Igalia Working Towards Faster 2D Rendering For Older Raspberry Pi Boards

[coder]

Basic desktop should be much less demanding than Quake 3. How is it possible that the board can run Quake 3 but not a basic desktop environment?

The article explained it. The reason is that the DE is running unaccelerated because they don't want to use up too much of the memory budget allocated to the GPU.

[coder]

Back when it was new, people were saying the Pi 2 was quite good for basic web browsing, email, etc. I have a Pi 3 that seems to do that fairly well, not that I've tried much. I think it's probably decent.

I don't know who these people were, but every time the Pi touched storage you felt it. Until Pi v4, storage was connected via USB2, even if you used a SD Card. I can't comment on desktop graphics performance, since my experience with Pi's has mostly been headless and using them remotely.

I guess one more thing I can say about general CPU performance is that I remember the first time I tried compiling a fairly simple C++ program on my Pi v1. It took so long that I thought the compiler might've hung! It was like 20 seconds, for something that compiled in well under a second on my PC. IIRC, the CPU cores in Pi v2 were each about 50% faster, and there were 4 of them. Still, I'd imagine its poor single-thread performance would significantly hamper its usability as a general purpose desktop.

[coder]

genune SoundBlaster 16 in my retro-computer

I have a Wave Blaster II that I never opened. If you want it, PM me and I'll see if I can find it.

I had another one which I did use. I only had one game that used it (Descent 2 or 3?), but I remember it sounded amazing at the time.

That reminds me: Quake gets a lot of well-deserved credit for its pioneering use of 3D rendering, but Descent was really up there as well. Unlike Quake, I don't think it was Z-buffered or used lighting maps, but they did perspective-correct texture mapping in software, and it ran at playable speeds on a 486.

[blackshard]

This is not accurate. I don't know where you get this BS, but the Pi's GPUs were never competitive.

Never competitive? I don't think so; VideoCore IV has strong 2D capabilities (it is full OpenVG 1.1 capable), full GLES2 3D GPU, a flexible hardware compositor that you can't even find on recent product or fully-fledged video cards, support for full-hd h.264 hardware decoding and all of this is a power envelope of few watts, ten years ago.

Also the 3D GPU of the RPi1 is exactly the same of the RPi3 you can see running 1080p at 80fps Quake3 on vulkan depicted in a post in this thread, which is yet quite decent performance.

What was the competition at that time? ARM Mali-450? Adreno 2xx series?

[hamishmb]

I don't know who these people were, but every time the Pi touched storage you felt it. Until Pi v4, storage was connected via USB2, even if you used a SD Card. I can't comment on desktop graphics performance, since my experience with Pi's has mostly been headless and using them remotely.

I guess one more thing I can say about general CPU performance is that I remember the first time I tried compiling a fairly simple C++ program on my Pi v1. It took so long that I thought the compiler might've hung! It was like 20 seconds, for something that compiled in well under a second on my PC. IIRC, the CPU cores in Pi v2 were each about 50% faster, and there were 4 of them. Still, I'd imagine its poor single-thread performance would significantly hamper its usability as a general purpose desktop.

Yeah, compiling on a Pi 1 isn't much fun, for sure.

My Pi 3 uses USB2 for the SD card too then, and that isn't that bad. I think it very much depends on your application and the quality of your SD card.

[caligula]

The longer, much less soundbite-y answer is a lot more involved, and TBH I'm not sure I can explain it well enough for normal people to understand without running on for pages, but I'll try....

Thanks - makes more sense now. I was originally questioning this because I've also experience with older hardware. My Windows 3.11 PC had a ISA GPU with less than 1 MB of VRAM. I used to set the desktop resolution to 800x600 (can't remember if it supported 256 colors with 512 kB of VRAM). My Windows 95(B) PC had a VLB GPU with exactly 1 MB of VRAM. The main CPU was 75 MHz Pentium. No vector instructions. Pretty slow RAM compared to DDR2. If you compare the hardware, I don't know all the specs, but I'm pretty sure all components of even RPi 1 are multiple times faster than the old crap.

E.g. when rendering a double-buffered scene, the blit operation was already using a large portion of the CPU power. Even pure software rendering should be tens of times faster. 1920x1080 x 24b contains only about 8 times more pixel data than 1024x768 x 8b. I know modern applications are quite a bit more demanding and less static than those back then. However even those systems had TTF font rendering, window decorations, etc. The compilers did suck a lot more. The performance gap between early 1990s hardware and RPi 2, at least, is staggering.

