Async Cores and Benchmarks

[BlueGoldAce]

I know this has been discussed off and on in the Sensation vs Galaxy II thread, but I would like to have a more focused discussion on this.

It seems to me that the the async cores could be a plausible reason for the lowe (for example) quadrant scores. What do you guys think? I mean, how can mt4g and desire HD, overclocked, score as high as the stock sensation? I am currently a graduation student at the University of Notre Dame. I am pursing a degree in nueroscience...so this is not my field of study. But I am very good friends with a few of the computer science and tech Phds up here, and I asked them about this. One of them is a avid phone tech geek, and he said that this is exactly the reason for the low benchmarks...even with smartbench. The reasons he rambled off are over my head (again I'm nueroscience), and he talked to me about for at least an hour. His conclusion was that at the very least the modified A8 snapdragons are as powerful, or very close to the A9's on the Galaxy S II. He said the async actually should allow the phone to perform better, sync individual cores can be assigned individual task, thus allowing increased multitasking and prevent a bottleneck at the processor level.

He also mentioned some of samsung's implantations in the phone, such as the ridiculously high read speed on the sd card (artifically made, faster than most solid state drives...somewhere in the 400 mb/s) seem to have been intended for the purpose of benchmarks, due to such programing would not benefit real world use. Smart move by samsung, though, as it makes for pretty numbers and fanfare.

I mentioned the browser, for example, difference and he said this is the result of software alone. As such, HTC will remedy this. The graphic excel is great on the Galaxy S II, but if HTC doesn't match this...Ice Cream will. Still...in my opinion...great thinking on samsung's part in this aspect (even if it will be short lived)

For the purpose of this argument, lets put aside the broswer examples (which does not speak for the actually processing power of the phone, and will not be a long lasting advantage). Also, there is debate around the camera, but again we are talking processing power here. Beyond these two examples, I have not seen a comparision in which the galaxy S II was faster in opening apps, or flying between various parts of the UI.

Leave out the sam dudes (sorry, don't remember full name) comparisions. Beyond the fact that they focus on the browser, etc, there also seems to be some controversy around that.

Anyway, I wanted to see what you guys think.

Take care, and enjoy the discussion.

 

[brusko1972]

I am no genius when it comes to chip technologies, but I have been thinking that maybe the async cores of the sensation are not running on full throttle with those benchmarks.

 

[DuoM]

I know this has been discussed off and on in the Sensation vs Galaxy II thread, but I would like to have a more focused discussion on this.

It seems to me that the the async cores could be a plausible reason for the lowe (for example) quadrant scores. What do you guys think? I mean, how can mt4g and desire HD, overclocked, score as high as the stock sensation? I am currently a graduation student at the University of Notre Dame. I am pursing a degree in nueroscience...so this is not my field of study. But I am very good friends with a few of the computer science and tech Phds up here, and I asked them about this. One of them is a avid phone tech geek, and he said that this is exactly the reason for the low benchmarks...even with smartbench. The reasons he rambled off are over my head (again I'm nueroscience), and he talked to me about for at least an hour. His conclusion was that at the very least the modified A8 snapdragons are as powerful, or very close to the A9's on the Galaxy S II. He said the async actually should allow the phone to perform better, sync individual cores can be assigned individual task, thus allowing increased multitasking and prevent a bottleneck at the processor level.

He also mentioned some of samsung's implantations in the phone, such as the ridiculously high read speed on the sd card (artifically made, faster than most solid state drives...somewhere in the 400 mb/s) seem to have been intended for the purpose of benchmarks, due to such programing would not benefit real world use. Smart move by samsung, though, as it makes for pretty numbers and fanfare.

I mentioned the browser, for example, difference and he said this is the result of software alone. As such, HTC will remedy this. The graphic excel is great on the Galaxy S II, but if HTC doesn't match this...Ice Cream will. Still...in my opinion...great thinking on samsung's part in this aspect (even if it will be short lived)

For the purpose of this argument, lets put aside the broswer examples (which does not speak for the actually processing power of the phone, and will not be a long lasting advantage). Also, there is debate around the camera, but again we are talking processing power here. Beyond these two examples, I have not seen a comparision in which the galaxy S II was faster in opening apps, or flying between various parts of the UI.

