OS2 World Community Forum

OS/2, eCS & ArcaOS - Technical => Hardware => Topic started by: Neil Waldhauer on October 21, 2021, 01:47:28 am

Title: OS/2 Video Performance
Post by: Neil Waldhauer on October 21, 2021, 01:47:28 am
People ask me what video card to buy to run OS/2. I usually tell them to run the graphics built into the motherboard. For some time, Intel CPU have graphics built in. This is an inherently better design than having a separate video card.

I ran the SysBench benchmark program on my Quebopeep (Lenovo M93p Tower computer) with Intel Core-i7 CPU connected to the monitor with a standard VGA cable. I installed the standard AMD Radeon HD8490 PCIe-16 video card with 1 GB video RAM. The graphics card has no VGA, but I used a DisplayPort to VGA adapter to connect the monitor. Here is how the two graphics setups compared.

Intel HDA Graphics
PM Graphics Marks: 268
DIVE Marks: 18062

AMD Radeon HD8490
PM Graphics Marks: 127
DIVE Marks: 2917

So you can see that the high performance video card is between 2 and 6 times slower than the built-in graphics on the motherboard.
Title: Re: OS/2 Video Performance
Post by: Dave Yeo on October 21, 2021, 05:53:59 am
OTOH, My old I5-2400's built in graphics gives,
PMGraphics score 190
Dive Marks 7511
Title: Re: OS/2 Video Performance
Post by: OS4User on October 21, 2021, 08:20:31 am
Intel HDA Graphics
PM Graphics Marks: 268
DIVE Marks: 18062

For OS/2  "DIVE -> Video bus bandwidth" is more significant. Since built-in graphics have bigger VBB  all other marks are also bigger for built-in. In most cases built-in graphics is even faster then HW solutions supported by SNAP (ATI x850).
Title: Re: OS/2 Video Performance
Post by: Dariusz Piatkowski on October 21, 2021, 04:53:41 pm
Interesting question Neil...the hardware has certainly gotten significantly faster, but our software has not kept up, so the challenge is in deciding what hardware speed increases translate to better OS/2 performance.

Here is an example, my old ATI X850 XT PE gives me the following results:

PM Graphics Marks: 1159
DIVE Marks: 2139

...and based on these hardware results, how does that translate to human perception of faster/slower screen updates?

So specifically, if the new hardware is so much faster, given that the current video drivers do not provide hardware acceleration, is it worth moving from something like the aged SNAP drivers (where I've got my two monitors) to a single screen, because all other applications become blazing fast?

Perhaps it is, but that would have to be driven by something like FF, Thunderbird, PMView, VLC and others working much quicker.

I have yet to see this. In fact, the combo of my old ATI X850 XT PE hardware and the supporting SNAP drivers will have to be pried out of my cold dead hands before I give up the functionality of that dual screen!!!  8)

Title: Re: OS/2 Video Performance
Post by: Sigurd Fastenrath on October 23, 2021, 12:18:07 am
My own build Thinkpad 25/2:

PM Graphic Marks: 58.294
Dive Marks: 7336.388

Title: Re: OS/2 Video Performance
Post by: Dave Yeo on October 23, 2021, 06:45:48 am
My own build Thinkpad 25/2:

PM Graphic Marks: 58.294
Dive Marks: 7336.388

That's sure slow drawing lines. Have you ever compared with running SNAP? Here SNAP is slightly faster and the individual tests are somewhat different in results. Luckily this Intel BIOS has wide screen resolutions in it.
Title: Re: OS/2 Video Performance
Post by: Andi B. on October 23, 2021, 11:51:10 am
Radeon550 with Panorama. Panorama can not get integrated Intel to output 3840x2160 so I need the Radeon.
Title: Re: OS/2 Video Performance
Post by: Martin Iturbide on October 23, 2021, 03:28:32 pm
Hi

Sysbench graphics tests was fun to watch :)

- Thinkpad L420 (7856-3WS) (https://www.os2world.com/wiki/index.php/Lenovo_ThinkPad_L420) with Panorama v1.17.
- Graphics:Intel HD Graphics 3000 (8086:0116)

Regards
Title: Re: OS/2 Video Performance
Post by: Dariusz Piatkowski on October 23, 2021, 03:53:23 pm
OK, so given the results everyone has shared so far, I interpret the classifications as:

1) PM Graphics Marks => higher value directly translates to FASTER WPS/PM updates
- majority of our applications today fall into this category

2) DIVE Marks => higher value directly translatest to FASTER updates for those apps that specifically use DIVE
- possibly video playback such as VLC, SMPlayer, mplayer? (K's releases primarily use the kva output driver here, but I see that some allow the selection between 'kva snap mode' and 'kva slower dive mode' - that's directly from SMPlayer config)

Not sure if stuff like FF for example uses DIVE or not? But at least here that's probably the most demanding applicatin (given that it's all about on-going screen updates) and from my perspective it seems like a higher 'PM Graphics Marks' value impacts this the most.
Title: Re: OS/2 Video Performance
Post by: Neil Waldhauer on October 23, 2021, 04:30:02 pm
I think a lot of newer graphics stuff, Mozilla and QT, don't use PM for graphics. We could use a better benchmark.

