BAFelton
Well-Known Member
GP MARK Caanoo 1.0.6 :
331.5
252.5
162.6
251.3
59.4
23.2
5.1
331.5
252.5
162.6
251.3
59.4
23.2
5.1
Gpmark
- Thread starter CKeichel
- Start date
BAFelton
Well-Known Member
Exophase
Nothing good will ever come of Exophase.
Thanks BAFelton, very helpful comparison. It shows to illustrate a little bitter what's actually happening, since I haven't seen the bench until now.
First a couple notes:
1) We don't know what refresh rate both are running at. It's possible that in very heavily memory bandwidth dominated tests, which some of these are, that a lower refresh rate will show up as an advantage, because potentially less bandwidth is taken by the framebuffer to display DMA transfer. I think this is the most likely explanation for the advantage that the left Caanoo has in some of the tests, as well as the disadvantage the right one has. I don't think that SDL soft-rotation can account for it because that'd show up as a bigger hit. This advantage is the sort of the thing that will show up in synthetic tests much more than any most real world applications, especially emulators, because few things will exhaust memory bandwidth.
2) Framebuffer is write only, so whether direct to framebuffer or allocated by SDL it won't go through cache because the ARM9 is configured as not allocating on write miss. But the write buffer will usually hide RAM latency, which is what we suspect to be the biggest difference between the Caanoo and Wiz in terms of hardware deficiency.
3) I'm making a lot of guesses because we don't have source. It'd help a lot if we had source.
These are my basic descriptions for the memory vs cache performance of the tests based on what it looks like they're doing (for more I really need to see source).. my apologies if this is redundant or worse, contradictory to what the on-screen comments say. I couldn't really read them as they scroll by. Maybe someone could post them for me, although I suppose I could run the damn test myself when I get home..
Blitter: 332.2 vs 331.5, Wiz + 0.21%
All we see drawn is a white bitmap, so it might be done in a memset style operation. This would be register to framebuffer, which is not going to exercise memory latency at all. But this would mean that it's not a real blit operation, which is memory to memory. I need Optimus's input on this.
The results are essentially a tie.
Plasma: 242.9 vs 252.3, Caanoo + 3.73%
The display is generated mathematically, so I expect it'll be working from registers/cache to memory and thus not be latency sensitive. Caanoo's mysterious advantage wins this one.
Rotozoomer: 169.6 vs 162.5, Wiz + 4.19%
The image Optimus uses is 80x80, presumably 16bpp, which is actually still small enough to fit into dcache. What I don't know is this image is repeated in memory or not. I presume it is, which means that it won't fit in dcache and will therefore exercise memory latency on the reads. This will hold part of the time; where it's zoomed out the Wiz has the advantage, when it's zoomed in the Caanoo does.
Rotozoomer near: 233.9 vs 251.3, Caanoo + 6.9%
This one will fit in dcache regardless of whether or not the image is tiled, since it's just using a small portion as its working set. Also note that here Caanoo tops out to near the same speed as it did in Plasma.
Rotozoomer far: 73.8 vs 59.3, Wiz + 19.65%
Never fitting in dcache, this will heavily exercise memory latency, especially if the image is not tiled. And we can see a big difference in the results. If whatever is giving the Caanoo the advantage the rest of the time is normalized out (sorry but I think there's something in terms of software configuration that is, either refresh rate or time spent in kernel routines) then Wiz will have an even bigger advantage.
Radial blur: 20.9 vs 21.8, Caanoo + 4.13%
This does appear to do memory to memory, but it's very slow, so it probably spends proportionately much more time working the ALU/cache and not latency sensitive operations, unless it's using huge LUTs which I imagine not. So Caanoo wins again.
3D Bunny: 5.3 vs 5.1, Wiz + 3.77%
This is non-textured, so it's going to be very ALU intensive. There will be some memory latency involved for fetching vertex data, which probably won't all fit in cache. This will only have a small impact.
So we can conclude from these tests that if the memory latency issue is nullified Caanoo has some advantage. There's no hardware explanation for this, so it's most likely a matter of software configuration, which can be refresh rate, kernel improvements, or other timing differences. The big change in performance between Caanoo firmware versions backs this up as well, but I suspect that different memory clock dividers are being used entirely there. We can also conclude that where memory latency is critical Wiz can have a big advantage.
