Related
So as we all know with our devs pushing the cpu like up to these ridiculous speeds and our benchmarks not really showing the type of jump that we were really expecting, Is it possible that the way the CPU works is causing these scores not to score as high but not affect it's speed in the same way?
So after reading a bit on our CPU, from what I got from it ours does has about the average amount of IPS(instructions per second) compared to the snapdragon and it's custom architecture. However ours is said to be designed in a way that it requires 25-50% less instructions for 20~% of it's functions. So if thats the case we should be doing a little better in different areas.
In turn my real question is...
Because these benchmarks were built prior to our CPU being out, will that affect the numbers considering they probably aren't built for the way ours functions. Or am I just thinking way to far into this and have no idea what i'm talking about .
Which benchmarks are you referencing?
MFLOPS?
lqaddict said:
Which benchmarks are you referencing?
MFLOPS?
Click to expand...
Click to collapse
Anything from Linpack to SetCPU native benchmarks.
xplanowestx said:
Anything from Linpack to SetCPU native benchmarks.
Click to expand...
Click to collapse
You should probably read a bit on LINPACK - it was originally designed in the late 70's, 1979 to be precise, it was a set of Fortran procedures to perform numeric linear algebra. It is quite affected by the OS, and that's why you see LINPACK scores on 2.2+ systems are quite "improved" over 2.1 or earlier systems running better or the same hardware.
As far as SetCPU is concerned - there is no point running the benchmark on the CPU's that do not support NEON instructions, NEON is a Cortex-A specific (our phone and iPhone4 are among the few that have it); running it on a non-Cortex-A CPU, like a G2's GenII Snapdragon/Sciorpion from Qualcomm, does nothing but confuse people with bunch of numbers (the numbers are correct, they show how long it took that CPU to ignore the instruction instead of execute it, ignoring takes a lot less time than executing ).
And the C number in native SetCPU, I would assume it is a condition set on the instruction set to Carry on even if the previous instruction resulted in NOP (No Operation), so in the case of the glorified G2 for instance (you prolly saw the pretty screenshots) after ignoring the NEON instruction, it carries on ignoring the next NEON instruction throwing at it, so I am not sure what it is indicative of though, have to ask the SetCPU dev.
I hope this helps.
lqaddict said:
You should probably read a bit on LINPACK - it was originally designed in the late 70's, 1979 to be precise, it was a set of Fortran procedures to perform numeric linear algebra. It is quite affected by the OS, and that's why you see LINPACK scores on 2.2+ systems are quite "improved" over 2.1 or earlier systems running better or the same hardware.
As far as SetCPU is concerned - there is no point running the benchmark on the CPU's that do not support NEON instructions, NEON is a Cortex-A specific (our phone and iPhone4 are among the few that have it); running it on a non-Cortex-A CPU, like a G2's GenII Snapdragon/Sciorpion from Qualcomm, does nothing but confuse people with bunch of numbers (the numbers are correct, they show how long it took that CPU to ignore the instruction instead of execute it, ignoring takes a lot less time than executing ).
And the C number in native SetCPU, I would assume it is a condition set on the instruction set to Carry on even if the previous instruction resulted in NOP (No Operation), so in the case of the glorified G2 for instance (you prolly saw the pretty screenshots) after ignoring the NEON instruction, it carries on ignoring the next NEON instruction throwing at it, so I am not sure what it is indicative of though, have to ask the SetCPU dev.
I hope this helps.
Click to expand...
Click to collapse
It does help very much so which is sort of what i was seeing regarding SetCPU scores, they didn't seem correct based off the newer frequency's of the TeamWhiskey kernel they are developing. This would make sense.
So regarding linpack and 2.2, why is it that the Snapdragon is able to pull out these 55+ linpack results? It doesn't make sense that our CPU wouldn't be capable to pull off the same with what were are able to OC our CPU too, were getting much lower results compared to other phones moving to 2.2, if you can please explain this a little further. Thanks!
xplanowestx said:
It does help very much so which is sort of what i was seeing regarding SetCPU scores, they didn't seem correct based off the newer frequency's of the TeamWhiskey kernel they are developing. This would make sense.