Be careful of confirmation bias; where you have a card that has better PM graphics so that must be what is important. OS/2 programs are old, and I suspect that your old graphics card works just as well as newer HDA graphics, and that all that extra speed is wasted. My reason for starting this thread was to remove the obstacle of using old hardware to get good graphics performance. I wanted to make it easier for people to select a good graphics platform to run OS/2.

This is going to change again with ArcaOS 5.1, which will have some new graphics features to replace some BIOS functions that are no longer there.
Title: Re: OS/2 Video Performance
Post by: Dariusz Piatkowski on October 23, 2021, 04:49:09 pm
Neil!

I think a lot of newer graphics stuff, Mozilla and QT, don't use PM for graphics. We could use a better benchmark.

Great point, exactly what I speculated above, although I am not at all concluding in one sweep that they "don't use PM for graphics", after all, a simple window, along with MIN/MAX controls is a PM window after all. If someone who knows these apps can comment further, we all would be in a better position to assess.

...Be careful of confirmation bias; where you have a card that has better PM graphics so that must be what is important...

Umm, this is not at all the point I am trying to make. What is relevant to each one of us is the improvement we see on our systems, that of course being based on our "runtime": system settings along with the standard apps we all use. By discussing in our forum we perhaps can formulate a definition of a NEW benchmark that would allow us to measure the performance across various applications - think of a app simulation benchmark!

What might that look like?

1) PM apps
SysBench covers that with it's 'PM Graphics Marks' score

2) DIVE / SNAP / KVA
SysBench covers part of that, but we need to better understand what new testing (if any) is required to cover the remaining ones

3) QT
Ahhh...that would be the new addition, right?...and maybe we need a 'PM Graphics Marks' equivalent, call it 'QT Graphics Marks'?

4) anything else...
This will only be arrived at by establishing some kind of consesus as to what matters to all of us.

And on that very point, you started this thread with a specific assumption that SNAP would not be used, and so you arrived at some results that clearly put the 'Intel HDA Graphics' in the lead...but isn't that the VERY definition of BIAS???
Title: Re: OS/2 Video Performance
Post by: Dave Yeo on October 23, 2021, 07:02:07 pm
Not sure if stuff like FF for example uses DIVE or not? But at least here that's probably the most demanding applicatin (given that it's all about on-going screen updates) and from my perspective it seems like a higher 'PM Graphics Marks' value impacts this the most.

It depends, DIVE is supposed to be disabled if using Panorama and enabled if using SNAP.  I have a vague memory of things getting backwards at some point, need to look. Also perhaps Panorama has improved its handling of the shadow buffer enough that DIVE would be a good choice.
The setting for DIVE should be "Full Hardware Acceleration enabled" or similar.
Title: Re: OS/2 Video Performance
Post by: Roderick Klein on October 23, 2021, 11:56:51 pm
Not sure if stuff like FF for example uses DIVE or not? But at least here that's probably the most demanding applicatin (given that it's all about on-going screen updates) and from my perspective it seems like a higher 'PM Graphics Marks' value impacts this the most.

It depends, DIVE is supposed to be disabled if using Panorama and enabled if using SNAP.  I have a vague memory of things getting backwards at some point, need to look. Also perhaps Panorama has improved its handling of the shadow buffer enough that DIVE would be a good choice.
The setting for DIVE should be "Full Hardware Acceleration enabled" or similar.

DIVE is not switched off with Panorama. Its something provided by MMOS/2.

Roderick
Title: Re: OS/2 Video Performance
Post by: Paul Smedley on October 24, 2021, 12:03:20 am
Not sure if stuff like FF for example uses DIVE or not? But at least here that's probably the most demanding applicatin (given that it's all about on-going screen updates) and from my perspective it seems like a higher 'PM Graphics Marks' value impacts this the most.

It depends, DIVE is supposed to be disabled if using Panorama and enabled if using SNAP.  I have a vague memory of things getting backwards at some point, need to look. Also perhaps Panorama has improved its handling of the shadow buffer enough that DIVE would be a good choice.
The setting for DIVE should be "Full Hardware Acceleration enabled" or similar.