Leave out the sam dudes (sorry, don't remember full name) comparisions. Beyond the fact that they focus on the browser, etc, there also seems to be some controversy around that.

Anyway, I wanted to see what you guys think.

Take care, and enjoy the discussion.

To put it simply either the Sensation is less powerful than the other dual core devices or HTC have done something wrong with the implementation of the hardware and/or the software.

Benchmarking software like Smartbench 2011 works just fine with multiple cores as does the version of Android that is shipped on the Sensation.

So the question is this, if the Sensation is more powerful than the benchmarks show, as some people believe it is, then why would they expect it to reach maximum performance when a user is going about their normal everyday tasks but not when a benchmark program is burning up the phone?

Obviously I'm looking at the phone as a whole since we can't just test the CPU alone but when talking about the CPU itself, like I said in another thread I always expected the Qualcomm dual-core offering to be No. 4 when it came to raw power. Behind Tegra 2, Exynos, and OMAP.

However it is strange how much further behind it scores in benchmarks that don't involve the screen. Is it possible that HTC used cheaper components in the phone to maximise profits and this is showing in the benchmarks?

I was set on getting the HTC Incredible S until the bootloader proved to be too much (three months now). One thing that was strange to see with that phone was that it was consistently benchmarking around the 1500 mark on Quadrant when last year's HTC Desire HD (Same SoC) would benchmark at around 2000 or more.

In other tests that are CPU rather then GPU based the Incredible S was behind again:

http://www.cnet.com.au/htc-incredible-s-339310045.htm

Perhaps it isn't different components but a change made by HTC to reduce CPU power but to increase battery life.

I don't know what it is but starting with the Incredible S new HTC phones have been getting lower scores expected even against earlier HTC phones using the same SoC.

 

[BlueGoldAce]

That idea is quite plausible.

But wouldn't it be qualcom who makes the processor? Unless you mean the components that bridge the rest of the phone together.

But could it also be the design/intent of the processor? I thinking like this. Single core processors are ample fast for the majority of android use. But when one is multitasking while performing other task, such as web browsing, a single core can struggle. So what if HTC designed the cores as such that allow to process independently. The theory is that maybe one of the cores is maxing out while the other is doing very little. This would mean that qualcom put restriction on the hardware so that when you are doing something like, say browsing a flash filled site, you still have ample power to perform other task. Does that make sense? This is a very basic version of the reason the Professor gave to me. He specializes in microchips, such as we see on mobile phones.

He also noted that the currents phones are nothing compared to what technology exisit, but won't be release anytime soon. Makes sense, keep slowly uping the phones = more profits. There is nothing to gain in releasing a super phone that will last 5 or + years.

 

[boostedb16b]

asynce is quite like your professor has explained to you one processor can be at 1.2ghz and the other be from 0 to 1.2ghz depending on the load

 

[brusko1972]

well, I am still getting the sensation when sim free handsets becomes available. I am upgrading from a touch hd wm 6.1 so I dont think a dual core will dissapoint me no matter buggy

 

[BlueGoldAce]

Overall...I understand the theory very well. I am just giving him credit. I had assumed this, based of some of my reading, but I didn't post since I don't have the credientials to make these claims...but he does. Now if you need some advice involving the nervous system, I'm you guy. Of course...I only have a bachelors...still a few years away from my PhD. Maybe I could then follow that up with a degree in computer science, you know essential make a career out of being a student .

 

[ceroglu]

Imho, htc become very greedy and they basically dont give a crap about the benchmarks and reviews anymore. This is why they locked the bootloaders and probably why they clocked down the processor to 1.2. In a way they must be thinking that they have proven themselves in the industry with quality products and they already have a customer base that who cares about their reputation not the benchmark scores. So I am hoping the petition for the locked bootloders and their initial sales will indicate that they are not alone in market anymore and people actually care about the preformance of their devices. May be they will get back into the game once again.