So regarding linpack and 2.2, why is it that the Snapdragon is able to pull out these 55+ linpack results? It doesn't make sense that our CPU wouldn't be capable to pull off the same with what were are able to OC our CPU too, were getting much lower results compared to other phones moving to 2.2, if you can please explain this a little further. Thanks!
Click to expand...
Click to collapse
It's the OS limitation, and I am trying not to make an excuse for it, but RFS as well.
LINPACK score is highly affected on what your OS is doing, and how it is doing what it does , given a fair chance, with Galaxy S hopefully getting 2.2 this year (LOL), and team z4mod developing the overhaul of the cripple RFS you will see comparable LINPACK scores.
lqaddict said:
It's the OS limitation, and I am trying not to make an excuse for it, but RFS as well.
LINPACK score is highly affected on what your OS is doing, and how it is doing what it does , given a fair chance, with Galaxy S hopefully getting 2.2 this year (LOL), and team z4mod developing the overhaul of the cripple RFS you will see comparable LINPACK scores.
Click to expand...
Click to collapse
You made some realy great points, it is always good to have knowledgeable people around to answer technical questions. However, I don't understand how RFS, a file system affects arithmetic calculation? (from what I read, Linpack is basically a really difficult math problem).
It's the OS limitation, and I am trying not to make an excuse for it, but RFS as well.
LINPACK score is highly affected on what your OS is doing, and how it is doing what it does , given a fair chance, with Galaxy S hopefully getting 2.2 this year (LOL), and team z4mod developing the overhaul of the cripple RFS you will see comparable LINPACK scores.
Click to expand...
Click to collapse
Thanks for the very in depth answers. Hopefully fixing the filesystem would fix the speeds and such of the CPU. I'm hoping your right on that idea. If thats the case, do you think we will be comparable to the G2's Extremely high scores?
You made some realy great points, it is always good to have knowledgeable people around to answer technical questions. However, I don't understand how RFS, a file system affects arithmetic calculation? (from what I read, Linpack is basically a really difficult math problem).
Click to expand...
Click to collapse
+1
On a second note, what would compel Samsung to use such filesystem? Is it because of the proprietary side of it?
xplanowestx said:
Thanks for the very in depth answers. Hopefully fixing the filesystem would fix the speeds and such of the CPU. I'm hoping your right on that idea. If thats the case, do you think we will be comparable to the G2's Extremely high scores?
+1
On a second note, what would compel Samsung to use such filesystem? Is it because of the proprietary side of it?
Click to expand...
Click to collapse
I might've gone there on a limp, but from my understanding is that floating point operations stressed out by computing NxN matrix (LINPACK benchmark) is optimized to take an advantage of the L2 cache on the processor. Well our processor has 512K L2 cache if I am not mistaken, and from what I read on Qualcomm CPU in G2 - it has no cache at all. So where to keep the data? The next fastest media is memory and then disk I/O, and idsk I/O is where the weakest link is, given the fact that we have a decaying FAT32 implementation the weakest link is even weaker.
Remember your system is as good as your weakest link.
Did I make a plausible excuse ?
lqaddict said:
I might've gone there on a limp, but from my understanding is that floating point operations stressed out by computing NxN matrix (LINPACK benchmark) is optimized to take an advantage of the L2 cache on the processor. Well our processor has 512K L2 cache if I am not mistaken, and from what I read on Qualcomm CPU in G2 - it has no cache at all. So where to keep the data? The next fastest media is memory and then disk I/O, and idsk I/O is where the weakest link is, given the fact that we have a decaying FAT32 implementation the weakest link is even weaker.
Remember your system is as good as your weakest link.
Did I make a plausible excuse ?
Click to expand...
Click to collapse
It COULD be possible, considering how freeing up RAM increases linpack scores (to an extend). But how about people who have voodoo? Isn't that ext 4, so linpack on those system should fly off the chart but they don't/
PaiPiePia said:
It COULD be possible, considering how freeing up RAM increases linpack scores (to an extend). But how about people who have voodoo? Isn't that ext 4, so linpack on those system should fly off the chart but they don't/
Click to expand...