Honestly, I think what these tests most illustrate is that synthetic benchmarks are misleading. Most of these tests aren't exercising memory latency at all and the one that is is a lot. In the real world very few useful programs will run completely out of cache and will only perform write-only transactions to memory, especially not for a 16KB data cache. Most will have a bigger working set and some number of dcache misses, whose latency can't be hidden at all on this architecture. The overall ratio of memory to cache/ALU will vary, as will the access pattern - higher spatial locality of reference will reduce cache misses.
Another factor is code size. Most likely, most of these tests fit in icache. Something that doesn't will have additional memory read (and hence memory latency) requirements. This is usually not a dominating factor in a program but it will often have at least some effect.
So I think we won't see the nearly 20% disadvantage for Caanoo in most real world programs, but I also think you won't tend to see Caanoo come out on top either. Lordus's original 10% numbers for the emulators he tested fits pretty nicely into this.
First a couple notes:
1) We don't know what refresh rate both are running at. It's possible that in very heavily memory bandwidth dominated tests, which some of these are, that a lower refresh rate will show up as an advantage, because potentially less bandwidth is taken by the framebuffer to display DMA transfer. I think this is the most likely explanation for the advantage that the left Caanoo has in some of the tests, as well as the disadvantage the right one has. I don't think that SDL soft-rotation can account for it because that'd show up as a bigger hit. This advantage is the sort of the thing that will show up in synthetic tests much more than any most real world applications, especially emulators, because few things will exhaust memory bandwidth.
2) Framebuffer is write only, so whether direct to framebuffer or allocated by SDL it won't go through cache because the ARM9 is configured as not allocating on write miss. But the write buffer will usually hide RAM latency, which is what we suspect to be the biggest difference between the Caanoo and Wiz in terms of hardware deficiency.
3) I'm making a lot of guesses because we don't have source. It'd help a lot if we had source.
These are my basic descriptions for the memory vs cache performance of the tests based on what it looks like they're doing (for more I really need to see source).. my apologies if this is redundant or worse, contradictory to what the on-screen comments say. I couldn't really read them as they scroll by. Maybe someone could post them for me, although I suppose I could run the damn test myself when I get home..
Blitter: 332.2 vs 331.5, Wiz + 0.21%
All we see drawn is a white bitmap, so it might be done in a memset style operation. This would be register to framebuffer, which is not going to exercise memory latency at all. But this would mean that it's not a real blit operation, which is memory to memory. I need Optimus's input on this.
The results are essentially a tie.
Plasma: 242.9 vs 252.3, Caanoo + 3.73%
The display is generated mathematically, so I expect it'll be working from registers/cache to memory and thus not be latency sensitive. Caanoo's mysterious advantage wins this one.
Rotozoomer: 169.6 vs 162.5, Wiz + 4.19%
The image Optimus uses is 80x80, presumably 16bpp, which is actually still small enough to fit into dcache. What I don't know is this image is repeated in memory or not. I presume it is, which means that it won't fit in dcache and will therefore exercise memory latency on the reads. This will hold part of the time; where it's zoomed out the Wiz has the advantage, when it's zoomed in the Caanoo does.
Rotozoomer near: 233.9 vs 251.3, Caanoo + 6.9%
This one will fit in dcache regardless of whether or not the image is tiled, since it's just using a small portion as its working set. Also note that here Caanoo tops out to near the same speed as it did in Plasma.
Rotozoomer far: 73.8 vs 59.3, Wiz + 19.65%
Never fitting in dcache, this will heavily exercise memory latency, especially if the image is not tiled. And we can see a big difference in the results. If whatever is giving the Caanoo the advantage the rest of the time is normalized out (sorry but I think there's something in terms of software configuration that is, either refresh rate or time spent in kernel routines) then Wiz will have an even bigger advantage.
Radial blur: 20.9 vs 21.8, Caanoo + 4.13%
This does appear to do memory to memory, but it's very slow, so it probably spends proportionately much more time working the ALU/cache and not latency sensitive operations, unless it's using huge LUTs which I imagine not. So Caanoo wins again.
