Something about the Low Linpack scores! - Galaxy S I9000 General

Hi All
I've found this article explaining why are the non Snapdragon devices scores lower (by much) !
http://www.greenecomputing.com/2010/09/24/mea-culpa-why-are-the-nexus-one-linpack-scores-so-much-higher-than-on-my-phone/

Yeah I remember reading something quite similar. Just simply that things aren't optimised enough for the hummingbird and are more optimised for the snapdragon. I think this is also the same reason to why JIT doesn't help much with the Galaxy S, although don't take my word for it.

Related

JP7 at Linpack

Found the following entry at: http://www.greenecomputing.com/apps/linpack/linpack-by-device/ (Samsung Galaxy S - Position 6)
1000.0MHz samsung/GT-I9000/GT-I9000/GT-I9000:2.2/FROYO/XXJP7:user/release-keys
I know thats possible to set the note manually, but maybe JP7 is inofficially out now.
Is there a download available?
See also: http://samsunggalaxysforums.com/showthread.php/1268-you-will-hate-me-for-this-but...
hmmm our galaxy benchmarks seems bit slow compare to desire doesn't it?
Think Froyo with JIT enabled will boost up our galaxy.
JIT is not working in JP1/2/3.
http://androidandme.com/2010/05/news/jit-performance-boost-coming-with-android-2-2/
Why do u think its not working?Look at the scores of Droid 2 which comes with FROYO. It has the same scores as SGS. Looks like the jit is more optimised fro qualcomm CPUs.
Aery said:
JIT is not working in JP1/2/3.
Click to expand...
Click to collapse
In fact, it's disabled in JP1 and JP2, but it is enabled in JP3.
Mikulec said:
Why do u think its not working?Look at the scores of Droid 2 which comes with FROYO. It has the same scores as SGS. Looks like the jit is more optimised fro qualcomm CPUs.
Click to expand...
Click to collapse
I was just thinking the same thing..
Mikulec said:
Why do u think its not working?Look at the scores of Droid 2 which comes with FROYO. It has the same scores as SGS. Looks like the jit is more optimised fro qualcomm CPUs.
Click to expand...
Click to collapse
I can understand it is more optimized for qualcomm's cpus but there is a 100% difference in the score.
Seems a bit odd to me.
Mikulec said:
Why do u think its not working?Look at the scores of Droid 2 which comes with FROYO. It has the same scores as SGS. Looks like the jit is more optimised fro qualcomm CPUs.
Click to expand...
Click to collapse
Sigh. Damn the N1.
More optimized for Qualcomm CPUs??? Isn't the CPU part of both the Samsung Hummingbird and the Snapdragon in essence the same ARM11 Cortex-A8? Yes they have some minor differences but please, nearly 3 times difference????? I'm beginning to think that Linpack is favoring Qualcomm chips!
And something else! How the hell did the Nexus One score 78 MFLOPS ?!?!?!?!?!?!?
Lets get back to the more important thing - what about the JP7, fake or real?
All I can say is that the snapdragon CPU and Hummingbird are built on the same architecture so they should perform quite close to each other.
Only thing that will prevent it from such a thing are drivers. But just wait I think Samsung will show the true power of the device with 2.2 soon.
they all based on coretex a8 more or less, if that what you say.
like Motorolas using TI OMAP 3 dont get high scores, so is galaxy s.
never get 30MFlops.
There is another thread somewhere that mentioned jp7 so yes i think its out in the wild somewhere.
I dont know , but maybe VFP (FPU hardware) functionality enabled in 2.2 snapdragon .
(limpack is FPO intensive).
For example: (in s3c6410 arm v6)
(mod enables VFP)
Average Linpack scores :
Stock: ~2.8 - 3.0
Stock w/ JIT: ~4.5
Stock w/ this mod: ~4.6
Stock w/ Jit & this mod: ~7.5 - 7.7
http://forum.sdx-developers.com/android-2-1-development/arm11-optimized-libdvm-so-3587/
edit: "(limpack is FPO (floating point operations) intensive)" instead "(limpack is FPU intensive)"
I'm hoping Samsung has put a performance-leach on the test versions because they want to blow peoples minds once they release it properly.
It's a bit of a stretch though...
pepitodequetequejas said:
(limpack is FPU intensive).
Click to expand...
Click to collapse
If true, this makes linpack a very very poor benchmark for phone speed... Nothing really uses the FPU on a phone besides for graphics, and that is all done by the graphics chipset and not the CPU anyway.
RyanZA said:
If true, this makes linpack a very very poor benchmark for phone speed... Nothing really uses the FPU on a phone besides for graphics, and that is all done by the graphics chipset and not the CPU anyway.
Click to expand...
Click to collapse
Yes , my slower (800Mhz armv6 wm) i8000 can easily beat my SGS in
FPO per second with chainfire moded libraries ( enables VFP ).
http://en.wikipedia.org/wiki/LINPACK
Pardon my language, but the benchmarks made me cry WTF is up with Samsung. but then Linpack may have a multiplier x2 if it sees Motorola as the brand name or something. LOL
After All why do we care about Linpack score?
OrionBG said:
After All why do we care about Linpack score?
Click to expand...
Click to collapse
Because the kind of geeks who read a forum like this our obsessed with benchmarks and anything they can point to to say they've got the best geek-toy on the planet currently!
MomijiTMO said:
Sigh. Damn the N1.
Click to expand...
Click to collapse
Yeah, damn the N1. LOL. Seriously, I have one, and it's such a love/hate relationship:
Perfect form factor/size/weight (well, wouldn't mind a 4" screen).
Great button layout.
Quality build.
Gets the latest greatest Google sh*t first.
Horrible touchscreen - almost impossible for me to type anything on it.
Horrible red bias to the screen/display.
Horrible pink dot in pictures.
Unusable in sunlight.
So close to the perfect phone, but not quite.
Same can be said for our SGS, though. Hardware pretty damned sweet. Screen accurate and beautiful and useable in sunlight, weight perfect, size perfect, works in sunlight. Of course, out of the box it's almost unusable because of the lag! At least the SGS can be fixed with software for the most part. The N1's flaws are hardware and permanent. I'm not even sure HTC attempted to fix any of them in later builds of the phone.