Click to collapse
Voodoo only addresses /data, I would assume instructions for LINPACK calculation will be punted to /cache
So then the new z4mod should (hopefully) raise our linpack scores fairly higher since thats a 100% complete conversion correct?
xplanowestx said:
So then the new z4mod should (hopefully) raise our linpack scores fairly higher since thats a 100% complete conversion correct?
Click to expand...
Click to collapse
I believe with mature z4mod and a stable froyo build the Galaxy S will shine, hopefully it is not a wishful thinking, and these things happen sooner than later for us.
lqaddict said:
I believe with mature z4mod and a stable froyo build the Galaxy S will shine, hopefully it is not a wishful thinking, and these things happen sooner than later for us.
Click to expand...
Click to collapse
+1 to that.
I am wondering what is the point of just pursuing higher Quadrant score?
Ok, you got a high quadrant score, does it mean your phone is running better?
Better has different aspects:
stability
speed
battery consumption
usability
so, come on guys, please stop pursuing higher score blindly!
I always see people posting how hight their quadrant score were, but not mentioning if their phone were working better with that high scores.
anguslaw said:
I am wondering what is the point of just pursuing higher Quadrant score?
Ok, you got a high quadrant score, does it mean your phone is running better?
Better has different aspects:
stability
speed
battery consumption
usability
so, come on guys, please stop pursuing higher score blindly!
I always see people posting how hight their quadrant score were, but not mentioning if their phone were working better with that high scores.
Click to expand...
Click to collapse
The point of going for a high quadrant score is speed & response, the higher quadrant is the higher the response and speed of the OS will be ,cause if you check in quadrant it checks CPU/RAM/3D/2D/I/O.
The 3D,2D and RAM doesnt change between OS's but CPU and I/O does,because I/O is the "Input / Output" that if it is higher,the OS will work alot smoother nevertheless because its clean and simple just how it was ment to be.
Also 2.1 and 2.2 does make a diference,i mean read about the diferences and they say the 2.2 can be between 100% to 500% faster,and if you check stock 2.1's 500 compared to stock 2.2's 1400,you get around 280% more speed.
Battery Consumption / Stability : It has nothing to do with Q's scores,thats more of the OS configuration and CPU Scaling.
Just try it out yourself : Use Normal SE 0.24 and do Q ,youll get around 500 or so,and you will notice the OS is very laggy even without a live wallpaper.
Now go for FreeX10 2.2 and do Q,youll get around 1400 and it will work awesomely fast even WITH a live wallpaper.
So in one brief description : The higher Q's score is,the better the device's Hardware interacts with its Software giving you smoother and faster usability of it,and stability + battery consumption is something totally diferent that if they dont work as they should,can be fixed by modifying the OS and its settings.
Edit : Thats why we are after high Q's scores rather than Linpack scores,Linpack only tests CPU speed by using the language defined in the OS (JIT etc)and even if it would be 120 would be useless if the I/O is rubbish.
So yeah,Q's score marks a diference.
Regards
Lawyered ?
How are other android touchpads doing? W / the 1.7ghz overclock I'm hitting a Max of 98ish mflops and 3300+ on quadrant! Which makes it my fastest Droid to date. Lets see if anybody can post a better quadrant screenshot?
richard head said:
How are other android touchpads doing? W / the 1.7ghz overclock I'm hitting a Max of 98ish mflops and 3300+ on quadrant! Which makes it my fastest Droid to date. Lets see if anybody can post a better quadrant screenshot?
Click to expand...
Click to collapse
i overclock my touchpad to 1.782 ghz. do you know how to overclock gpu
adreno 220 is in 266 mhz original but i need to overclock it to 300 mhz because its a very good frequency or 320 mhz
Sent from my SGH-T989 using xda premium
Gotcha
Think I beat you
Fastest I've gotten. Have not really tried with the latest versions.
general feeling is that as far as benchmarks go, quadrant is crap
should try antutu instead.
ive a thread in this section somewhere where some people have posted results
This discussion seems a bit flawed, in that you're gauging performance based upon benchmark software, etc. We've talked about this a lot over the past year, and most of these tests are very dependent upon specific hardware setups, usually in much older devices with older chipsets. My Nexus S w/ a single-core 1ghz UNDERCLOCKED to 880mhz scores a 3500 all the time in Quadrant. No way in hell it's better than the touchpad. It's not even a fair fight, if you compare raw specs.