3D Bunny: 5.3 vs 5.1, Wiz + 3.77%
This is non-textured, so it's going to be very ALU intensive. There will be some memory latency involved for fetching vertex data, which probably won't all fit in cache. This will only have a small impact.
So we can conclude from these tests that if the memory latency issue is nullified Caanoo has some advantage. There's no hardware explanation for this, so it's most likely a matter of software configuration, which can be refresh rate, kernel improvements, or other timing differences. The big change in performance between Caanoo firmware versions backs this up as well, but I suspect that different memory clock dividers are being used entirely there. We can also conclude that where memory latency is critical Wiz can have a big advantage.
Honestly, I think what these tests most illustrate is that synthetic benchmarks are misleading. Most of these tests aren't exercising memory latency at all and the one that is is a lot. In the real world very few useful programs will run completely out of cache and will only perform write-only transactions to memory, especially not for a 16KB data cache. Most will have a bigger working set and some number of dcache misses, whose latency can't be hidden at all on this architecture. The overall ratio of memory to cache/ALU will vary, as will the access pattern - higher spatial locality of reference will reduce cache misses.
Another factor is code size. Most likely, most of these tests fit in icache. Something that doesn't will have additional memory read (and hence memory latency) requirements. This is usually not a dominating factor in a program but it will often have at least some effect.
So I think we won't see the nearly 20% disadvantage for Caanoo in most real world programs, but I also think you won't tend to see Caanoo come out on top either. Lordus's original 10% numbers for the emulators he tested fits pretty nicely into this.
Exophase
Nothing good will ever come of Exophase.
Can you link me to where someone has done an apples to apples test of pcsx4all on Caanoo vs Wiz?
ruffnutts
Active Member
Pcsx4All is deffo faster on firmware 1.0.6 not sure by how much though
BAFelton
Well-Known Member
Exophase said:
I made this vid a few days ago. But I didn't check the FPS
Last edited by a moderator:
Exophase
Nothing good will ever come of Exophase.
Thanks BAFelton. It's unfortunately difficult to visually conclude much from that test. Are you using the same type of SD card for both?
Exophase
Nothing good will ever come of Exophase.
CKeichel said:
Strange how you can make that claim without any meaningful data. It seemed for a lot of it the speed was throttled, and where things really got ahead one way or the other (and yes, that included the Wiz getting ahead) was during transitions. Which would probably be highly affected by CD read time. Which, zomg, would be highly affected by SD speed, hence why I asked?
Last edited by a moderator:
One thing I noticed is the Caanoo has a blue gradient background where as the Wiz is black (on black or dark graphics). What it up with that? Is that some weird thing different in the emulator or something? It is quite different, not just a litle.BAFelton said:Exophase said:
I made this vid a few days ago. But I didn't check the FPS
Last edited by a moderator:
Exophase said:
Meaningful data? You mean apart from the fact, that we can clearly see it? What do you mean by transitions, a lot of what we saw were videos, while playing, there is a constant data reading from the card.
In my experience at least the Caanoo reads not very fast from a SD card, so the effective data transfer time from different SD cards is often the same, even if the cards have different speeds. If it would be only the SD speed one device would always be faster then the other.
Apart from that, the PSX had a 2x CD-Rom, the maximum speed was 300 KB/s, the videos are played with 25 FPS, so there shouldn't be a change in places.
Last edited by a moderator:
Exophase
Nothing good will ever come of Exophase.
You can clearly see what? That they're throttled to the same speed for most of the time when running and one only gets ahead of the other after things fade to black and change to a different scene? Of course, it happens that both get ahead of each other. Then when it gets to the actual slow parts things are already long since desynched due to buttons being pressed at different times. So yes, it's really hard to conclude anything meaningful from this.
SD card reading can still cause sudden lag even for emulating a CD, especially if there's preloading going on in the OS. Ask notaz about the problems he had with Sega-CD which is a 1x CD.. Of course this won't make the emulator "faster" except for during times people usually won't even care about, but for this "test" it might cause you to draw the conclusions you are.