DIVE is not switched off with Panorama. Its something provided by MMOS/2.

The way I read Dave's comment, he's saying that within *firefox* dive is supposed to be disabled using panorama, and enabled using Snap....
Title: Re: OS/2 Video Performance
Post by: Dave Yeo on October 24, 2021, 04:18:26 am
Not sure if stuff like FF for example uses DIVE or not? But at least here that's probably the most demanding applicatin (given that it's all about on-going screen updates) and from my perspective it seems like a higher 'PM Graphics Marks' value impacts this the most.

It depends, DIVE is supposed to be disabled if using Panorama and enabled if using SNAP.  I have a vague memory of things getting backwards at some point, need to look. Also perhaps Panorama has improved its handling of the shadow buffer enough that DIVE would be a good choice.
The setting for DIVE should be "Full Hardware Acceleration enabled" or similar.

DIVE is not switched off with Panorama. Its something provided by MMOS/2.

Roderick

As Paul said, the Mozilla apps turn off DIVE when they detect Panorama or in safe mode. from SeaMonkey's stdout,
Code: [Select]
DIVE is disabled - Panorama's shadow-buffer is enabled
the code from nsWindow.cpp,
Code: [Select]
  // Don't use DIVE if the Panorama video driver is in use
  // unless its shadow buffer is turned off.
  HMODULE hmod;
  if (!DosQueryModuleHandle("PANOGREX", &hmod)) {
    char      str[8];
    if (PrfQueryProfileString(HINI_USERPROFILE, "PANORAMA", "VBEShadowBuffer",
                              0, str, sizeof(str)) && !strcmp(str, "0")) {
      sUseDive = TRUE;
      printf("Video driver is Panorama - shadow-buffer is disabled\n");
    }
    else
      printf("DIVE is disabled - Panorama's shadow-buffer is enabled\n");

    return;
  }

  sUseDive = TRUE;

I think it was Rich who made that decision based on DIVE having crappy performance with the Panorama shadow buffer. If the shadow buffer is turned off in the Screen object, Firefox will use DIVE.
Other video drivers use DIVE but hide the pointer when writing to the framebuffer.
As said, it is also controlled by the video acceleration preference.
Title: Re: OS/2 Video Performance
Post by: Neil Waldhauer on October 24, 2021, 03:47:32 pm

And on that very point, you started this thread with a specific assumption that SNAP would not be used, and so you arrived at some results that clearly put the 'Intel HDA Graphics' in the lead...but isn't that the VERY definition of BIAS???

I agree that HDA does well in SysBench DIVE measurements. I have run SNAP measurements for a long time, and unless the graphics card is 15 years old, they are never far from Panorama. For some systems, there are performance issues with SNAP, but the Arca Noae version seems to solve them. I've gotten used to all the goodies that come with Panorama, so I don't give SNAP much attention.

But if you do have one of those old graphics cards that support SNAP, I agree that things are pretty rosy. I used to sell only systems with SNAP graphics accelerated support. You can count me as a fan. But with systems that only have SNAP accelerated support, I'm going to assume Arca Noae version of SNAP will be about the same as Panorama.
Title: Re: OS/2 Video Performance
Post by: jailbird on November 06, 2021, 12:48:48 am
I agree that HDA does well in SysBench DIVE measurements. I have run SNAP measurements for a long time, and unless the graphics card is 15 years old, they are never far from Panorama. For some systems, there are performance issues with SNAP, but the Arca Noae version seems to solve them. I've gotten used to all the goodies that come with Panorama, so I don't give SNAP much attention.

Forgive me if I'm being naive or dense, I'm not an OS/2 expert at all.  From what I've picked up so far in this thread is:

- On "modern" graphics, SNAP and Panorama are about equal in performance
- DIVE seems to always been faster than PM
- In SM/FF/TB, Panorama with shadow buffers enabled disables DIVE
- Disabling shadow buffers in Panorama lowers PM performance, I believe?

So if Panorama has a tradeoff of either slower SM/FF/TB by preventing the use of DIVE or slower everything else in the disabling of the shadow buffer and SNAP *doesn't* have this issue as far as I can tell, then how in Panorama superior to SNAP on modern cards?
Title: Re: OS/2 Video Performance
Post by: Dave Yeo on November 06, 2021, 01:26:34 am
I have installs with both Panorama and SNAP. I can't notice any difference on the speed of SeaMonkey in either install. Scrolling is smooth without too much CPU. This is an old I5 with Intel graphics.
The big problem with SNAP is lack of wide screen support generally though this video bios does support wide screens with SNAP, it is the only one I've come across. Currently running 1920x1200x32.