Message boards :
Number crunching :
Confused..about what GPU support is coming 1st
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 . . . 6 · Next
Author | Message |
---|---|
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
Because some bright spark reckoned that ATI was faster than nVidia, and had a go at writing an optimized app for MilkWay while the MilkyWay admin were still thinking about CUDA, and some of us dashed out and got ATI cards and proved the bright spark right - the ATIs are ..... fast! Travis is still working on CUDA and an ATI version for MW_GPU. It’s just that the bright spark and his helpers got there first with the ATI rocket science. |
Send message Joined: 8 Mar 08 Posts: 17 Credit: 4,411,459 RAC: 0 |
Well said! :P |
Send message Joined: 21 Feb 09 Posts: 180 Credit: 27,806,824 RAC: 0 |
But why then is a ATI app out already, when Cuda is the one that is supported by BOINC? There are two halves of Milkyway: MW, and MW_GPU. MW_GPU is a different URL, and a different project in BOINC terms. The ATI app that is out is for this half of MW. Not the other half of MW, titled MW_GPU. It's the other half that will get the CUDA app - this half wont get a CUDA app yet - the focus is for CUDA to MW_GPU first. The ATI app isn't written by the project, but a very helpful third party well versed in ATi code, whereas the CUDA app is being written by the project developers. I know that..I'm not stupid. I was under the assumption that if you didn't understand the parity between MW and MW_GPU then you may not have known that info. It wasn't to call your intellegence into question, but often on the boards we get similar questions where people do not know the whole story (or haven't searched the forums for it), so I was just covering all the bases. Apologies, I didn't mean any offence by any of what I said. |
Send message Joined: 8 Dec 07 Posts: 60 Credit: 67,028,931 RAC: 0 |
The ATI app only processes CPU WUs on an ATI GPU. Once the Cuda code is released and MW GPU, which is a separate project, begins sending work an ATI app will be made available to process Cuda WUs on an ATI GPU. Afterward, the ATI app that processes CPU WUs will no longer be allowed. |
Send message Joined: 27 Dec 07 Posts: 2 Credit: 115,550,606 RAC: 0 |
Maybe Mr N00BS, before posting, you can learn somethings like : hello, please and thanks !!!!! Nvidia is bad !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! |
Send message Joined: 4 Jun 09 Posts: 45 Credit: 447,355 RAC: 0 |
Ah ic. But why would someone create an app to do CPU work on a GPU when they know the server can't handle the requests? That doesn't make any sense to me.. |
Send message Joined: 29 Sep 08 Posts: 1618 Credit: 46,511,893 RAC: 0 |
|
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
Try because they did not know that the server could not scale up to the demand. Sadly, there also have been changes to the clients so that there are more requests coming from the clients at the same time we ran into this ... the idea is that one or two more straws on the back of the camel cannot possibly hurt ... until you add one straw too many ... That is more or less what happened here ... high and rising demand, more "empty" requests and pop goes the weasel ... |
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
And don't forget the number of users has increased and the number of people using GPUs has grown at the same time. So all up, a few concrete blocks have been added to the camels back. |
Send message Joined: 4 Jun 09 Posts: 45 Credit: 447,355 RAC: 0 |
Wouldn't it of made sense then to ask the project lead, if the server could handle the load? I would think its pretty much common sense though..crunching WU's at what 8 seconds? I mean come on.. I think the people that made the app and the project lead just doesn't care if there are any WU's to go around..if they did they would block the ATI app. Sorry if I'm coming out as a little bit harsh, but this doesn't exactly seem "professional" to me. If you are going to be using peoples hardware for free, those people (us) should expect some professionalism. |
Send message Joined: 12 Apr 08 Posts: 621 Credit: 161,934,067 RAC: 0 |