Is galaxy s Gpu really that power

i have heared that galaxy s Gpu can give 90M triangles/sec is that true as some sources claming that it only gives 28M tri/sec http://en.wikipedia.org/wiki/PowerVR , and the higher one sgx 545 gives 40 m so how the sgx 540 gives 90M
hoss_n2 said:
i have heared that galaxy s Gpu can give 90M triangles/sec is that true as some sources claming that it only gives 28M tri/sec http://en.wikipedia.org/wiki/PowerVR , and the higher one sgx 545 gives 40 m so how the sgx 540 gives 90M
Click to expand...
Click to collapse
I don't think the number listed on wikipedia is 'triangles' per second... It just says polys... So it could be a different shape thats harder to render?
Just my guess.
Besides if the 90M claimed is actually the 28 million then don't worry because the same thing for the iPhone's GPU (the 535) claims around 22m and wiki is listing it as 14.
Aaannnnddd if you're worried about the GPU feel comforted that no 3D benchmarks I've seen have even slowed it down so far and you can see tons of videos on youtube of Galaxy S series phones face rolling every single other Android device in gaming FPS benchmarks. Even if it isn't as amazing as the numbers they claimed there is no doubt that it's the best in the market at the moment, and by quite a lot too!
I'm not going to pretend that I read the comment thoroughly, but I've read a similar question. The person who seemed to know what they were talking about, said that essentially the 90m is a "theoretical number" and that about half of that number is what the phone should? can? will? potentially? do...(skimming, memory and probably comprehension make that a very difficult word to fill in accurately)....but this is how all manufacturers report their graphics capabilities (at least in smartphones, but I'll assume the same holds true for the desktop/laptop graphics cards).
So, while the number is definitely overstated, it's within the standard reporting convention...and relative to other numbers, still accurate (2x as many triangles is 2x as many whether everything gets cut in half of cut by a factor of 3).
*I'll remove my fuzzy language when someone better informed than me responds*
I also read a good article (don't know where it is now sorry) all about how the GPU relies heavily on the memory and bus between them etc and for example there could be a phone running the same GPU as another and have much less performance because they don't use much memory, or use slow memory. Apparently our SGS have done pretty well in all departments.
To untangle the confusion-
Triangles= "polys" (polygons)
The SGS does nothing bear 90M, but on the other side, none of the other phones are doing what the manufacturers are claiming them to do.
Plus, the wikipedia article is FAR from being reliable, it's been edited more than 5 times over the past 2 months, with ever changing results. No official specs are available from imgtec.
One thing i CAN tell you is that the GPU on the SGS is nothing less than a monster.
http://glbenchmark.com/result.jsp?benchmark=glpro11&certified_only=2
I'd like you to take as a refrence the Compal NAZ10 that uses the ever-glorified Nvidia TEGRA 2, and the iPhone 4 (SGX535)
I don't know what trick Samsung used, but there shouldn't be such a massive difference between the 535 and the 540.
Appearently someone over at Sammy did something right.
Extremely right.
Pika007 said:
...
One thing i CAN tell you is that the GPU on the SGS is nothing less than a monster.
http://glbenchmark.com/result.jsp?benchmark=glpro11&certified_only=2
I'd like you to take as a refrence the Compal NAZ10 that uses the ever-glorified Nvidia TEGRA 2, and the iPhone 4 (SGX535)
I don't know what trick Samsung used, but there shouldn't be such a massive difference between the 535 and the 540.
Appearently someone over at Sammy did something right.
Extremely right.
Click to expand...
Click to collapse
Well, one important fact is the pixelcount in the glbenchmark link you sent. iPhone4 and iPad share the same GPU. The difference in pixels is about 20%, and hence the difference between those two.
Let me make one ugly calculation to map SGS's score to iPhone4's. Pixelcount difference between i4 and SGS is a factor 0.625. That we would make the SGS score 1146 on the iPhone resolution. (or 1723 for i4 on 800*480 resolution). Offcourse there are more factors involved but this the best estimate i can make at the moment.
Difference turns out not te be that great after all.
I knew this argument was going to pop up soon enough, so i'll add one VERY important factor-
Score doesn't decrease proportionally to an increase in resolution.
For example, doubling the resolution won't give half the score. More like 70%~
Try running 3Dmark on your PC in different resolutions, you'll see some interesting results.
Personally, GLmark 1.1 for me is just a very crude example, for general demontstrations. It's not really close to be very accurate.
I'm waiting for GLmark 2.0 that should be a great tool to effectively compare the devices.
Who cares if the phone is powerful when there are no great games that take advantage of the power and when you have an OS that lags all the damn time despite the fact that Quadrant gives me 2100+. Even opening the PHONE app can take up to 10 seconds. This thing can drive me crazy at times.
Pika007 said:
To untangle the confusion-
Triangles= "polys" (polygons)
The SGS does nothing bear 90M, but on the other side, none of the other phones are doing what the manufacturers are claiming them to do.
Plus, the wikipedia article is FAR from being reliable, it's been edited more than 5 times over the past 2 months, with ever changing results. No official specs are available from imgtec.
One thing i CAN tell you is that the GPU on the SGS is nothing less than a monster.
http://glbenchmark.com/result.jsp?benchmark=glpro11&certified_only=2
I'd like you to take as a refrence the Compal NAZ10 that uses the ever-glorified Nvidia TEGRA 2, and the iPhone 4 (SGX535)
I don't know what trick Samsung used, but there shouldn't be such a massive difference between the 535 and the 540.
Appearently someone over at Sammy did something right.
Extremely right.
Click to expand...
Click to collapse
yes it is edited more than 5 times but there is an offcial sources says that sgx 454 gives only 40M polygons so hw sgx450 gives 90M i know numbers are not important if there is nothing to use it but i only wanted to know
I think its due to fact that older chip has 2d acceleration too, while 450 is pure 3d and we use cpu for 2d. Thats why its faster.
It is important to note that PowerVR does not do 3D rendering using the traditional 3D polygon based pipeline, like those used in nVidia and ATi cards. It uses the unique tile based rendering engine. This approach is more efficient and uses less memory bandwidth as well as RAW horse power. IIRC, the original PowerVR 3D PC card is a PCI card that can compete head to head with AGP based cards from 3dfx and ATi at that time. Unfortunately, its unique rendering engine does not fit well with Direct3D and OpenGL which favor traditional polygon-based rendering pipelines.
So, the 90M figure could well be the equivelent performance number when using traditional 3D rendering pipeline as compared to Tile-based PowerVR setup.
Power VR does indeed use the traditional 3D polygon based pipeline.
Tile based rendering is in addition, not instead.
Do note that not all games (and actually, far from it) are using TBR properly (if at all).
Read the release notes and press release, it has enough details.
hoss_n2 said:
yes it is edited more than 5 times but there is an offcial sources says that sgx 454 gives only 40M polygons so hw sgx450 gives 90M i know numbers are not important if there is nothing to use it but i only wanted to know
Click to expand...
Click to collapse
All the given numbers for "official" specs about PowerVR GPU's are for a frequenct of 200mhz.
Those chips can do well above 400mhz, so for example, if an SGX530 does 14M polygons and 500Mpixels per second @200mhz, if you clock it up to 400, it'll do 28Mpolys/1Gpixels.
Though i extremely doubt samsung has the SGX540 clocked at 600mhz in ths SGS...
A pratical and good exaple that shows of the power of the Galaxy S is Gameloft's Real Football 2010 game. The game hasn't got a framelock so it's playable on the Desire and Nexus One. Since pictures tell a thousand words and videos even moreso, I'll provide you this YouTube link: http://www.youtube.com/watch?v=S0DxP0sk5s0
Pika007 said:
All the given numbers for "official" specs about PowerVR GPU's are for a frequenct of 200mhz.
Those chips can do well above 400mhz, so for example, if an SGX530 does 14M polygons and 500Mpixels per second @200mhz, if you clock it up to 400, it'll do 28Mpolys/1Gpixels.
Though i extremely doubt samsung has the SGX540 clocked at 600mhz in ths SGS...
Click to expand...
Click to collapse
This is true however overclocking the GPU to those numbers is silly because the memory & memory bus can't support that much data throughput anyway. I don't even think there is enough to support the amount of the standard clock rate. There is a lot more to consider than just the GPU when it comes to graphics here
You're taking that article you read way too seriously.
Plus, we have no idea what is the bandwidth limit of the galaxy S, we don't know what kind of memory is used, how much of it, at what frequency, etc.
WiseDuck said:
Who cares if the phone is powerful when there are no great games that take advantage of the power and when you have an OS that lags all the damn time despite the fact that Quadrant gives me 2100+. Even opening the PHONE app can take up to 10 seconds. This thing can drive me crazy at times.
Click to expand...
Click to collapse
+1
Re: lag, I want doing bad until I installed one of the fixes. Now I've officially entered crazy-town.
If I would have to guess it has to do with S5PC110 optimizations. When rendering polygons there are many things that contribute aside from the GPU. Think of it maybe similar to hybrid-sli...(but this is just a guess)
but if you want to look at it in more detail, someone posted the official documentation and spec sheet for the S5PC110 a while back..I ddint get a chance to look at it but my guess the clock speeds and other stuff would be there :/
WiseDuck said:
Who cares if the phone is powerful when there are no great games that take advantage of the power and when you have an OS that lags all the damn time despite the fact that Quadrant gives me 2100+. Even opening the PHONE app can take up to 10 seconds. This thing can drive me crazy at times.
Click to expand...
Click to collapse
Well i dont have any lags, what so ever after lag fix. Something else must be troubleing your phone. Auto memory manager is a need tho if you want to keep it real snappy.
Sent from my GT-I9000 using XDA App

