Message boards :
Number crunching :
AMD vs. Intel
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · Next
Author | Message |
---|---|
The Gas Giant Send message Joined: 22 Nov 01 Posts: 1904 Credit: 2,646,654 RAC: 0 |
Who?. ....and I look forward to reading every word who? writes! Keep it up. Reading about how Barcelona is in reality is an amazing let down (even at this early stage in its release). |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
Who?. ... and I would be just as uninterested if an AMD employee were bad-mouthing the Intel parts. While we're on the subject, where is the magical hyper-optimized SETI app. that Who? has been promising? |
archae86 Send message Joined: 31 Aug 99 Posts: 909 Credit: 1,582,816 RAC: 0 |
While we're on the subject, where is the magical hyper-optimized SETI app. that Who? has been promising? Assuming you were actually interested in an answer to this question, rather than just criticizing, his most recent comment is: message 645690 |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
While we're on the subject, where is the magical hyper-optimized SETI app. that Who? has been promising? Ah, just a few more weeks of tuning. |
kittyman Send message Joined: 9 Jul 00 Posts: 51469 Credit: 1,018,363,574 RAC: 1,004 |
While we're on the subject, where is the magical hyper-optimized SETI app. that Who? has been promising? LOL....I like the guy, but I hope he had better luck with that gearbox.... "Freedom is just Chaos, with better lighting." Alan Dean Foster |
tullio Send message Joined: 9 Apr 04 Posts: 8797 Credit: 2,930,782 RAC: 1 |
They say generals like to fight past wars. Folding@home has passed the petaflops limit due to the CELL processor on Playstation3. Since I am still using an Intel PII Deschutes running SETI, Einstein and QMC I am a little surprised of these Intel vs AMD wars. What about a good RISC chip? Tullio |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
They say generals like to fight past wars. Folding@home has passed the petaflops limit due to the CELL processor on Playstation3. Since I am still using an Intel PII Deschutes running SETI, Einstein and QMC I am a little surprised of these Intel vs AMD wars. What about a good RISC chip? i am french! don t shoot ! hahhaha who? Skulltrail D5400XS |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
So, i can tell you what it is, they will try to beat the 3dmark score with a phenom proto and 2 overclock GPU. as soon as they do, i ll do the same with my skulltrail and we will see :) i am ready , hehehehe i thought it is funny to share. who? Skulltrail D5400XS |
Fivestar Crashtest Send message Joined: 10 Dec 99 Posts: 226 Credit: 5,377,978 RAC: 0 |
The fight is here. |
Sirius B Send message Joined: 26 Dec 00 Posts: 24881 Credit: 3,081,182 RAC: 7 |
Since it's release, several people have attached their barcelona systems to the project. Whenever new equipment is released, there are always those who want it straight away. If common sense is applied with patience, once things have settled down, the cost drops. Once the cost from competitors is in the same ballpark, it is down to the individual's choice. From what I have seen in this & other threads, it seems to me that some people are building systems just to run s@h. The majority of people build/purchase computers to aid in education/everyday life, i.e., Internet, accounts/word processing etc. Does it really matter whether they are Intel/AMD, Windows/Linux/OS X? Most comsumers are cost conscious. As for myself, just because my systems are AMD & attached does not mean I don't build Intel systems. I build what the customer wants, & unfortunately, I could not get the Intel customers to have boinc on their systems. Why are mine AMD? Simple, I got a great deal on the equipment & could not get similar deals with Intel based kit. My network cost £1,400 whereas had I chosen Intel at the time, it would have cost £1,870. That difference enabled me to build another system for a customer & help reduce the cost of the network. To me that is sound business. When things have settled down, if barcelona proves too expensive, then my server will be Intel, if not, then AMD. |
Sirius B Send message Joined: 26 Dec 00 Posts: 24881 Credit: 3,081,182 RAC: 7 |
[Edit] Question answered [/edit] |
Osiris30 Send message Joined: 19 Aug 07 Posts: 264 Credit: 41,917,631 RAC: 0 |
Hi Everyone. seems to be a problem - is it my end or s@h end? The thread is being editted and a bunch of who? related material have been put in another thread.... |
Sirius B Send message Joined: 26 Dec 00 Posts: 24881 Credit: 3,081,182 RAC: 7 |
Hi Everyone. seems to be a problem - is it my end or s@h end? Thanks, Alinator pipped you to the post. Regards PJ |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
While we're on the subject, where is the magical hyper-optimized SETI app. that Who? has been promising? well, my gearbox is still not in, i am too busy getting my product ready as soon as possible. my personal life and hobbies are far behing for more than a year, people should be able to understand this. In the mean time, look at the amazing hardware I am leading ... people need to understand that between SETI opti and my product design and tuning, the choice is obvious. V8 is already dominating BOINC, and SkullT will increase this lead, the optimization for me is the cherryl on the cake, not the cake. You all understood now? I am not Intel, and you don t have to blame me for the entire company stuff. now, if you feel like releasing all your complain on me, well, I can't help you, you are an iD... Without me, there is NO vectorization in the SETI code, because every body was trying to optimized the FFT, while the critical path was findPulse(), so, stop decreasing my contribution here, and show a little respect too, it is not because I have a blue badge that I do not deserve respect for my past contribution, you could be all looking for a faster FFT ... clear too? thanks, have a good day. who? |
Alex Kan Send message Joined: 4 Dec 03 Posts: 127 Credit: 29,269 RAC: 0 |
Without me, there is NO vectorization in the SETI code, because every body was trying to optimized the FFT, while the critical path was findPulse(), so, stop decreasing my contribution here, and show a little respect too, it is not because I have a blue badge that I do not deserve respect for my past contribution, you could be all looking for a faster FFT ... clear too? You're taking far too much credit for the past year's advances in optimized apps if you make a claim like this. Joe Segur pointed this out last time you made the claim, so I guess it's my turn. If you're referring to your post in September where you vectorized one of the array-averaging loops, which I'm assuming you are, Ben Herndon responded to it, pointing out that he had something similar long ago, to the point where some of it was already checked into the SETI CVS server at the time of your post. He also tried to open a dialogue with you about your optimization proposals, but I don't see a response from you there or on Simon's forums. If you look in the source code to apps that predate your post, I imagine you'll find that your post is definitely not the first time that that loop has been vectorized. I don't think any of us have spent time on optimizing the FFT beyond picking the fastest libraries we could find and choosing between in-place and out-of-place transforms where advantageous. Our efforts have mostly been directed elsewhere. |
1mp0£173 Send message Joined: 3 Apr 99 Posts: 8423 Credit: 356,897 RAC: 0 |
... and with you, there is one machine running vectorized code? |
Trinitron Send message Joined: 11 Jan 00 Posts: 16 Credit: 305,768 RAC: 0 |
Share that love for my P3 too, please. Yes, no? ;) |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
Without me, there is NO vectorization in the SETI code, because every body was trying to optimized the FFT, while the critical path was findPulse(), so, stop decreasing my contribution here, and show a little respect too, it is not because I have a blue badge that I do not deserve respect for my past contribution, you could be all looking for a faster FFT ... clear too? Well, if you were aware of findpulse() being the critical path, it is very strange that you left it the way it was ... come on! it was bunch of float to int and int to float conversion. I don't buy this. I still have a copy of it, you want to me show what you was supposely focusing on optimizing? May be some of you was aware of it, but none of you ever writen a code like I did, and your today's code , except the array of functions, look like my code in many of the sub-routings, like it or not. anybody can compare your today's code, with my 1st posting with the intrinsics. who? |
Francois Piednoel Send message Joined: 14 Jun 00 Posts: 898 Credit: 5,969,361 RAC: 0 |
Without me, there is NO vectorization in the SETI code, because every body was trying to optimized the FFT, while the critical path was findPulse(), so, stop decreasing my contribution here, and show a little respect too, it is not because I have a blue badge that I do not deserve respect for my past contribution, you could be all looking for a faster FFT ... clear too? And I don't think you try to fall back your optimization in windows build either. Do i need to give a copy of the thread you posted on a mac web site about my vectorized code? (You is the seti MAC group) who? |
Alex Kan Send message Joined: 4 Dec 03 Posts: 127 Credit: 29,269 RAC: 0 |
Well, if you were aware of findpulse() being the critical path, it is very strange that you left it the way it was ... come on! it was bunch of float to int and int to float conversion. I don't buy this. I still have a copy of it, you want to me show what you was supposely focusing on optimizing? This is true. Anyone can compare today's code with your first posting. They can also compare code from August 2006 with your first posting. Those of you who actually download the tarball, go to line 1175 in client/pulsefind.cpp. This is only the first Intel source release (hand-vectorized primarily because I was compiling with GCC at the time); IIRC, gradually-evolving PPC variants of the same functions go as far back as the source tarballs I've made available. Going back even further, you can also compare with Ben Herndon's variants of the same functions, committed to a Sourceforge CVS server nearly three years ago. The intrinsics are obfuscated somewhat by the use of preprocessor defines, but the structure of those loops is the same. Yes, I goofed in that this was an unofficial CVS tree rather than the official project tree, but the point still stands. Looking back at the discussions that have taken place on Simon's forums, some of which are not publicly-accessible, I can find a few examples of cases where find_pulse and its child functions were identified as being potential hotspots. More interesting is the observation in the second thread that ICC's autovectorizer is easily capable of vectorizing those array-summing loops. True, without the proper #pragma alignment hints, ICC won't pick up that the vector load for the low elements could be done with movaps. Still, in conjunction with the fact that pulse-finding wasn't nearly as big a hotspot in the pre-Enhanced days, this goes a long way towards explaining why optimizers were spending more time on less easily-autovectorizable functions prior to your post. And I don't think you try to fall back your optimization in windows build either. Do i need to give a copy of the thread you posted on a mac web site about my vectorized code? (You is the seti MAC group) What exactly are you asking here? I'm assuming that you're referring to the Team MacNN forums, since that's the only other place I post. I looked through the threads I've posted in, and I see two mentions of your work (only one of which I made), both concerning the structure-of-structure FFT optimizations you mentioned in August 2006. Am I missing something? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.