Do you judge the performance of your computer based on how fast it can perform iterations of Prime 95? No, you judge it based on user experience.
Is your user experience with CM7 very fast? I know mine is. Based on that, and the experience I get with gfx-intense apps like GTA3, etc, I say the performance is very good.
PS. The fact that Quadrant to this day still has the Nexus One as its top reference device says something about it being old and outdated. The Nexus One came out in Jan 2010, 2 years ago.
Hello XDA,
I was just wondering if anyone else have encountered the same issue (not sure if it is an issue).
When I run Quadrant Benchmark, there is only 8 CPU showing. I have googled and have seen a few videos shows 12 CPU during a Quadrant Benchmark.
I have attached screenshots of the information Quadrant Benchmark shows.
Also Quadrant seems to only show 1 Core in System Information.
Not sure that the device being a refurb would have anything to do with it. But I just want to make sure that I am getting the most out of my phone, and may be also possibly increases by benchmark scores to the 4000-5000 range as even with just 8 CPU I am averaging between 3000-4000 benchmark points.
Any clarification and assistance would be greatly appreciated.
Thank you in advance!!!
I think you're quite confused . The "8 CPUs" are the number of tests it runs to test the CPU portion. Could be as simple as they previously used 12 tests but have now managed to test the same things with 8 tests. You phone doesn't have 8 CPUs
Quadrant often only displays 1 core because the other "sleeps" when not needed. Once the demand for more power is there it'll wake the second core.
If you want higher scores you can overclock, change governor, etc. but really benchmark scores are meaningless
Quadrant was updated several months ago and the number of CPU tests was reduced to 8 from 12. See this post (its a june 2010 post) http://www.aurorasoftworks.com/products
Wink
Thank you
Hello All, Thank you to the both of you for the clarification and guidance. Much appreciated!!!
Have a great day/night.
Close
Ok, How do I go about closing this thread? Edit Post maybe?
twiggums said:
I think you're quite confused . The "8 CPUs" are the number of tests it runs to test the CPU portion. Could be as simple as they previously used 12 tests but have now managed to test the same things with 8 tests. You phone doesn't have 8 CPUs
Quadrant often only displays 1 core because the other "sleeps" when not needed. Once the demand for more power is there it'll wake the second core.
If you want higher scores you can overclock, change governor, etc. but really benchmark scores are meaningless
Click to expand...
Click to collapse
Beyond [email protected] said:
Ok, How do I go about closing this thread? Edit Post maybe?
Click to expand...
Click to collapse
You have to wait for one of the mod's to close it. Maybe they'll close it, now that you've requested it. Or maybe it'll just stay open and get buried, which is fine too.
Pm a mod and request it to be closed.
Sent from my SAMSUNG-SGH-I777 using XDA
My understanding is that you need to edit the title to say [SOLVED] and it will either get buried and die or Red will close it.
I do not yet have a Nexus 7 to test this with. You use this at your own risk!
This is the first real kernel work I've done, so don't be surprised if it doesn't work. I've only provided a boot.img as fastboot is easy enough to use on the Nexus 7.
Features (If it works):
CPU OC to 1.7GHz maximum
CPU over volt to hopefully reach 1.7GHz
GPU OC to 600MHz
Simple IO scheduler
SmartAssV2 CPU governor
The boot.img is attached. Source can be found at my github.
If anyone here has a Nexus 7 it would be very helpful to know if it works. I should have mine soon though. If it works well, enjoy. Feedback is always welcomed, as are benchmarks. Thanks.
Removed link until fixed!
This is scary looking, an untested Overclock that's never been run on the hardware before.
I'm guessing that the T30L is just a speed binned T30 and as such this shouldn't damage it. The same overvolt (and higher) has been successful on the T30 to get even higher clocks (1.8GHz). I would test this given hardware, however I don't yet have my Nexus 7.
ben1066 said:
I'm guessing that the T30L is just a speed binned T30 and as such this shouldn't damage it. The same overvolt (and higher) has been successful on the T30 to get even higher clocks (1.8GHz). I would test this given hardware, however I don't yet have my Nexus 7.
