Slow Redraw Of Crosshair Cursor in NT8

[NinjaTrader_Brett]

Is this the standard crosshair or the global crosshair?

[Taddypole]

This is the standard crosshair. I didn't even think to try global crosshairs since the standard crosshairs weren't working to my expectations.

I've run some performance tests on my FirePro 2450 graphics card using PassMarks "Performance Test" software (version 8 and the new beta version 9) and the performance doesn't look that bad. It doesn't do 3D well but that shouldn't be a factor.

My card has DirectX 10.1 and as far as I have read, anything after 9.0 should provide hardware acceleration which is what WPF needs so as not to run it all in software.

I did find some very interesting information at: http://kynosarges.org/WpfPerformance.html

In that article it talks about Anit-Aliasing settings and that using anti-aliasing may be a big hit on WPF performance.

In reading other forum posts on this subject, it doesn't seem clear that it is purely a graphics card issue. If it is, I'm willing to upgrade but I need to know what the issue is before doing so as not to spend money foolishly.

regards,
taddypole...

[NinjaTrader_Brett]

Taddypole,

Thanks for that info.

Here is the thing, we're just as much in the dark as you are since we haven't run WPF in the field large enough yet to get enough experience with WPF driven crosshair to really make any firm recommendations.

I've got an Nvidia GT660 and a laptop with Nvidia Quadro card in it and another laptop with intel integrated HD 600 series and both run the crosshair great for me for comparison.

The crosshair performance is completely video card driven since its the one doing the rendering. I do know that its easier to render solid lines then dashed/doted lines, please make sure your crosshair is set to be a solid line.

If your still getting bad performance with the crosshair with a solid line then it almost has to be video card limiting at this point. Would you have access to any other PC's with a different video card you could try out on to confirm before you go out and purchase a video card?

-Brett

[sledge]

This is the standard crosshair. I didn't even think to try global crosshairs since the standard crosshairs weren't working to my expectations.

I've run some performance tests on my FirePro 2450 graphics card using PassMarks "Performance Test" software (version 8 and the new beta version 9) and the performance doesn't look that bad. It doesn't do 3D well but that shouldn't be a factor.

My card has DirectX 10.1 and as far as I have read, anything after 9.0 should provide hardware acceleration which is what WPF needs so as not to run it all in software.

I did find some very interesting information at: http://kynosarges.org/WpfPerformance.html

In that article it talks about Anit-Aliasing settings and that using anti-aliasing may be a big hit on WPF performance.

In reading other forum posts on this subject, it doesn't seem clear that it is purely a graphics card issue. If it is, I'm willing to upgrade but I need to know what the issue is before doing so as not to spend money foolishly.

regards,
taddypole...

Try changing to windows basic mode as a test to see if any performance is gained.

http://windows.microsoft.com/en-us/windows7/change-the-theme-to-improve-performance

[Taddypole]

Taddypole,

If your still getting bad performance with the crosshair with a solid line then it almost has to be video card limiting at this point. Would you have access to any other PC's with a different video card you could try out on to confirm before you go out and purchase a video card?

-Brett

I checked and do have solid crosshairs, so no problem there. I also checked Global Crosshairs again with no improvement.

I do have a 3rd computer that is mostly used for utility/non trading stuff. It has no video card but does have the Intel integrated HD Graphics 4000.

I have been thinking of setting that one up to run NT8 all day instead of just when I get time. I have 2 LifeTime licenses already. Is it possible to get a test license to run NT8 on that machine?

regards,
taddypole...

[Taddypole]

Try changing to windows basic mode as a test to see if any performance is gained.

http://windows.microsoft.com/en-us/w...ve-performance

Thanks for the suggestion Sledge but no Joy there either....

regards,
taddypole...

[sledge]

Thanks for the suggestion Sledge but no Joy there either....

regards,
taddypole...

Are you running 1 chart per screen in full size mode or something?

If I run my laptop (3k screen (2880x1620)), with a Quadro K3100m, and 2 attached Dell U3415W (mini display port and hdmi both at (3440x1440)), if I put my 3 charts in full screen mode on each, global crosshairs does "slow down" slightly.

The smaller the chart - on each the better the performance. The bigger I make the chart I start getting some lag setting in. It's definitely noticeable compared to smaller sized charts. I couldn't imagine running a lesser card with these full screen charts.

As a test - make smaller charts per window and see if that helps. (I only have 1 indicator on one of these charts).


Maybe this weekend I'll put NT8 on my 2008 vista laptop and see what happens. (1 vga output, so no multi monitor stress).

edit - I see the same "lag" creeping in when I put all 3 charts full size on the same monitor "stacked".

[Taddypole]

Are you running 1 chart per screen in full size mode or something?

That's it Sledge. The smaller I make the chart, the better the performance. My chart also has no indicators on it. Just 5 min bars.

If I make the chart 1/8th the size of the screen, there is no noticeable lag. Nearly equal to NT7 with charts at full screen. "So back to, Which card will help me out!?"