 

[BlueGoldAce]

Imho, htc become very greedy and they basically dont give a crap about the benchmarks and reviews anymore. This is why they locked the bootloaders and probably why they clocked down the processor to 1.2. In a way they must be thinking that they have proven themselves in the industry with quality products and they already have a customer base that who cares about their reputation not the benchmark scores. So I am hoping the petition for the locked bootloders and their initial sales will indicate that they are not alone in market anymore and people actually care about the preformance of their devices. May be they will get back into the game once again.

Benchmarks are not everything. If a piece of hardware can perform as well, or even better, but doesn't display a long bar in quadrant, who cares? HTC Sensation does beat the Galaxy II in linpack. Yes it has a slightly lower FPS, but it has a higher resolution.

I doubt benchmarks are at the forefront of there minds. The majority of the market doesn't even know what quadrant is.

 

[DuoM]

That idea is quite plausible.

But wouldn't it be qualcom who makes the processor? Unless you mean the components that bridge the rest of the phone together.

But could it also be the design/intent of the processor? I thinking like this. Single core processors are ample fast for the majority of android use. But when one is multitasking while performing other task, such as web browsing, a single core can struggle. So what if HTC designed the cores as such that allow to process independently. The theory is that maybe one of the cores is maxing out while the other is doing very little. This would mean that qualcom put restriction on the hardware so that when you are doing something like, say browsing a flash filled site, you still have ample power to perform other task. Does that make sense? This is a very basic version of the reason the Professor gave to me. He specializes in microchips, such as we see on mobile phones.

He also noted that the currents phones are nothing compared to what technology exisit, but won't be release anytime soon. Makes sense, keep slowly uping the phones = more profits. There is nothing to gain in releasing a super phone that will last 5 or + years.

Nope, I was thinking HTC. Qualcomm may make the asych processor but how it is actually setup and how the software and other hardware components function with it is up to HTC. However it is possible that the design or feedback system is faulty so only one core is working or the second core never goes beyond 50% of the speed of the first core. This could also be down to configuration/other components in the phone, which would be down to HTC.

I don't think that Qualcomm or HTC would have deliberately limited the way the cores can function in the way that you suggest (beyond making them async, which can most likely be turned off too). As I understand it doing something like playing a flash video only uses one core anyway (due to the nature of flash). With a phone like the Sensation this should mean that even if core 1 is at 1.2GHz, core 2 could be at 300 MHz or whatever is required to fetch your email, run the phone, Wi-Fi, etc. With other dual-core implementations both cores will run at 1.2GHz, which will also allow you to do other stuff. The benefit of having async cores is supposed to be battery life.

As for holding back stuff, this is how consumer electronics firms work. If you could get access you could go and look at the next few iterations of products which exist now in various forms and which will be released over the next few years unless there is major disruption to a market by something unexpected.

 

[BlueGoldAce]

Nope, I was thinking HTC. Qualcomm may make the asych processor but how it is actually setup and how the software and other hardware components function with it is up to HTC. However it is possible that the design or feedback system is faulty so only one core is working or the second core never goes beyond 50% of the speed of the first core. This could also be down to configuration/other components in the phone, which would be down to HTC.

I don't think that Qualcomm or HTC would have deliberately limited the way the cores can function in the way that you suggest (beyond making them async, which can most likely be turned off too). As I understand it doing something like playing a flash video only uses one core anyway (due to the nature of flash). With a phone like the Sensation this should mean that even if core 1 is at 1.2GHz, core 2 could be at 300 MHz or whatever is required to fetch your email, run the phone, Wi-Fi, etc. With other dual-core implementations both cores will run at 1.2GHz, which will also allow you to do other stuff. The benefit of having async cores is supposed to be battery life.

As for holding back stuff, this is how consumer electronics firms work. If you could get access you could go and look at the next few iterations of products which exist now in various forms and which will be released over the next few years unless there is major disruption to a market by something unexpected.

So you are positive the benchmarks are reading it correctly? The async cores are new, so the benchmarks may not fully utilize them.

I hope you are wrong about the quality of the phone, beyond the processor.

Sense could be the problem it as well. I know with the Evo, take off sense and you score in the 1500s, with sense stock you get 1000.

 

[Killbynature]

Been saying this for the longest also to mention since gingerbread using ext 4 with Samsung internal memory about 16 gigs to can easily make quadrant ridiculous high by getting a ridiculously fast sd card partition ext4 and get data2ext 4 on the sd card. Samsung could of did the same with their rom with their internal memory and rom. All is theory but in test we really won't know unless we test to computers. Which we can perform better test on.

Sent from my demonSPEED Glacier using XDA Premium App

 

[DuoM]

So you are positive the benchmarks are reading it correctly? The async cores are new, so the benchmarks may not fully utilize them.

I hope you are wrong about the quality of the phone, beyond the processor.

Sense could be the problem it as well. I know with the Evo, take off sense and you score in the 1500s, with sense stock you get 1000.

I believe that they are reading it correctly but that doesn't mean that the phone is performing correctly. The benchmarks just fire off a number of threads, in the case of Smartbench 2011 four threads, which should engage both processors since the benchmarks are designed to stress the phone.

If that doesn't get both cores to respond, what will?

Sense could be the problem. It just gets heavier and heavier. I know that some people are buying this phone because of Sense but for me it is just bloatware now. Some things it adds are nice but it seems that instead of improving the areas where it adds true functionality e.g. email and SMS, HTC have added more eye candy.

I didn't know about the EVO benchmarks though.

 

[DuoM]

Been saying this for the longest also to mention since gingerbread using ext 4 with Samsung internal memory about 16 gigs to can easily make quadrant ridiculous high by getting a ridiculously fast sd card partition ext4 and get data2ext 4 on the sd card. Samsung could of did the same with their rom with their internal memory and rom. All is theory but in test we really won't know unless we test to computers. Which we can perform better test on.

Sent from my demonSPEED Glacier using XDA Premium App

The author of Smartbench was in another thread, I don't know if you saw his comments. Anyway he said that he had reduced the weighting of I/O in Smartbench to prevent fast I/O for skewing the results so much.

Also because the results are so much lower than expected he is also going to take a look at the Sensation just to be sure that his software is reporting correctly with that phone.

 

[tolis626]

I believe that they are reading it correctly but that doesn't mean that the phone is performing correctly. The benchmarks just fire off a number of threads, in the case of Smartbench 2011 four threads, which should engage both processors since the benchmarks are designed to stress the phone.

If that doesn't get both cores to respond, what will?

Sense could be the problem. It just gets heavier and heavier. I know that some people are buying this phone because of Sense but for me it is just bloatware now. Some things it adds are nice but it seems that instead of improving the areas where it adds true functionality e.g. email and SMS, HTC have added more eye candy.

I didn't know about the EVO benchmarks though.

Bullshit.All you guys blame Sense for lags etc,but it's not the case.Not completely at least.
Try a Desire HD.Single core an all as it is,Sense 3 runs like heaven,nicer than Sense 2-2.1 .So Sense doesn't **** up your phone that much.
If you want my opinion,wait till some respectable devs get their hands on the phone(Not that devs who already have it aren't respectable,but we can't know until we have something custom made ).Then,if and when they start tinkering with it,getting it to perform as it should,we'd see everything work better.Maybe it's something in the kernel limiting the CPU or something.And even GPU benchmarks aren't fair,as HTC's devices' GPUs always underperform with the default Ondemand governor.It would only be fair if we benchmarked both phones(GS2 and Sensation) rooted,on stock roms with performance governor.Only then we'll be comparing fairly.

 

[tolis626]

Also,we could wait until the authors of Linpack update it to support multi-threading to further test the CPU.

 

[Beefheart]

I can't help but get the feeling this thread is grasping at straws somewhat. It could simply be a case of the Sensation not being as fast as people believed it would be. We shall see if the developer of Smartbench finds changes he needs to make, the next version should tell us where the Sensation stands and if this Async argument holds water.

 

[Acei]

Also because the results are so much lower than expected he is also going to take a look at the Sensation just to be sure that his software is reporting correctly with that phone.

Correct.

And here are the series of charts I have put together so far:

This one shows the progress of Productivity Index over time: http://bit.ly/lWC3gb
And this one shows the progress of Games Index: http://bit.ly/jBu8Lx

Interesting? May be. Perhaps you guys can tell me what you see in these.

I will also plot similar charts for the individual tests as well (this helps because each tests are supposed to stress the CPU/GPU in its own unique ways). I've looked at it briefly, but I can already see few strange results within. As soon as I find more time, I'll post more articles on the above site.

EDIT: Fixed the link to the games index page.

 

[steve austin]

What is is maybe what is...

Been saying this for the longest also to mention since gingerbread using ext 4 with Samsung internal memory about 16 gigs to can easily make quadrant ridiculous high by getting a ridiculously fast sd card partition ext4 and get data2ext 4 on the sd card. Samsung could of did the same with their rom with their internal memory and rom. All is theory but in test we really won't know unless we test to computers. Which we can perform better test on.

Sent from my demonSPEED Glacier using XDA Premium App

Ok, initial reports were that because the dual cores are so ne that the quadrant trackers could not "properly" measure. HOWEVER, no mater what review I have seen so far comparing the Galaxy and the Sensation, the Sensation ALWAYS looses. If a top of the line sports car cna go from 0-60 at .X seconds faster than the other it is something that is going to be touted and used as a benchmark. Therefore, i don't buy that the quadrant scored don't really matter all the time.

Now, more to your point about Samsungs speed. It seems like HTC, by the way they allocated memory, may have done the phone a speed disservice.


"The HTC Sensation measures in at 4.96 x 2.57 x 0.44 inches and 5.22oz is what it weighs. You’ll find that the unibody metal chassis combined with the soft-touch plastic inserts make for a rather comfortable phone to hold, and nice to rub up against your beard if you’ve got one. So smooth! The back is creak-free, the entirety of the battery cover coming off at once, the battery, SIM card, and microSD card popping in and out without an issue like they truly should.

Flaws in this whole situation include the fact that while the Galaxy S II pairs its 1.2GHz dual-core with 1GB of RAM, the HTC Sensation uses just 768MB. Samsung’s device has 16GB of memory packed in, while HTC makes due with just 1GB. There’s an 8GB microSD card in there, but you can add a bunch more memory to the GSII the same way if you want. Davies notes the truth: “here’s just no way that even a fast microSDHC card can be as quick as NANDFlash connected directly to the logic board, and that leaves the Sensation at a speed disadvantage.”

http://androidcommunity.com/htc-sensation-review-via-slashgear-20110523/

 

[DuoM]

Correct.

And here are the series of charts I have put together so far:

This one shows the progress of Productivity Index over time: http://bit.ly/lWC3gb
And this one shows the progress of Games Index: http://bit.ly/lWC3gb

Interesting? May be. Perhaps you guys can tell me what you see in these.

I will also plot similar charts for the individual tests as well (this helps because each tests are supposed to stress the CPU/GPU in its own unique ways). I've looked at it briefly, but I can already see few strange results within. As soon as I find more time, I'll post more articles on the above site.

Thanks for this. What I see in this is that the Sensation is simply not as powerful as the Galaxy S II.

BTW Sensation is a released device in Europe and the later batch of scores you are seeing are from the final released device. I believe that the UK got it on Thursday the 19th of May but I it is in several countries now.

I also noticed that you posted this score: "Samsung Galaxy S2 at 1GHz: 3229", do you have figures for it at 1.2GHz?

The links you posted both go to the Productivity Index, do you have the link for the Games Index?

Thanks again.