[coder]

Never competitive? I don't think so; VideoCore IV has strong 2D capabilities (it is full OpenVG 1.1 capable), full GLES2 3D GPU, a flexible hardware compositor that you can't even find on recent product or fully-fledged video cards, support for full-hd h.264 hardware decoding and all of this is a power envelope of few watts, ten years ago.

Compared with tablet-grade SoCs of its time, it was at or below-par. OpenGL ES 2.0+ and H.264 decode acceleration was already the norm, and even OpenCL support was more common than not - something the Pi still lacks to this day.

Also the 3D GPU of the RPi1 is exactly the same of the RPi3 you can see running 1080p at 80fps Quake3 on vulkan depicted in a post in this thread, which is yet quite decent performance.

The 3D capability of the VideoCore VI actually does differ from the original VideoCore IV and a Pi v1 sure isn't going to run Quake3 at 1080p @ 80 fps!

BTW, Quake 3 was already 12 years old, when the original Pi launched. Hardly a cutting-edge workload.

What was the competition at that time? ARM Mali-450? Adreno 2xx series?

Mali T604 launched in 2010; the T658 in 2011. Both supported OpenGL ES 3.1 and OpenCL 1.1. PowerVR SGX544 and SGX554 both launched in 2010. Looks like Adreno 220 launched in 2010-2011.

When the original Pi launched, its GPU was acceptable, not cutting-edge. The main problem with the Pi's GPU is that even the VideoCore VI isn't much better than the original VideoCore IV. Meanwhile, the industry standard improved by orders of magnitude.

[coder]

My Pi 3 uses USB2 for the SD card too then, and that isn't that bad. I think it very much depends on your application and the quality of your SD card.

The quality of the SD card has almost nothing to do with it. You can't read or write more than about 25 MB/s, period. That's well below hard disk speeds!

For reference, SATA 1 first shipped in ~2004 with a speed of 150 MB/s, and a modern, typical HDD can surpass that.

Granted, you'll get more IOPS from a Pi v3 + SD Card than a HDD, but not remotely close to a PC with a SATA SSD.

[hamishmb]

The quality of the SD card has almost nothing to do with it. You can't read or write more than about 25 MB/s, period. That's well below hard disk speeds!

For reference, SATA 1 first shipped in ~2004 with a speed of 150 MB/s, and a modern, typical HDD can surpass that.

Granted, you'll get more IOPS from a Pi v3 + SD Card than a HDD, but not remotely close to a PC with a SATA SSD.

Yeah, of course a HDD is much quicker than that. You can get more like 4 MB/s if you choose a bad SD card though - I have an original Pi 1 with just such a card, and it makes it horrendously slow for a lot of things.

Of note might be that the IDE HDDs in my circa 2005 Dell D600 and iBook G4 got a maximum of about 40MB/s. Obviously much older though.

[coder]

Thanks - makes more sense now. I was originally questioning this because I've also experience with older hardware. My Windows 3.11 PC had a ISA GPU with less than 1 MB of VRAM. I used to set the desktop resolution to 800x600 (can't remember if it supported 256 colors with 512 kB of VRAM).

If you do the math, 800x600 @ 8-bit should fit in 512 kB. So, it's mainly just a question of whether the RAMDAC could handle it. And the term "GPU" was first used in 1999. Before then, they were just "graphics cards" and IIRC the chips on them were referred to as "video chipsets". You'd sometimes see the more advanced ones referred to as "Windows accelerators".

My Windows 95(B) PC had a VLB GPU with exactly 1 MB of VRAM. The main CPU was 75 MHz Pentium. No vector instructions. Pretty slow RAM compared to DDR2.

I had an ISA graphics card with a Tseng Labs ET4000 and 1 MB of VRAM. It could do 1024x768 @ 256 colors. We got that card with our 386, and then kept it when upgrading to a Pentium 75. The monitor was a whopping 13", but I mostly gamed at fiddled around at 320x200. Normal text mode was 640x480. I mostly stayed in DOS and just used Windows 3.1 for MS Word and Internet.

Vector instructions weren't introduced until the Pentium MMX, which ran at 233 MHz. RAM was definitely like sub-100 MB/s, but I don't remember exactly.

If you compare the hardware, I don't know all the specs, but I'm pretty sure all components of even RPi 1 are multiple times faster than the old crap.

At launch, the Pi v1 was advertised as being somewhat comparable to a Pentium 200 (a 2-way, in-order core).

I know modern applications are quite a bit more demanding and less static than those back then.

Software has a tendency to run well only on the hardware of its day.