Hummingbird

http://www.gsmarena.com/samsung_i8700_omnia_7_joins_the_wp7_game_with_some_big_guns-news-1991.php
This may be a mistake in the article, but if it's true then it may mean that Samsung realized its' Hummingbird processor, although great looking on paper it's not as good as Snapdragon in everyday use (like the issue with 128bit registers allowing Snapdragons to gain much more performance boost on 2.2 than Hummingbird is able to).
Maybe. We can only speculate... But Its a different phone, different OS, different target audience, maybe snapdragon is better suited to that?
Correct me if im wrong.. But if i remember well.. Microsoft said few months ago that all windows phones will be based on same hw specs.. means all gonna have snapdragon!
There are higher throughput SIMD FP units in Qualcomm's CPUs, which mean they score higher in Linpack and the CPU segment of Quadrant, but it's a performance gain that doesn't translate to smoother day to day usage.
Samsung are using the older 65nm snapdragon processor. It's not better than their Hummingbird.
morvaeldd said:
http://www.gsmarena.com/samsung_i8700_omnia_7_joins_the_wp7_game_with_some_big_guns-news-1991.php
This may be a mistake in the article, but if it's true then it may mean that Samsung realized its' Hummingbird processor, although great looking on paper it's not as good as Snapdragon in everyday use (like the issue with 128bit registers allowing Snapdragons to gain much more performance boost on 2.2 than Hummingbird is able to).
Click to expand...
Click to collapse
Samsung aren't stupid, and I'm sure they are well aware of what their processor is capable of.. Remember, our JIT scores have been based entirely on pre-release versions of Froyo thus far.
And everyone is recycling the same information about the registers. That article was based on an assumption though.