Wouldn't it of made sense then to ask the project lead, if the server could handle the load? I would think its pretty much common sense though..crunching WU's at what 8 seconds? I mean come on.. The problem with scalability issues in computers is that it is the proverbial camel's back scenario... one microsecond you are fine, the next you are dead ... Having dealt with these my self when I was a systems engineer I know how insidious they can be and one of the reasons that I have been historically so hard on the development team for their lack of consideration for the accumulation of all the little things that kill scaleability ... I could list more including the issues that SaH is having with their database woes because of design issues in the table structures I warned about ... As to professionalism ... well, nothing much in the BOINC world is handled professionally ... why should this be different? As to blocking and Tasks to go around? All the project cares about, or should care about, is getting the work done ... who does it and how matters not at all from the goals of the project... though it does say nice things about the staff here that they do try to keep the rioting hordes placated ... As to harshness, well, you are asking for what no project promises ... they are doing science on a shoestring and a prayer and allow us to join in the processes ... if you want something more ... well, you are going to be disappointed ... |
Send message Joined: 4 Jun 09 Posts: 45 Credit: 447,355 RAC: 0 |
So to sum it up..they don't care; as long as the WU's are being crunched..Even though their server can't handle the load, lol. |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
So to sum it up..they don't care; as long as the WU's are being crunched..Even though their server can't handle the load, lol. The admin here are building a new project, MilkWay_GPU, to take the GPU load away from this one. The have said again and again that this was a better way to resolve the situation than continually trying to tweak this project, thus giving MilkyWay_GPU their attention, and leaving this project, MilkyWay, 'as is' until MilkyWay_GPU is ready. So to sum it up.. not as you said. |
Send message Joined: 4 Jun 09 Posts: 45 Credit: 447,355 RAC: 0 |
Yes I know this..its http://milkyway.cs.rpi.edu/milkyway_gpu/ My comment was about what is CURRENTLY going on. |
Send message Joined: 21 Feb 09 Posts: 180 Credit: 27,806,824 RAC: 0 |
Yes I know this..its http://milkyway.cs.rpi.edu/milkyway_gpu/ They are building the CUDA app for MW_GPU. It's almost done, apart from checkpointing. When Travis has finished this, according to one of the other threads, he will work on getting some WUs out over at MW_GPU. The ATi app, developed by Cluster Physik (a third party, he did it to help the science be computer faster) is being developed alongside MW - he's in constant conversation about the new app. The ATi app was made because the source code went public and CP wanted to help. Unfortunately this is what made the server get so many requests, after a WU which did take 10-30 minutes, now only takes 30 seconds. This action caused MW_GPU to be spawned. If it hadn't, we'd still be several months back doing CPU calculations. In essence, the MW_GPU side is being born out of necessity, and will (hopefully) make everyone happy :) Given the MW_GPU latest update (i.e. February and 'Almost There'), and that Travis has almost finished the CUDA app, I'd give it another month or so until MW_GPU is up and running for work - maybe not fully operational at that point, but in the right direction (from own personal experience I make this estimate). |
Send message Joined: 29 Sep 08 Posts: 1618 Credit: 46,511,893 RAC: 0 |
Yes I know this..its http://milkyway.cs.rpi.edu/milkyway_gpu/ Ahaa.. ;-) What is currently going on is this; You were mad for signing up to a project that doesnt work yet. And now its this one youre not happy with. The sun always shine on TV. You may go and watch some. ;-) aye.. i know, im not enough fed up with these projects. Its a shame. |
Send message Joined: 13 Feb 08 Posts: 1124 Credit: 46,740 RAC: 0 |
Sorry if I'm coming out as a little bit harsh, but this doesn't exactly seem "professional" to me. If you are going to be using peoples hardware for free, those people (us) should expect some professionalism. Well, you weren't asked to volunteer, youre welcome to come back in a few months and see if its more to your liking then. |
Send message Joined: 4 Jun 09 Posts: 45 Credit: 447,355 RAC: 0 |
Yes I know this..its http://milkyway.cs.rpi.edu/milkyway_gpu/ Thanks for replying. I guess I'll sit by and wait and check back every so often. |
Send message Joined: 6 Apr 08 Posts: 2018 Credit: 100,142,856 RAC: 0 |
|
Send message Joined: 24 Dec 07 Posts: 1947 Credit: 240,884,648 RAC: 0 |
youre welcome to come back in a few months and see if its more to your liking then. Normality to occur..... |
©2024 Astroinformatics Group