Click to expand...
Click to collapse
This sounds interesting. Especially the 600mhz gpu OC but may I ask if you are thinking of implementing some kind of app interface to control gpu clock and voltages etc like Extweeks on google play please? As I am guessing a lot of people won't be able to go to 600mhz stably, so a way to change the OC to something like 520mhz (to bring it to T30 speed) would be a good option
I've seen voltage tweaks controlled from userspace on other devices but not the GPU clock. I'd like to get it working first, then I guess I'll look at such things, especially if there is interest.
Cel1084 said:
This is scary looking, an untested Overclock that's never been run on the hardware before.
Click to expand...
Click to collapse
you go first!
bencozzy said:
Glados kernel on the galaxy nexus allows gpu oc control.
Click to expand...
Click to collapse
Siyah kernel for the galaxy s2 and galaxy s3 both let you control gpu clock speed and voltages, might be cool to have something similar
I don't even have my Nexus yet, and i'm already downloading things to flash to it, hahah. Will report back once Google ships to the US!
if this kernel works can we control the clocks with antutu or similar?
The CPU clock should be controllable, and I'm working on making the overvolt controllable. The GPU clock is not yet controllable, and I'm not so sure where to start on that.
ben1066 said:
The CPU clock should be controllable, and I'm working on making the overvolt controllable. The GPU clock is not yet controllable, and I'm not so sure where to start on that.
Click to expand...
Click to collapse
you should put the gpu clock to 520mhz arnt t30l 413mhz stock and kai would be even slower assuming its the budget tegra3 soc
Right, here's the thing. I've spoofed the SoC speedo ID to be that of the standard T30, however, without looking through with a fine tooth comb, it seems that the top that that id goes is 600MHz. In usage, it may be 520MHz, but I'm not sure. In addition I'm fairly sure these are just speed binned, and can probably run at the higher clocks if we just add a bit more voltage, or they get a little hotter. If anyone can tell me if this actually works, then I can adjust either way.
Look at tegra3_dvfs.c, line 256-262. It seems to indicate a maximum of 600MHz.
This should be helpful to you. tegra3 technical reference manual. everything there is to know about all variants of the chip. how it works, what its capable of, schematics, diagrams, chip layout, etc,,
http://db.tt/vWWou2Fu
Thanks but I already have access to NVidia's Tegra portal, which includes the TRM for Tegra 2 and Tegra 3. I'm hoping I shouldn't have to mess with it that low level
I don't understand why anyone would want to overclock a Tegra3, which is plenty fast enough already, especially when they have never even touched the device.
Also, I don't understand why anyone with any sense would use Simple IO scheduler, which has a higher latency and lower throughput than deadline, or even the bloat that is CFQ for that matter.
And finally, I don't understand why any real 'developer' would release something like this without testing it, especially with possibly dangerous overclocking and overvoltage settings. Only on XDA...
With all due respect, you should remove it until you have tested it *yourself* and confirmed that it doesn't make your Nexus 7 vanish in a cloud of smoke.
When I feel the need the need for speed owww.
_thalamus said:
I don't understand why anyone would want to overclock a Tegra3, which is plenty fast enough already, especially when they have never even touched the device.
Also, I don't understand why anyone with any sense would use Simple IO scheduler, which has a higher latency and lower throughput than deadline, or even the bloat that is CFQ for that matter.
And finally, I don't understand why any real 'developer' would release something like this without testing it, especially with possibly dangerous overclocking and overvoltage settings. Only on XDA...
With all due respect, you should remove it until you have tested it *yourself* and confirmed that it doesn't make your Nexus 7 vanish in a cloud of smoke.
Click to expand...
Click to collapse
Our Tegra 3 CPU is a lower clock version that the normal T30, it's the T30L. I have no doubt that this will not damage your device, the voltages used are still less than used by some TF201 ROMs (the TF201 uses the T30). I included Simple IO scheduler since it is something that seems popular, latency isn't the only thing that matters (read http://www.vincentkong.com/wiki/-/w...42041E#section-Android+IO+Schedulers-Deadline). I have seen benchmarks that show both SIO and deadline as better than each other, it depends what metric you record. I didn't remove CFQ, it's not that I've added it. The scheduler can be changed if you so desire anyway.
I have not provided a simple flash package and I've clearly stated in red writing that this is UNTESTED. I do not have the device, and it is yes untested however I didn't see the point on keeping something potentially useful private. If you have the knowledge to use fastboot to flash a boot.img, you probably know how to flash back the old one too.
_thalamus said:
I don't understand why anyone would want to overclock a Tegra3, which is plenty fast enough already, especially when they have never even touched the device.
Also, I don't understand why anyone with any sense would use Simple IO scheduler, which has a higher latency and lower throughput than deadline, or even the bloat that is CFQ for that matter.
And finally, I don't understand why any real 'developer' would release something like this without testing it, especially with possibly dangerous overclocking and overvoltage settings. Only on XDA...
With all due respect, you should remove it until you have tested it *yourself* and confirmed that it doesn't make your Nexus 7 vanish in a cloud of smoke.
Click to expand...
Click to collapse
seriously harsh man, just because you don't understand doesn't mean its wrong, or right for that matter
ben1066 said:
Our Tegra 3 CPU is a lower clock version that the normal T30, it's the T30L.
Click to expand...
Click to collapse
I take it you understand why similar chips are rated at various speeds for different devices? Because they are designed with a lower thermal output and / or the cooling characteristics / power characteristics of the device are different. The T30L has lower speed apps processors, lower speed GPU and lower speed memory. All in all, it will pump out much less heat than a T30.
I have no doubt that this will not damage your device, the voltages used are still less than used by some TF201 ROMs (the TF201 uses the T30).
Click to expand...
Click to collapse
You don't *know* that it won't damage someones device, you are assuming that it won't. The likelihood is that it probably won't, but would you stake your life on it? I wouldn't, and I've been doing Android kernel development for some time.
Also, this isn't the TF201, and it isn't the T30. It is a different device with different thermal characteristics and a different SoC, you can't compare them like that.
I included Simple IO scheduler since it is something that seems popular, latency isn't the only thing that matters
Click to expand...
Click to collapse
Latency of reads and writes and throughput are the only 2 things which matter (and I mentioned both), and SIO is poor at both of them. Justin Bieber is popular, but he's still ****, so including something which is popular isn't really a good reason.
---------- Post added at 06:44 PM ---------- Previous post was at 06:33 PM ----------
foxorroxors said:
seriously harsh man, just because you don't understand doesn't mean its wrong, or right for that matter
Click to expand...
Click to collapse
Harsh perhaps, but I prefer honest. Necessary, most certainly.
It is stupid and irresponsible to release something which is untested and potentially dangerous as it isn't fair on the poor muppet that flashes it and then f**ks their device up.
It has only been released because some 'developer' wants to make his epenis bigger by releasing something for a brand new device on XDA. Not that I am saying that he is the only one, there's plenty of others that do it, but as I have one of these on order I am taking an interest in these threads and was quite surprised with what I saw.
As someone who has done kernel development for some time now, I would never dream of releasing something I haven't tested thoroughly myself, or which I have got a trusted tester to thoroughly test, but hey, this is XDA and the standards are low.
ben1066 just out of curiosity may I ask how the gpu scales frequencies on the Tegra 3 t30l please? As I am used to the galaxy s2 and s3 where you have numerous frequency steps like 166mhz, 260mhz, 350mhz and 440mhz and you have an up and down threshold to govern whether you jump up or down the available frequencies, is this similar to how the gpu in works on the tegra 3 please?
Also when you say overclock the gpu, is it replacing 416mhz with 600mhz or is it adding an extra gpu frequency step after 416mhz, so 416mhz is still available to be used if needed? Sorry one last question, if the gpu does have frequency steps like other gpus, what ones are available for use please?
I am sorry to ask, I am just so curious about these questions, and I can't find them anywhere on the internet, so any help would be greatly appreciated. Thank you so much