Galaxy S2 Early Benchmark

http://images.anandtech.com/graphs/graph4177/35412.png
Some of you might know, the Galaxy S uses the PowerVR SGX-540 GPU, which will be in the upcoming Omap 4 SoC. And Samsung decided to switch from PowerVR to ARM for the GPU in the SoC, the Mali-400. The image you see is the early benchmarks of that new SoC, the Samsung Exynos SoC (formally Orion). My friend shared the image and right now he's not here for me to link the original site, but i'll edit that in later. Also as a note that mali chip is currently at 65nm in the benchmarks, the final chip will be in 40nm? But still, as someone who has been excited to see "5x the performance of the galaxy s" idc if it's an early build, I'm still
How about you guys?
information like this is suspect and rife with no base standard so it little or no meaning. The other phones are upgrades and most likely have tweaks of varying degrees. So this provides really nothing........ Beside the TMO Vibrant is at the top of most of the bench tests t to see the fascinate so much above...... mames one question the test results
kanwal236 said:
http://images.anandtech.com/graphs/graph4177/35412.png
Some of you might know, the Galaxy S uses the PowerVR SGX-540 GPU, which will be in the upcoming Omap 4 SoC. And Samsung decided to switch from PowerVR to ARM for the GPU in the SoC, the Mali-400. The image you see is the early benchmarks of that new SoC, the Samsung Exynos SoC (formally Orion). My friend shared the image and right now he's not here for me to link the original site, but i'll edit that in later. Also as a note that mali chip is currently at 65nm in the benchmarks, the final chip will be in 40nm? But still, as someone who has been excited to see "5x the performance of the galaxy s" idc if it's an early build, I'm still
How about you guys?
Click to expand...
Click to collapse
Wow the Tegra 2 doesn't beat the SGX-540 by much talking about making a powerful phone,wow the Galaxy S is a damn powerful phone,now if game companies tarted to actually use its powerful GPU,it would be great.
Extremely interesting, this.
Well, I hope it gets better as Samsung updates the software, drivers and other parts of the UI. (lolz)
Take a closer look at the benchmark, this is a graphics benchmark. The cpu itself has great performance, but what made the S so great was that it had the best gpu, no game lagged. Samsung is changing up the gpu provider and the new one benchmarks lower than the S, by final release at most they could hope for is getting the same performance in graphics. But the point is, samsung promised 5x the graphics performance, we're currently at half.
Sent from my SXY-T959
Are you surprised?

All exynos users...your reviews?

Whoever have the exynos version please post a review on all details.
StylishMafia said:
Whoever have the exynos version please post a review on all details.
Click to expand...
Click to collapse
+1
please!!!
Just got one yesterday and its really fast than my Note 2 here is my first antutu score on it
just got mine.loving the device. SM-N900 So far I can say that it is blazing fast. cam has no lag. Although my hand would hurt using this as my primary phone, the one handed operation features are quite brilliant. they've taken it to the next level. Screen is beautiful.Will play with it more
I wrote some impressions in a thread in the general section. Someone asked for benchmarks, and while google was resyncing apps, data, etc, I got a 17K quadrant score. I reran it a couple of times since then and my highest broke 22k.
Synthetic benchmarks aside, it absolutely flies. However, without an S800 to compare it to, I cannot say which is faster. They're both beasts.
On the other hand... it seems like they're generally pretty close, with the S800 edging out the exynos 5420 in most CPU-bound benchmarks. This makes me wonder - perhaps the exynos will be the better chip once the firmware update for heterogeneous big.little processing rolls out? Time will tell.
My money is on the Mali GPU, too - this is extremely unscientific, but the 604 was light years ahead of the Adreno 225 and better than the 320 in many cases, and the 6xx (628? I forget) sounds like such an enormous improvement. I'm sure the 330 is no slouch, but I think it could get edged out... again, time will tell.
Of course, none of this really matters. What really matters is the software, and I expect S800 will be EXTREMELY well supported. Every high-end (and even mid-range; did you see the new kindle fire?) has an S800, whereas the Exynos 5420 has almost no design wins and a tarnished reputation from the chip in the galaxy s4 (5410?).
If you're on the fence, I would say go with an S800. If you want to take a little gamble, get an Exynos - it won't be much worse than the S800, and in "Q4" (is that now? I don't know what samsung's fiscal quarter schedule is like) or later, I think there's a very real chance Exynos will come out on top.

Categories

Resources