Since the rendering techniques are so different this comparison might be futile, but I thought it was interesting that the old GeForce4 MX 460 put out about 38 Million triangles per second. Some of the mobile GPU's are getting up there, and the MP2 of course is already well past it with 65 million.
Triangles really aren't a good axis to compare one once you start getting to GPUs more than 7 or 8 years old. Although geometry processing has surged in the most recent generation of cards, it only increased bot about 2x between the GF5xxx and 2xx series. Improved texture, shading, and anti-aliasing capability returned much better improvements in image quality. Mobile GPUs probably won't evolve quite the same way (higher DPI makes AA much less needed); but they're getting to the point where geometry is going to be mostly maxed out in terms of farther returns for a while.
I know that ARM is generally very good on power utilization, but do we have any power utilization comparisons of these GPUs (or the whole SOC/phone). 2x-4x the performance is great, if it doesn't kill your battery life.
Also, any explanation for the difference in performance between the Nexus S (1GHz Hummingbird and an SGX 540) vs the Optimus 3D (1 GHz OMAP4430 + SGX 540) or the Infuse 4G (1.2GHz Hummingbird + SGC 540)? Different GPU clocks? Different number of GPU cores?
The article/review of the actual phone is clearly linked above and if you bothered to look it you'll see that the battery performance is actually better than most other smartphones, even with this being the fastest phone/SoC on the planet. They might not have battery performance tests for playing games but the GPU is still used when using the browser and other stuff, which was tested in the review.
<i>and if you bothered to look it you'll see that the battery performance is actually better than most other smartphones, even with this being the fastest phone/SoC on the planet. </i>
I always find it interesting how there's always someone who jumps in on a poster with a smart-ass comment. It would be so much easier for you to skip the post and shut the hell up. Ooops, look at me; I'm doing it too.
I do believe he succeeded in his goal then. People who post moronic comments will not understand how moronic those comments are until someone takes it to the extreme so that they can get a taste of their own medicine, per se. His comment effectively points out how dumb those types of comments are. Particularly the one being sarcastic about him not looking at the link. Gaming deserves its own benchmarks on battery life.. the way gaming uses a GPU is nothing like how a browser uses it.
If I were interested in reading about the phone, I would have read the article. This was a separately published link with the title "ARM's Mali-400 MP4 is the Fastest Smartphone GPU...for Now". If it's about which phone is fastest, then the title is misleading. I came to find about CPUs and GPUs, not about specific phone models.
It would be a challenge, but what I would really love to see is a ballpark comparison of these mobile GPUs to their desktop counterparts.
The question that bugs the living juices out of me when seeing these kinds of reviews is: “What year of desktop graphics hardware are we approximately at on mobiles?”
Does anyone know of any comparisons of that nature?
Very nice! The only thing faster this year will most likely be the A5 in the iPhone 5 since OMAP5, Krait and Tegra 3 AFAIK won't be available in a smartphone this year.
According to this Engadget article, the ARM T604 is likely going to be paired with the Cortex-A15 for late 2012, which means we won't be seeing in the Galaxy S III, but maybe the S IV in 2013.
iPhone 5 could be faster, it depends how much they've had to scale back the performance to make the A5 work in a phone form factor. They have the advantage of being last out the gate and can tune to beat this, if they can find the power/thermal headroom.
The best thing I love about these comments and this site is that for the average person none of this matters. It is only important to guys who still live in their parents basement, have a half eaten pizza next to their 2 liter bottle of diet Pepsi, and a tube of Clearasil on the shelf. Anantech has been a great source of info to me in many ways. Long live the geeks. We shall inherit the Earth. By the way, I do like girls and talk to them frequently. Have been know to get intimate from time to time as well. :) Also, I know what the sun looks like. :)
Why is the Sensation missing form gl2.1 benches? Also, no where in the vellamo chart does the sgs2 get anywhere close to 100% faster than the evo 3d. I'm a huge adreno fan, and feel that it performs WAY better than most people give it credit for, usually due to it's higher resolution in tests that can't be ran off screen.
Earlier version of glbechmark has the following result for IPAD-2 offscreen test @ 1280x720 Egypt offscreen: 38 PRO offscreen: 55
the new version from this report indicates 85/148.
I suspect that the new version result on IPAD-2 is for 800x480 resolution and not 1280x720.Could be a bug in iOS port of the new GLbenchmark version... result is very suspicious considering it is almost 3X higher than previous version number.
IPAD-2 results @ 1280x720 offscreen with vsync should be as below: Egypt ~ 50 PRO ~ 102 which is still better than galaxy S2. most likely latest glbenchmark has bug in iOS port... watch out new release.
yes, it did have. what they had was an offscreen 720p test (but still with vsync on).
Egypt was ~ 38FPS for 720p offscreen with vsync on, on the IPAD-2 The next version of glbenchmark removes the vsync on, and the result now shoots up to 85.7! this is too good to be true. For people familiar with opengl, onscreen tests require eglswap() to be called, for offscreen test if you remove the eglswap() and do glflush instead, you can get the FPS without vsync limitation.
But there is no guarantee that the underlying drivers and h/w do the same work as onscreen test. For e.g. a smart driver can figure out that it gets a bunch of render buffer commands routed to offscreen FBOs, but the result is not even used (no pixel read calls or texture loads from these FBOs), and so can skip the rendering.
Unless you have tools to extract frames out of the GPU, you can never be sure if the results of these offscreen tests make sense or not.
So, i dont think it is right to assume that IPAD-2 GPU is much better than Mali400MP just by looking at one data point from glbenchmark.
I'm wondering Why do not include the Motorola Atrix in your investigation.
Specially now when Google had bought Motorola. No matter what, I think Atrix must be a reference in this stuff, because Google is owner of Android and Motorola.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
30 Comments
Back to Article
tipoo - Sunday, September 11, 2011 - link
Since the rendering techniques are so different this comparison might be futile, but I thought it was interesting that the old GeForce4 MX 460 put out about 38 Million triangles per second. Some of the mobile GPU's are getting up there, and the MP2 of course is already well past it with 65 million.DanNeely - Sunday, September 11, 2011 - link
Triangles really aren't a good axis to compare one once you start getting to GPUs more than 7 or 8 years old. Although geometry processing has surged in the most recent generation of cards, it only increased bot about 2x between the GF5xxx and 2xx series. Improved texture, shading, and anti-aliasing capability returned much better improvements in image quality. Mobile GPUs probably won't evolve quite the same way (higher DPI makes AA much less needed); but they're getting to the point where geometry is going to be mostly maxed out in terms of farther returns for a while.gstrickler - Sunday, September 11, 2011 - link
I know that ARM is generally very good on power utilization, but do we have any power utilization comparisons of these GPUs (or the whole SOC/phone). 2x-4x the performance is great, if it doesn't kill your battery life.Also, any explanation for the difference in performance between the Nexus S (1GHz Hummingbird and an SGX 540) vs the Optimus 3D (1 GHz OMAP4430 + SGX 540) or the Infuse 4G (1.2GHz Hummingbird + SGC 540)? Different GPU clocks? Different number of GPU cores?
jordanclock - Sunday, September 11, 2011 - link
Differences in the CPUs and driver efficiencies.B3an - Sunday, September 11, 2011 - link
The article/review of the actual phone is clearly linked above and if you bothered to look it you'll see that the battery performance is actually better than most other smartphones, even with this being the fastest phone/SoC on the planet. They might not have battery performance tests for playing games but the GPU is still used when using the browser and other stuff, which was tested in the review.room200 - Sunday, September 11, 2011 - link
<i>and if you bothered to look it you'll see that the battery performance is actually better than most other smartphones, even with this being the fastest phone/SoC on the planet. </i>I always find it interesting how there's always someone who jumps in on a poster with a smart-ass comment. It would be so much easier for you to skip the post and shut the hell up. Ooops, look at me; I'm doing it too.
ph0tek - Sunday, September 11, 2011 - link
And yet you post the most stupid and pointless comment here. Nice one moron.coder543 - Sunday, September 11, 2011 - link
I do believe he succeeded in his goal then. People who post moronic comments will not understand how moronic those comments are until someone takes it to the extreme so that they can get a taste of their own medicine, per se. His comment effectively points out how dumb those types of comments are. Particularly the one being sarcastic about him not looking at the link. Gaming deserves its own benchmarks on battery life.. the way gaming uses a GPU is nothing like how a browser uses it.SlyNine1 - Monday, September 12, 2011 - link
Until you look at the ones preceding it. AKA Mine, Yours and anyone else that keeps this crap going.gstrickler - Sunday, September 11, 2011 - link
If I were interested in reading about the phone, I would have read the article. This was a separately published link with the title "ARM's Mali-400 MP4 is the Fastest Smartphone GPU...for Now". If it's about which phone is fastest, then the title is misleading. I came to find about CPUs and GPUs, not about specific phone models.bh64 - Sunday, September 11, 2011 - link
It would be a challenge, but what I would really love to see is a ballpark comparison of these mobile GPUs to their desktop counterparts.The question that bugs the living juices out of me when seeing these kinds of reviews is: “What year of desktop graphics hardware are we approximately at on mobiles?”
Does anyone know of any comparisons of that nature?
DanNeely - Sunday, September 11, 2011 - link
You could go digging for old quake 3 benches and see what cards gave similar perormance levels.Mike1111 - Sunday, September 11, 2011 - link
Very nice! The only thing faster this year will most likely be the A5 in the iPhone 5 since OMAP5, Krait and Tegra 3 AFAIK won't be available in a smartphone this year.ph00ny - Sunday, September 11, 2011 - link
It's not just those two in the running. ARM already announced the T604 which is supposedly several times faster than 400MPGnillGnoll - Monday, September 12, 2011 - link
ARM also said they expect T604 to be available in devices in late 2012 at the earliest.dagamer34 - Monday, September 12, 2011 - link
According to this Engadget article, the ARM T604 is likely going to be paired with the Cortex-A15 for late 2012, which means we won't be seeing in the Galaxy S III, but maybe the S IV in 2013.http://www.engadget.com/2010/11/10/arm-intros-next...
jlago - Sunday, September 11, 2011 - link
iPhone 5 could be faster, it depends how much they've had to scale back the performance to make the A5 work in a phone form factor. They have the advantage of being last out the gate and can tune to beat this, if they can find the power/thermal headroom.inplainview - Sunday, September 11, 2011 - link
The best thing I love about these comments and this site is that for the average person none of this matters. It is only important to guys who still live in their parents basement, have a half eaten pizza next to their 2 liter bottle of diet Pepsi, and a tube of Clearasil on the shelf. Anantech has been a great source of info to me in many ways. Long live the geeks. We shall inherit the Earth. By the way, I do like girls and talk to them frequently. Have been know to get intimate from time to time as well. :) Also, I know what the sun looks like. :)Arnulf - Sunday, September 11, 2011 - link
You are so cool !!! May I adorn myself in your awesome stellar glow ?inplainview - Monday, September 12, 2011 - link
I prefer money. Nothing else will do... Now move along. You are not nearly as funny as me.Stormkroe - Sunday, September 11, 2011 - link
Why is the Sensation missing form gl2.1 benches? Also, no where in the vellamo chart does the sgs2 get anywhere close to 100% faster than the evo 3d. I'm a huge adreno fan, and feel that it performs WAY better than most people give it credit for, usually due to it's higher resolution in tests that can't be ran off screen.sprockkets - Sunday, September 11, 2011 - link
You need to read the article; they no longer have the phone to test with it.ab303 - Sunday, September 11, 2011 - link
Earlier version of glbechmark has the following result for IPAD-2 offscreen test @ 1280x720Egypt offscreen: 38
PRO offscreen: 55
the new version from this report indicates 85/148.
I suspect that the new version result on IPAD-2 is for 800x480 resolution and not 1280x720.Could be a bug in iOS port of the new GLbenchmark version... result is very suspicious considering it is almost 3X higher than previous version number.
iwodo - Monday, September 12, 2011 - link
May be worth Anand to look at it?ab303 - Tuesday, September 13, 2011 - link
IPAD-2 results @ 1280x720 offscreen with vsync should be as below:Egypt ~ 50
PRO ~ 102
which is still better than galaxy S2. most likely latest glbenchmark has bug in iOS port... watch out new release.
ab303 - Tuesday, September 13, 2011 - link
sorry meant with vsync OFFGnillGnoll - Wednesday, September 14, 2011 - link
Versions of GLBenchmark before 2.1.0 don't even have the offscreen test.ab303 - Friday, October 7, 2011 - link
yes, it did have. what they had was an offscreen 720p test (but still with vsync on).Egypt was ~ 38FPS for 720p offscreen with vsync on, on the IPAD-2
The next version of glbenchmark removes the vsync on, and the result now shoots up to 85.7! this is too good to be true. For people familiar with opengl, onscreen tests require eglswap() to be called, for offscreen test if you remove the eglswap() and do glflush instead, you can get the FPS without vsync limitation.
But there is no guarantee that the underlying drivers and h/w do the same work as onscreen test. For e.g. a smart driver can figure out that it gets a bunch of render buffer commands routed to offscreen FBOs, but the result is not even used (no pixel read calls or texture loads from these FBOs), and so can skip the rendering.
Unless you have tools to extract frames out of the GPU, you can never be sure if the results of these offscreen tests make sense or not.
So, i dont think it is right to assume that IPAD-2 GPU is much better than Mali400MP just by looking at one data point from glbenchmark.
Josemarti - Tuesday, September 13, 2011 - link
I'm wondering Why do not include the Motorola Atrix in your investigation.Specially now when Google had bought Motorola. No matter what, I think Atrix must be a reference in this stuff, because Google is owner of Android and Motorola.
I think you are in time to include it.
PS: Sorry my bad English.
randombytes - Thursday, November 10, 2011 - link
Was release in Oct, but it is using same A5 chip as of iPad2.