gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sat Dec 10, 2011 3:47 am

What is the performance of the Raspberry Pi? Specially, the \"Dual Core VideoCore IV® Multimedia Co-Processor\" is not exactly helpful. I\'m not expecting it to run Crysis, but what about simpler games like Minecraft*?

*I know minecraft is for java, I\'m just using it as a graphical comparison.

semisight
Posts: 16
Joined: Thu Dec 01, 2011 1:54 am
Contact: Website

Re: Performance

Sat Dec 10, 2011 4:00 am

It has been said to have roughly the GPU performance of a first generation XBox. It has been shown to run Quake III at full HD. (It is the most interesting computing platform in the world...)

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sat Dec 10, 2011 4:33 am

[[[ Please disregard this speculation and see my updated post on page 2 ]]]

semisight
Posts: 16
Joined: Thu Dec 01, 2011 1:54 am
Contact: Website

Re: Performance

Sat Dec 10, 2011 7:46 am

Well, yes. The Pi has a less technically advanced core than the iPhone. That\'s what you get for $25.

None of these are really brute force statistics. Extensions = polish, and the rest are just because the Pi displays at a greater resolution than the iPhone. What you really want to see to \"compare\" the two are stats like vertices/polygons per second. The first gen xbox can do 115M vertices/s, while the SGX543 in the iPhone 4S can do 105M vertices/s effective.

SteGriff
Posts: 8
Joined: Mon Nov 21, 2011 2:34 pm
Contact: Website

Re: Performance

Sat Dec 10, 2011 9:41 am

I am not expecting Minecraft to run with any kind of fluidity on the RasPi if you can get it to run at all (is there enough RAM?) Minecraft not only puts a lot of pressure on video and processing, but also storage; often the hard disk access is a substantial bottleneck when running Minecraft on a PC. I\'m not sure how your average SD card will compare in terms of R/W speeds.

Thought I\'d mention it since you brought Minecraft up. The rest of your questions have been addressed very well above.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Sat Dec 10, 2011 9:54 am

Also note that the Videocore IV GPU uses a *lot* less power than than the Core in the latest iPhone4, but doesn\'t quite have the same performance

The camera ISP in the Videocore IV is better than iPhone, in fact I think I am right in saying it\'s one of the best available (same with the power requirements - best in class).
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sat Dec 10, 2011 9:59 am

[quote]Quote from jamesh on December 10, 2011, 09:54
Also note that the Videocore IV GPU uses a *lot* less power than than the Core in the latest iPhone4, but doesn\'t quite have the same performance

The camera ISP in the Videocore IV is better than iPhone, in fact I think I am right in saying it\'s one of the best available (same with the power requirements - best in class).[/quote]

I\'m absolutely not expecting an iPhone 4S performance out of a $25 / $35 credit card sized computer. That is just unreasonable.

However, I was hoping for more of the iPhone 3GS\'s graphical power. Seems it\'s a bit less than iPhone 3G.

Apologies for referencing with apple products, it\'s the easiest to compare for me.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sat Dec 10, 2011 10:12 am

Wait a moment, comparing triangles/s, the pi is better than the iPhone 4! Is this because of screen resolution or what? And if it is, is it retina display (960 by 640)?

ross
Posts: 13
Joined: Fri Aug 26, 2011 8:18 pm

Re: Performance

Sat Dec 10, 2011 12:03 pm

[quote]Wait a moment, comparing triangles/s, the pi is better than the iPhone 4! [/quote]
This isn\'t an apples-to-apples comparison (because the screen resolutions are quite different between all three) but should give a general flavour of the relative GPU performance:
http://www.glbenchmark.com/compare.jsp? ... iPhone%204

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Sat Dec 10, 2011 12:38 pm

That\'s quite interesting. The VC4 seem to really beat the iPhones up to the 4 (no results for the 4S). In fact much faster than I was expecting, with the obvious proviso that the Nokia phones have much slower CPU\'s as Symbian doesn\'t need so much horsepower as Apple OS.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

obarthelemy
Posts: 1407
Joined: Tue Aug 09, 2011 10:53 pm

Re: Performance

Sat Dec 10, 2011 1:29 pm

Just to be very clear, the Nokia uses a ideocore 4, in the above comparison.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sun Dec 11, 2011 12:45 am

Nokia: 230,400 pixels (640 by 360)
iPhone 4: 614,400 pixels (960 by 640)

That means, the score for the iPhone 4 must be multiplied by 614 400 / 230 400 = 2.66666667.