SD card reading can still cause sudden lag even for emulating a CD, especially if there's preloading going on in the OS. Ask notaz about the problems he had with Sega-CD which is a 1x CD.. Of course this won't make the emulator "faster" except for during times people usually won't even care about, but for this "test" it might cause you to draw the conclusions you are.
Exophase said:
First of all, if it is during video streaming like here, I wouldn't say, it is something people usually won't care about.
You say, "it is long since desynched due to buttons being pressed at different times", we can see, the buttons are pressed at once to start the emulation, there may be a delay of a few milliseconds, but nothing, that can cause this lag.
Afterwards, no buttons are pressed for a long time.
You can clearly see, that the anti piracy logo appears faster on the Caanoo and is displayed not as long as on the Wiz, in this moment, the Caanoo is far ahead of the Wiz, the first video starts on the Caanoo about half a second earlier, then on the Wiz, but if you pay attention, you will see, the second video starts almost in sync with the Wiz. This is one of the reasons, why I said this is strange. Because, if you continue to watch the video, you can see, that the Wiz is now overtaking the Caanoo. This isn't something, that is happening in the transitions between the videos, but it is the video itself, that is now played faster on the Wiz, then on the Caanoo.
Then, strange again, the Wiz enters the menu screen about a second earlier, then the Caanoo, but the demo on both machines again starts almost synched again and in this demo, the Caanoo overtakes the Wiz another time.
I found that rather odd, nothing I can explain with different card speeds only.
Last edited by a moderator:
Exophase
Nothing good will ever come of Exophase.
Seriously CKeichel, I'm sorry to say this but this comparison is bullshit. There's no point drawing a lot from it, especially when you're looking at how fast something at the VERY START happens, which is undoubtedly due to the Caanoo loading it faster for whatever reason, not emulating it faster. You say the gameplay stuff and not transitions/CD-ROM accessing is going faster but I don't see it. Why not wait for actual FPS comparisons for something that runs slow?
Exophase said:
Sorry to say this is bullshit, the loading procedures are part of the emulation. If the loading on the Caanoo is faster, then the emulation runs faster. Period.
You should better read, what I wrote Exophase, I never talk about gameplay stuff, but about videos, but video playback is an integral part of any Playstation emulation, this was what I talked about.
I really don't understand, why you don't want to see, what the video shows, I never said, the Caanoo is here faster then the Wiz in general. I only pointed out, that I find it strange, that sometimes the Caanoo is faster and sometimes the Wiz.
And the fact, that the video with the Umbrella Logo runs considerably faster on the Wiz, is undeniable.
You say wait for actual FPS comparsions, this is fine for me, all I pointed out was, that I found the behaviour in this particular case strange. You seem to try to make something different of my words.
We can surely say, your explanation with the different speed of the SD cards is wrong, cause it would put one device in front all the time, there wouldn't be a change.
PS: If the Caanoo loads faster then the Wiz, it would lead to faster emulation in certain cases, something which you try to proof wrong since the Caanoo was launched.
Last edited by a moderator:
RatoZumbi said:
Of what use should an FPS counter be here? You can see the different speeds with the naked eye, the question is, why are both devices running the same games with varying speeds, why is the Caanoo sometimes faster and then the Wiz? An FPS-Counter wouldn't help to answer this questions.
Last edited by a moderator:
DaveC said:
That's AMOLED (Wiz) vs TFT (Caanoo). AMOLED got a much higher contrast since there is no back light. Each pixel is its own light source. It also got a much higher viewing angle (about 170° in both directions, I'd say).
Well, AMOLEDs are more expensive, have a shorter lifetime, and a higher failure rate. The smallest imperfection can reduce its lifetime drastically. You can also get burn-in fairly quickly.
It's a pretty new technology. In a few years AMOLEDs will be a lot better. Right now they are somewhat meh... but at least the image is very pretty and amazingly crispy.
---
Looking at those numbers, the Caanoo looks pretty disappointing. Not really much of an upgrade. They should have skipped that one. Without GLES 2.0 and without an FPU it's not very interesting hardware. I don't need another Wiz.
Last edited by a moderator: