli 690 690 sli etc. etc.
When i'm being paid i always do my job through.
IMO you don't really need a graph to know how good or bad the frametime performance of a card is and an average would be sufficient. All spikes caused by system "errors" are eliminated by the 1 percentile and every other spikes, which would probably be lower, are steady and caused by the graphics card/drivers of the graphics card. Even though the spikes are not mentioned, the average of all frametimes will give a pretty good indication of the actual ingame performance/experience. No matter how you look at it, lower average frametimes still give you the best experience. Feel free to think differently, you don't have to agree to this, I guess there's pretty much nothing that all people would think the same about.
If we(yes, I work there) would have a 10-20 card comparison it would become a mess with that much of graphs, or a lot of pages and I don't think anyone is waiting for that.
Actually for me it is not sufficient.
ok i can get the logic behind this but how can we be sure that %1 is absolutely because of system errors ???All spikes caused by system "errors" are eliminated by the 1 percentile and every other spikes, which would probably be lower, are steady and caused by the graphics card/drivers of the graphics card.
Average of frametimes is exactly 1000ms/average fps nothing more.Even though the spikes are not mentioned, the average of all frametimes will give a pretty good indication of the actual ingame performance/experience.
of course we can and should think differently, if we all think the same then nothing happens but i am sure you will also agree with me that if we have 680, 680 sli 690 and 690 sli graphs it would better.No matter how you look at it, lower average frametimes still give you the best experience. Feel free to think differently, you don't have to agree to this, I guess there's pretty much nothing that all people would think the same about.If we(yes, I work there) would have a 10-20 card comparison it would become a mess with that much of graphs, or a lot of pages and I don't think anyone is waiting for that.
also other then showing the graphs i think you can other better ways to express frametimes. like after calculating the avg frametime (ft) you can take the average of < avg ft and > avg ft times and give them. also min and max frametimes. and if i am not wrong what causes that disturbing stutters are big fluctuations on frametimes so with the worst frametime maybe worst fluctuation number may also be useful too.
When i'm being paid i always do my job through.
It's exceedingly frustrating that it seems these two companies can't get their sht straight as far as software support. It's embarrassing that the MORE money you spend with a company to buy 2/3/4 cards from them you actually get punished with poor performance or just outright instability. I'd be more than happy to spend the dimes on a 4 way Xfire/SLI setup, but not if opening a game equates to a round of Russian roulette stability wise.
Seriously they should just outsource the driver coding to India or something. Paying someone $60-70k a year to code doesn't appear to be working for them; so send it to a coder farm of 50 people for the same money. At this point it couldn't hurt to try
Posting Permissions
Reply With Quote
Bookmarks