Now, a more accurate (but not totally accurate) result of performance: (rounded to make it more pretty)

Multi-textured fill rate: 392,000 (Pi) vs 741,000 (iPhone 4)

Flat shaded triangles (OpenGL ES 1.1): 10,300 (Pi) vs 19,800 (iPhone 4)

Textured Triangles: 17,200 (Pi) vs 17,400 (iPhone 4)

So, no, the VCS 4 does not beat the iPhone 4.

However, if the results are all using a common resolution (which I highly doubt), the VCS blows the i4 out of the water.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sun Dec 11, 2011 12:47 am

[quote]Quote from obarthelemy on December 10, 2011, 13:29
Just to be very clear, the Nokia uses a ideocore 4, in the above comparison.[/quote]
Ok, I guess we can expect performances higher than the nokia 701, since pi uses dual core?

obarthelemy
Posts: 1407
Joined: Tue Aug 09, 2011 10:53 pm

Re: Performance

Sun Dec 11, 2011 8:37 am

The Pi\'s ARM is not dual core.

gsh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 1601
Joined: Sat Sep 10, 2011 11:43 am

Re: Performance

Sun Dec 11, 2011 9:00 am

So most of those GLBenchmark tests actually run at a specified resolution (for example 512x512) therefore they are directly comparable...

Otherwise whats the point of a benchmark if I can create a phone with a 10x10 screen that theoretically outperforms a high end PC!!!

So yes the Videocore 4 performance is better than iPhone 4S, although they are getting closer (we\'ve had it running this performance for a couple of years! Shame its taken so long to get phones into the market!)
--
Gordon Hollingworth PhD
Raspberry Pi - Director of Software Engineering

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Sun Dec 11, 2011 10:32 am

[quote]Quote from gladoscc on December 11, 2011, 00:47
[quote]Quote from obarthelemy on December 10, 2011, 13:29
Just to be very clear, the Nokia uses a ideocore 4, in the above comparison.[/quote]
Ok, I guess we can expect performances higher than the nokia 701, since pi uses dual core?[/quote]

The Nokia phone uses a Nokia Arm (?) core + a separate Videocore GPU (2763). The Pi has a single Arm core built in the the same die as the Videocore GPU (2835). However, the Pi Arm core is slower than the Nokia one. (700 vs 1000?? Don\'t know the Nokia core speed)

So, there is less work required in getting data to and from the host processor as they share memory, so it can be a bit faster in that respect. However, the raw power of the GPU is the same.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

User avatar
Emanuele
Posts: 182
Joined: Wed Aug 03, 2011 5:28 pm
Contact: Website

Re: Performance

Sun Dec 11, 2011 10:53 am

[quote]Quote from gsh on December 11, 2011, 09:00
So most of those GLBenchmark tests actually run at a specified resolution (for example 512x512) therefore they are directly comparable...

Otherwise whats the point of a benchmark if I can create a phone with a 10x10 screen that theoretically outperforms a high end PC!!!

So yes the Videocore 4 performance is better than iPhone 4S, although they are getting closer (we\'ve had it running this performance for a couple of years! Shame its taken so long to get phones into the market!)

[/quote]

To be honest, I would expect just the opposite. What is the point of having a benchmark at 512x512 when what matters to both users and developers is full screen / native resolution performance?

Here is my speculation. The bottleneck in Raspi is going to be texture resolution (regardless of the OpenGL limit). Because of the titled architecure, you should be able to scale up final resolution to HDTV without impacting the frame rate too much. However, you would not be able to scale up the textures accordingly. Since the core is fast, you might be able to rely more on complex shaders and less on textures.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Sun Dec 11, 2011 11:43 am

[quote]Quote from Emanuele on December 11, 2011, 10:53
[quote]Quote from gsh on December 11, 2011, 09:00
So most of those GLBenchmark tests actually run at a specified resolution (for example 512x512) therefore they are directly comparable...

Otherwise whats the point of a benchmark if I can create a phone with a 10x10 screen that theoretically outperforms a high end PC!!!

So yes the Videocore 4 performance is better than iPhone 4S, although they are getting closer (we\'ve had it running this performance for a couple of years! Shame its taken so long to get phones into the market!)

[/quote]

To be honest, I would expect just the opposite. What is the point of having a benchmark at 512x512 when what matters to both users and developers is full screen / native resolution performance?

Here is my speculation. The bottleneck in Raspi is going to be texture resolution (regardless of the OpenGL limit). Because of the titled architecure, you should be able to scale up final resolution to HDTV without impacting the frame rate too much. However, you would not be able to scale up the textures accordingly. Since the core is fast, you might be able to rely more on complex shaders and less on textures.[/quote]

Yeah, 128 / 256MB of ram is a bit too small.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Sun Dec 11, 2011 11:56 am

[quote]Quote from Emanuele on December 11, 2011, 10:53
[quote]Quote from gsh on December 11, 2011, 09:00
So most of those GLBenchmark tests actually run at a specified resolution (for example 512x512) therefore they are directly comparable...

Otherwise whats the point of a benchmark if I can create a phone with a 10x10 screen that theoretically outperforms a high end PC!!!

So yes the Videocore 4 performance is better than iPhone 4S, although they are getting closer (we\'ve had it running this performance for a couple of years! Shame its taken so long to get phones into the market!)

[/quote]

To be honest, I would expect just the opposite. What is the point of having a benchmark at 512x512 when what matters to both users and developers is full screen / native resolution performance?

Here is my speculation. The bottleneck in Raspi is going to be texture resolution (regardless of the OpenGL limit). Because of the titled architecure, you should be able to scale up final resolution to HDTV without impacting the frame rate too much. However, you would not be able to scale up the textures accordingly. Since the core is fast, you might be able to rely more on complex shaders and less on textures.[/quote]

The ONLY way to compare performance of a GPU is to compare it with the same code at the same resolution. Because, as GSH said, a device with a lower resolution than another, but with the same GPU would appear faster - but in fact the GPU is the same speed. If you want to measure GPU performance that\'s what you need to do. If you want to measure DEVICE performance, that\'s different.

Texture size is also a different argument.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

User avatar
Emanuele
Posts: 182
Joined: Wed Aug 03, 2011 5:28 pm
Contact: Website

Re: Performance

Sun Dec 11, 2011 1:00 pm

@gladoscc More than the size of the memory, I was thinking about the bandwidth of the memory. We had some discussions about the raspi\'s memory bandwidth a while ago but, IIRC, we didn\'t end up with a number. I don\'t expect the memory on a raspi to be faster than the memory on an iphone.

@jamesh True, but my point is this: is the linked GLBenchmark testing the GPU or the whole graphics pipeline (CPU+GPU+RAM+display)? I would expect the latter.

gsh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 1601
Joined: Sat Sep 10, 2011 11:43 am

Re: Performance

Sun Dec 11, 2011 1:33 pm

What is bandwidth going to tell you, the raw bandwidth is 3.2GiB/s but it really depends upon the hardware that is using it what the actual throughput is going to be, somewhere between 2.1 and 2.9 GiB/s
--
Gordon Hollingworth PhD
Raspberry Pi - Director of Software Engineering

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Sun Dec 11, 2011 1:59 pm

[quote]Quote from Emanuele on December 11, 2011, 13:00
@gladoscc More than the size of the memory, I was thinking about the bandwidth of the memory. We had some discussions about the raspi\'s memory bandwidth a while ago but, IIRC, we didn\'t end up with a number. I don\'t expect the memory on a raspi to be faster than the memory on an iphone.

@jamesh True, but my point is this: is the linked GLBenchmark testing the GPU or the whole graphics pipeline (CPU+GPU+RAM+display)? I would expect the latter.
[/quote]

It\'s not quite as fast (but still pretty high compared with other SoC\';s).

The point about the GL benchmark is that by testing at 512x512, you can take away any issues you might have with the display, so you can remove that from the list of things being tested (since they vary dramatically from device to device). Of course the GPU is still driving a display, its just that the size being constant make a fairer test.

So the test does test CPU, GPU, bandwidth and display driving efficiency- although there isn\'t much CPU work required - the GPU does all the heavy lifting.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

ChrisR
Posts: 14
Joined: Tue Dec 06, 2011 3:11 pm

Re: Performance

Sun Dec 11, 2011 8:47 pm

[quote]Quote from jamesh on December 11, 2011, 10:32
The Nokia phone uses a Nokia Arm (?) core + a separate Videocore GPU (2763). The Pi has a single Arm core built in the the same die as the Videocore GPU (2835). However, the Pi Arm core is slower than the Nokia one. (700 vs 1000?? Don\'t know the Nokia core speed)
[/quote]

The Nokia hardware has a customized ARM11 at 1GHz, which is pretty much the same in terms of supported features as the core in the Pi, so is about 30% faster based on clock speed only but there may be a difference in memory performance which would probably impact a GL benchmark. Also, the Nokia system has 128MB of dedicated RAM attached to the GPU which is likely to be faster than the shared memory system used in the Pi. In total, the available headroom for graphics performance should be higher on the Nokia 701/700/603 family.

With that in mind, it\'s (IMO) very unlikely that the Pi will beat the 701 benchmarks, but there are significant architectural differences in both hardware and software which makes predicting Pi performance based on 701 performance rather troublesome.

gladoscc
Posts: 58
Joined: Sat Dec 10, 2011 3:32 am

Re: Performance

Mon Dec 12, 2011 9:53 am

[quote]Quote from ChrisR on December 11, 2011, 20:47
[quote]Quote from jamesh on December 11, 2011, 10:32
The Nokia phone uses a Nokia Arm (?) core + a separate Videocore GPU (2763). The Pi has a single Arm core built in the the same die as the Videocore GPU (2835). However, the Pi Arm core is slower than the Nokia one. (700 vs 1000?? Don\'t know the Nokia core speed)
[/quote]

The Nokia hardware has a customized ARM11 at 1GHz, which is pretty much the same in terms of supported features as the core in the Pi, so is about 30% faster based on clock speed only but there may be a difference in memory performance which would probably impact a GL benchmark. Also, the Nokia system has 128MB of dedicated RAM attached to the GPU which is likely to be faster than the shared memory system used in the Pi. In total, the available headroom for graphics performance should be higher on the Nokia 701/700/603 family.

With that in mind, it\'s (IMO) very unlikely that the Pi will beat the 701 benchmarks, but there are significant architectural differences in both hardware and software which makes predicting Pi performance based on 701 performance rather troublesome.

[/quote]
If that is true, then the graphical power of the pi is really disappointing.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 27390
Joined: Sat Jul 30, 2011 7:41 pm

Re: Performance

Mon Dec 12, 2011 10:14 am

[quote]Quote from gladoscc on December 12, 2011, 09:53
[quote]Quote from ChrisR on December 11, 2011, 20:47
[quote]Quote from jamesh on December 11, 2011, 10:32
The Nokia phone uses a Nokia Arm (?) core + a separate Videocore GPU (2763). The Pi has a single Arm core built in the the same die as the Videocore GPU (2835). However, the Pi Arm core is slower than the Nokia one. (700 vs 1000?? Don\'t know the Nokia core speed)
[/quote]

The Nokia hardware has a customized ARM11 at 1GHz, which is pretty much the same in terms of supported features as the core in the Pi, so is about 30% faster based on clock speed only but there may be a difference in memory performance which would probably impact a GL benchmark. Also, the Nokia system has 128MB of dedicated RAM attached to the GPU which is likely to be faster than the shared memory system used in the Pi. In total, the available headroom for graphics performance should be higher on the Nokia 701/700/603 family.

With that in mind, it\'s (IMO) very unlikely that the Pi will beat the 701 benchmarks, but there are significant architectural differences in both hardware and software which makes predicting Pi performance based on 701 performance rather troublesome.

[/quote]
If that is true, then the graphical power of the pi is really disappointing. [/quote]

Whuh? What do you mean? How can you make that assessment from the comment you quote?

In fact, the GPU in the 701 accesses memory in the same way as the BRCM2835, it\'s just that in the 2835, it has to share the bus with the Arm, so it that is doing a lot (and I mean a lot) of memory accessing for some reason then there may be an impact on the GPU memory access speed, but to be honest it won\'t be that great - you need to be really REALLY hammering it to have an noticable effect.

Haven\'t done the tests yet, but I think the Raspi benchmarks will be pretty much the same or better that the 701.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed.
I've been saying "Mucho" to my Spanish friend a lot more lately. It means a lot to him.

Return to “General discussion”