FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,532
AMD & NVIDIA GPU VR Performance: Space Pirate Trainer - Do you remember playing Galaga? Space Pirate Trainer is a VR game that puts you into an arcade shooter, except as you might have guessed, gameplay is a bit more involved. So how do all the latest GPUs from AMD and NVIDIA hold up to SPT abuse? We wore our arms out, and even knocked a picture off the wall to find out.
 
This was the game I demoed and sold me on the Vive in about thirty seconds. When the girl told me to reach like I am itching my back and I came back with a shield... That was pretty cool.

Great review. It's fantastic to know which settings to load up.
 
went to buy a used gtx 780 off a kid on cl just a few hours ago and the address he texted me was to a vr arcade that recently opened up in town. apparently its a business his old man owns and he let me try out this very game. i was totally blase about vr until i gave it a go. man am i impressed. i guess seeing or rather trying is believing.
 
This game was already good but the recent update has added way more variety and polish. I hope they continue to improve it.
 
Kyle. With the summary list do you think you could add a column where you throw out the "best/worst" game for each card? This should minimize some sensitives to data fluctuation and give users a better idea how constant the card is. IIRC there is one game trucked all cards but the titan which is why I think you "average graph" might be a bit skewed or misleading.
 
Kyle. With the summary list do you think you could add a column where you throw out the "best/worst" game for each card? This should minimize some sensitives to data fluctuation and give users a better idea how constant the card is. IIRC there is one game trucked all cards but the titan which is why I think you "average graph" might be a bit skewed or misleading.
I am not sure of what you are asking for. You want me to sum up all our VR articles on the Leaderboard?
 
My guess/assumption is you have a spreadsheet/table with all the data. You sum up the N tests you have done for a card and divide by N. However, the problem with that is while the data has correlation (all VR games), there might be odd balls. So imagine if you 10 games and 9 of them had average render times of 5 ms per frame but you had one game that was just bad at was 30 ms and thus the average would be 7.5 ms. This makes the card look 50% worse than it really is because of "one game". So, if you throw out the odd ball it makes the data appear reasonable. The same goes the other way where a card does badly in general but 1 game might skew the numbers making it look much better than it really is.
OK, and what is your suggestion as to how you want us to represent this data exactly?

That said, we have not seen any game that scales differently in the big picture. And we do write articles about specific games that cover the points you are talking about getting more granularity on.
 
This game is a lot of fun. It's pretty basic overall but the latest update did add a ton of content. It's a nice game to just listen to the music (great techno soundtrack IMO) and shoot some stuff.
 
My guess/assumption is you have a spreadsheet/table with all the data. You sum up the N tests you have done for a card and divide by N. However, the problem with that is while the data has correlation (all VR games), there might be odd balls. So imagine if you 10 games and 9 of them had average render times of 5 ms per frame but you had one game that was just bad at was 30 ms and thus the average would be 7.5 ms. This makes the card look 50% worse than it really is because of "one game". So, if you throw out the odd ball it makes the data appear reasonable. The same goes the other way where a card does badly in general but 1 game might skew the numbers making it look much better than it really is.

Well so far I believe Raw Data has been the worst performing, but it was so on ALL cards. The data is only skewed a bit for the Titan because it wasn't out when some of the tests were done. (EDIT: Pointed out as incorrect by Kyle) However, it still tops the charts for every other game so I think the ranking is still valid regardless of the numbers.
 
Last edited:
The data is only skewed a bit for the Titan because it wasn't out when some of the tests were done
You are incorrect, I went back and retested and added the Titan data for all games.
 
My mistake then - I went back and checked the article but didn't see that the Titan had been added. Nice!
I did not go back and change the articles since that would like require a lot of rewriting. I did however go back and do the needed testing and add the data to the VR Leaderboard.
 
I dont like the new graphs D:

It really seems the 480/1070/1080 are not very good.

I hope next gen cards can handle VR and 4k.
 
I dont like the new graphs D:

It really seems the 480/1070/1080 are not very good.

I hope next gen cards can handle VR and 4k.

1070/1080/Titan XP do just fine... maybe you don't know what you are looking at?

Realize on page 5 Kyle had 8x AA and super sampling at 1.4x...
 
yeah maybe I dont know what I am looking at because graph quality is so bad.

1060/480 are great if you only want 1080p at 60fps.
So what particularly is your issues with the graphs exactly?

And you realize that there is no 60fps.
 
This means I need to click on them to view the larger version and they look great but thats just annoying and breaks the ability to digest the content easily.
Roger that.

p.s. would it be possible to represent avg min frame time and avg max frame time with a line on the graph?
Yes, but I do not think that is conducive to the message.

Bottom line there are over 30,000 data points on those graphs. Once you squeeze it down so much it I am not sure what it really shows.
 
Then again maybe dumbing it down is what needs to happen to appeal to a wider body of readership. (I know you are talking about formatting.) That is fairly technical stuff to grasp for a lot of folks I am sure, and I am sure most do not read the page on how to understand it.
 
The lines for the reprojection and dropped frames could be made thicker so they are easier to read. Alternately you might consider having them filled in if it's not a PITA in the software you're using.
 
The lines for the reprojection and dropped frames could be made thicker so they are easier to read. Alternately you might consider having them filled in if it's not a PITA in the software you're using.
When you make those thicker so they show better, they totally overlap on the vertical lines creating an issue where they are much too prominent and do not represent even close to properly. That said, it still does it now. Think about drawing 30,0000 vertical lines across that graph.....
 
The graphs seem fine to me. You can tell when you drop into reprojection and how long you're in reprojection overall, which is the main thing you care about for VR.
 
Tossing in compiled test results on a "low end" VR PC. This is the system I built for the kids to play on for now.

System:
FX-8320 @ 4.5
R9 390 @ 1100/1650
8GB DDR3 @1600

settings: Better, 1.0, 4xAA (same as in the review) (reprojection was turned off)

7.37 ms average
total frames: 38,963
dropped frames: 55
frames where render time exceeded 10ms: 415 (1.07% would be in reprojection)


I toyed with max settings, but 1.4 and 8xAA put the card into reprojection just idling at the menus.
 
Back
Top