To answer that question, I just found another utility called GUPZ. Very much like the processor utility "CPUZ", it runs and displays information about your GPU. (See attachment)

The software can be found at: http://www.techpowerup.com/downloads...-gpu-z-v0-8-7/

And with the GPU info display, there is a "Lookup" button that takes you to a page that display your video card and a good description of it.

Would be nice if users would take a snapshot of their GPU info and relate it to their crosshair performance rating to help other users who need to upgrade GPU hardware to get acceptable performance.


regards,
taddypole...

[crokusek]

As a software developer of 25 years, I'm fairly confident that the right place to fix the Cursor drawing algorithm in NT8 is in the rendering code itself--not the hardware. It is so much slower than NT7. We are in beta so it goes with the territory. Perfectly fine.

Nearly all modern graphic toolkits provide techniques to superimpose images temporarily on top of an existing images efficiently. Compared to video games of even 20 years ago, drawing a couple lines non-destructively on top of an existing image should be child's play so I hope the NT team will take some time to analyze alternative drawing order, buffering, etc. For example a good fix might be to just buffer the portion of the screen that will be destroyed by the moving cursor instead of the whole screen (this would explain why the effect gets worse with larger screen sizes).

I'm worried about the comments that try to shift the issue toward the hardware or sealed software toolkits. Of course hardware can be used to "fix" underlying issues successfully but its more more likely to be something simple in the code itself.

Thanks in advance,
Chris

Using Build 8.0.0.12

[Segwin]

I've found that the global crosshair really lags vs local.

@Brett: perhaps a post of minimum computer specs may help clarify.

[NinjaTrader_Brett]

Segwin,

Don't have any minimum specs for video card, though global crosshair can be expected to be a tad bit slower. Since as you may or may not know with NT8 we are multithreaded, each window has access to each core on your processor (therefore more bandwidth) but to cross the thread boundary with data of crosshair position on one window and pass it to other windows on other threads is expensive. So we try to do it as little as possible which is why global crosshair you see a little lag. More CPU resources wont necessarily resolve that part. If you see it seriously lagging that's a different story though, the lag I'm talking about might be milliseconds.

Crokusek,

If your comment is in respect to the regular crosshair. Then wanted to thank you for your well thought out post. We're on the same page and take the feedback to heart and chased all such 'simple' fixes to the limit we could with current implementation (Its fairly heavily optimized with current approach as is). We don't have any immediate plans yet to do more with crosshair rendering since to do so likely requires re-coding the entire thing to another implementation which may or may not result in better performance on slower video card hardware. Therefor I'm heavily monitoring feedback at this point and based on that work with development on if we needed to pull the trigger on that research or not. Thanks for the feedback.

-Brett

[molecool]

Having been in the SW dev racket for at least the same amount of time I was very concerned that this would be the case when I originally submitted this thread. I haven't been privy to the various optimizations that have been introduced since (work keeps me busy) but to me it sounds like as if the NT team is facing an API wall here. If I understand it correctly the implications would be substantial and cause a significant delay of the first full release of NT8. I really hope that won't be necessary but I have not been a big fan of the new GUI toolkit from the get-go. Maybe it's just me... I hope NT can make it work.

As a software developer of 25 years, I'm fairly confident that the right place to fix the Cursor drawing algorithm in NT8 is in the rendering code itself--not the hardware. It is so much slower than NT7. We are in beta so it goes with the territory. Perfectly fine.

Nearly all modern graphic toolkits provide techniques to superimpose images temporarily on top of an existing images efficiently. Compared to video games of even 20 years ago, drawing a couple lines non-destructively on top of an existing image should be child's play so I hope the NT team will take some time to analyze alternative drawing order, buffering, etc. For example a good fix might be to just buffer the portion of the screen that will be destroyed by the moving cursor instead of the whole screen (this would explain why the effect gets worse with larger screen sizes).

I'm worried about the comments that try to shift the issue toward the hardware or sealed software toolkits. Of course hardware can be used to "fix" underlying issues successfully but its more more likely to be something simple in the code itself.

Thanks in advance,
Chris

Using Build 8.0.0.12

[exupblue]

same problem here

The standard crosshair is working fine, but local and global the both have an important lag.

My version is NT8 8.0.0.12

Running in w10 nvidia Gforce 940M, 10gb ram and I7 4720hq
with those specs i don't thing my laptop should be the problem.

for the rest after one week testing it really looks like a nice upgrade form NT7

Exupblue

[BearingHH]

If you could make an improvement of the performance on the crosshair it would be fantastic, as it is very slow on RDP connections for example (NT7 was definitively faster).

Another fix would be nice:

I switch between the normal cursor and local crosshair with keyboard shortcuts, so I could point my normal cursor to a destination and switch to the crosshair. But your NT8 code keeps track of the last position from the last crosshair "session" and display it there. After the mouse moves for one pixel, it will be redrawn on the right position.
Would be nice if the crosshair would be directly drawn on the new position when switched to it by keyboard shortcut.

[NinjaTrader_Brett]

BearingHH,

Let me check into that with development and see what could be done. I agree would be more correct to just display at current mouse position after hotkey selection.

-Brett