Topic: Why X11 runs cooler - an answer. (Read 2654 times)
I may be going further with X11 now, to make it run REAL hot - we'll see. SIMD, the largest bottleneck, has yet to be done properly, so that's up first.
whats up with the furry wallpaper man
It's funny how that's still news.
it was new to me... I went wtf?
Nice analysis, the thought I'd like to add is that even if you had the most possibly efficient x11 implementation and it still runs cooler than, say, scrypt, then you should still prefer scrypt over x-whatever. If the most optimized implementation is cooler, you are wasting power in keeping all the system working compared to scrypt, so you get a worst "network security / watt spent" ratio.
I think it would be hotter than scrypt - although I haven't gone that far yet.
whats up with the furry wallpaper man
It's funny how that's still news.
Is it the same for FRESH algo?
No idea, I'd have to look at it.
Oh ok. It runs cooler and acts a lot like x-11.
Oh god yes, it's affected. Uses the PHM code, and consists of Shavite, SIMD, and Echo, which are probably the three with the worst implementations.
Okay, so I took 15min to do a rough hack job on Freshcoin's kernel - just for shits and giggles. It's not nearly as good as it should be, because it relies heavily on SIMD, and my SIMD sucks - one big reason my X11 isn't up to par yet - but anyway, results were:
Code:
Stock:
290X: 5.65MH/s
280X: 4.59MH/s
Modded:
290X: 7.91MH/s
280X: 6.66MH/s
Difference:
290X: 40.00%
280X: 45.09%
Settings:
290X @ 1000/1250 with xI 64, worksize 64, and 2 GPU threads
280X @ 1140/1500 with xI 128, worksize 64, and 2 GPU threads
Are you saying X11 is better than other coin protocol?
See my comment here:
wolf, I don't get what you are trying to say. Can you put in layman terms please?
X11 runs cold because the implementation sucks and a good part of the GPU is idle.
wolf, I don't get what you are trying to say. Can you put in layman terms please?
X11 runs cold because the implementation sucks and a good part of the GPU is idle.
Hehe I've read in many forums x11 runs cooler cause it sucks at hashing. I'm surprised nobody has bothered to fix this until now...
Or perhaps it was such a large job to do properly it required some long term planning.
Do me a favour mate, "basically the same - the extra wave in flight helps hide memory access times, causing it to be faster", can you explain why the extra wave in flight helps hide memory access times? Why is hiding memory access times a good thing for speed?
Because the GPU schedules other waves to execute while the memory access is going on. So you basically lose no hashing time. If you have less waves in flight, you can end up stalling on a memory access.
OK so why is this so? Because the GPU schedules other waves to execute while the memory access is going on.
Why would it bother scheduling waves if it can operate more efficiently doing things a different way?
What more efficient way? You get to execute x amount of waves on a compute unit, determined by how much of the CU's resources the wave uses. This is why high register use is bad. If the CU doesn't have enough resources to execute concurrent waves, then it basically HAS to do nothing while waiting for memory - it can't do other work because the resources of the CU are being used.
wolf, I don't get what you are trying to say. Can you put in layman terms please?
X11 runs cold because the implementation sucks and a good part of the GPU is idle.
Hehe I've read in many forums x11 runs cooler cause it sucks at hashing. I'm surprised nobody has bothered to fix this until now...
Or perhaps it was such a large job to do properly it required some long term planning.
Do me a favour mate, "basically the same - the extra wave in flight helps hide memory access times, causing it to be faster", can you explain why the extra wave in flight helps hide memory access times? Why is hiding memory access times a good thing for speed?
Because the GPU schedules other waves to execute while the memory access is going on. So you basically lose no hashing time. If you have less waves in flight, you can end up stalling on a memory access.
Is it the same for FRESH algo?
No idea, I'd have to look at it.
Oh ok. It runs cooler and acts a lot like x-11.
Oh god yes, it's affected. Uses the PHM code, and consists of Shavite, SIMD, and Echo, which are probably the three with the worst implementations.
Very nice write up. I've always wanted a good explanation. Thanks.
nicely explained Wolf0, thanks.
whats up with the furry wallpaper man
It's funny how that's still news.
it was new to me... I went wtf?
Nice analysis, the thought I'd like to add is that even if you had the most possibly efficient x11 implementation and it still runs cooler than, say, scrypt, then you should still prefer scrypt over x-whatever. If the most optimized implementation is cooler, you are wasting power in keeping all the system working compared to scrypt, so you get a worst "network security / watt spent" ratio.
whats up with the furry wallpaper man
Nice write up OP. Btw Spoetnik, you're wrong most of the time.
I am alwaysrightwrong, come to terms with it 
I am always
i don't open my mouth with out the ability to prove it 99% of the time and you all make vague snotty little insults at me.. as usual
and being an obnoxious asshole has no bearing on whether i am right or not sweety.. go back to flogging shit-coins and Trolling prick.
sad tons of you out there are mentally challenged idiots that could never master the concept of staying on topic on a forum.
click topic.. see Spoetnik say something.. insult him and then leave.
and i am the loser ?
mmmmm hmmmmmm
I can't even describe with words how far you are out of your league Come-In-Behind
..scurry off or i will give you the infamous and legendary Spoetnik Treatment !
Your only reply should be thank you, for me giving you my opinion (The correct one)
The words of Spoetnik are a gift to the Altcoin scene and you all should feel privileged you even get to read them !
..your welcome now be quiet i am the smart one here !
@UnicronFarts
I think they are pretty cheap so prob not a big risk eh (picking up some JPC coins) LOL
I have seen crappier coins go the moon hahahah
Are you saying X11 is better than other coin protocol?
decent guide !
wolf, I don't get what you are trying to say. Can you put in layman terms please?
X11 runs cold because the implementation sucks and a good part of the GPU is idle.
Hehe I've read in many forums x11 runs cooler cause it sucks at hashing. I'm surprised nobody has bothered to fix this until now...
Or perhaps it was such a large job to do properly it required some long term planning.
Do me a favour mate, "basically the same - the extra wave in flight helps hide memory access times, causing it to be faster", can you explain why the extra wave in flight helps hide memory access times? Why is hiding memory access times a good thing for speed?
Because the GPU schedules other waves to execute while the memory access is going on. So you basically lose no hashing time. If you have less waves in flight, you can end up stalling on a memory access.
OK so why is this so? Because the GPU schedules other waves to execute while the memory access is going on.
Why would it bother scheduling waves if it can operate more efficiently doing things a different way?
wolf, I don't get what you are trying to say. Can you put in layman terms please?
X11 runs cold because the implementation sucks and a good part of the GPU is idle.
Hehe I've read in many forums x11 runs cooler cause it sucks at hashing. I'm surprised nobody has bothered to fix this until now...
Or perhaps it was such a large job to do properly it required some long term planning.
Do me a favour mate, "basically the same - the extra wave in flight helps hide memory access times, causing it to be faster", can you explain why the extra wave in flight helps hide memory access times? Why is hiding memory access times a good thing for speed?
Is it the same for FRESH algo?
No idea, I'd have to look at it.
Oh ok. It runs cooler and acts a lot like x-11.
Is it the same for FRESH algo?
Jump to: