Message boards :
Number crunching :
Future of Milkyway@Home
Message board moderation
Author | Message |
---|---|
![]() ![]() Send message Joined: 13 Dec 12 Posts: 101 Credit: 1,782,758,310 RAC: 15 ![]() ![]() ![]() |
I'm just curious if anyone else has the same thoughts that I do around this project in the years to come. And by that I'm referring to newer GPU's. For the first time in decades, I will have not 1 but 2 NVIDIA gpu's and as we all know..... they SUCK at Milkyway. I'll be losing an old stalwart in the old 280X which is probably capable of doing a few thousand less credits than a newer NVIDIA gpu I'm gussin. Is it possible that Milkyway@Home could slow down in future if the best GPU's become project specific? Radeon VII's, Titans. I find that NVIDIA cards finally do well on Einstein so for me personally, I gain on Primegrid & Einstein. The point is, when I upgrade my cards as they get older, it doesn't look like the newer cards are doing so well with M@H due to the FP64 gimping. Thoughts. ps I see the 7900XTX is still gimped 3.838 TFLOPS (1:16) vs a Radeon VII 3.360 TFLOPS (1:4) For reference, my 3070Ti - FP64 (double) performance 339.8 GFLOPS (1:64) (urgh) ![]() ![]() |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
I'm just curious if anyone else has the same thoughts that I do around this project in the years to come. And by that I'm referring to newer GPU's. MOST people who crunch at Boinc Projects never read the forums at all, so if they have a problem they assume it's the Project and move on to a different one that hopefully works for them, that indicates the lack of willingness to tweak their pc's which leads to NO I don't think the AMD cards will ever not be here simply because they are a pain in the butt to make work in most Linux distros. Nvidia gpu's are MUCH easier to setup in Linux but in Windows they are the same as in Linux, download the drivers, install them, reboot the pc and you are gaming or crunching. AMD gpu's can also pull their own weight and be cost effective too for gaming then if you add in the new Intel gpu's that are coming that also 'just work' in Windows pc's and MilkyWay and othrr Projects that benefit from the AMD cards have a good long run ahead of them. YES AMD is dropping the FP32/FP64 differences between them and Nvidia and it's no wonder Nvidia is doing better at some Projects right now. It turns out gaming doesn't benefit as much from the FP32/FP64 ratios as AMD thought. |
![]() Send message Joined: 18 Nov 22 Posts: 84 Credit: 640,530,847 RAC: 9 ![]() ![]() |
Nvidia Titan Vs will probably stay the best affordable high performing card. Maybe some V100s if you can get a deal on the infrastructure for an SXM2 setup. The A100s will remain priced out of reach for most people for a while I think. AMD CDNA cards could be a viable option, but it seems like they gatekeep the drivers. ![]() ![]() |
![]() ![]() Send message Joined: 13 Dec 12 Posts: 101 Credit: 1,782,758,310 RAC: 15 ![]() ![]() ![]() |
Yes but that's kinda my point. Sure.... I'd love to buy an old (old because the Titan days are gone) Titan because it's smashes Milkyway, but it gets left behind in other projects and newer cards. Unless I'm specifically targeting Milkyway@Home, I'm no longer buying GPU's that have great FP64 performance. As these great FP64 cards become old, there are no new FP64 cards really coming through at the consumer level. That's why I wonder if this project will slowly produce less results. If I had infinite money $$ sure, I'd load up on all the Titan V's I could find and plant them with pc's and smash out Milkyway until I reached my desired rank. Then I'd rip them all out, put them into storage & plug in shiney, new high end GPU's to smash other gpu projects. That's what I'd do. (I'd also be retired and have time to do all that sh^te) ![]() ![]() |
![]() Send message Joined: 18 Nov 22 Posts: 84 Credit: 640,530,847 RAC: 9 ![]() ![]() |
Titan V still does really well at Einstein too. but what's the alternative? for the project to reduce their precision so the normie cards can play? ![]() ![]() |
![]() ![]() Send message Joined: 13 Dec 12 Posts: 101 Credit: 1,782,758,310 RAC: 15 ![]() ![]() ![]() |
Yes.... what is the alternative? I assume that nothing will change but over time people will upgrade their GPU's but that will result in less work being completed for this project. I'm hoping my Radeon VII's kick on for a while longer but any replacement card, won't be as good for M@H. ![]() ![]() |
Send message Joined: 28 May 22 Posts: 17 Credit: 402,111,833 RAC: 0 ![]() ![]() ![]() |
Yes.... what is the alternative? There are a couple possible alternatives; I believe the AMD MI50's and MI60's are comparable to VII. The MI60 has 32GB of HBM2 memory (and a 1/2 FP64 divider, which gives it a 7+ TB FP64 rating, much faster than the VII's 1/4 FP64 divider). MI50 has 16GB of HBM2 memory just as the VII does. Both are rated at 300 Watts, same as VII. Info from the TechPowerUp and other various web pages. I have my VII undervolted <200 Watts and it still performs fine. Martin |
![]() Send message Joined: 19 Jul 10 Posts: 618 Credit: 19,254,980 RAC: 50 ![]() ![]() |
That's why I wonder if this project will slowly produce less results. Not really, the SP:DP ratio is maybe lower than it was before on some selected cards (4:1 IIRC only Cayman, Tahiti and Radeon VII), but on average the DP performance is still a lot higher than it used to be and CPUs are getting faster too. I don't see there any slowdown. Not sure what cards were released at the end of 2019, but that's when the last significant increase of monthly credit happened, you can clearly see it in the graph, quite stable since than with small after-christmas-bumps. ![]() |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
That's why I wonder if this project will slowly produce less results. On top of that with all the new more than 16/24/32, without HT, core cpu's for pc's the trajectory should be nothing but upwards as their price comes down to affordable levels. |
![]() ![]() Send message Joined: 13 Dec 12 Posts: 101 Credit: 1,782,758,310 RAC: 15 ![]() ![]() ![]() |
Does anybody crunch Milkyway on their CPU? Not me personally. Credits are too low. It will be interesting to see how the credits pan out. I'd say the handful of big hitters skew these figures a bit. I just know that from my point of view, future GPU's I buy will not perform as well at Milkyway@Home due to the DP gimping of consumer cards. I was just curious what others though. ![]() ![]() |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
Does anybody crunch Milkyway on their CPU? I personally have several pc's crunching MilkyWay cpu tasks, I crunch for wuprop which awards badges for the number of hours you put in not the credits a project gives you, so longer tasks are not always a bad thing. The pc's doing the cpu work are ones that really struggle to meet deadlines at other projects, they meet them but not always by a few days which is what I prefer. They are older dual quad core Intel Zeon cpu's that just crunch 24/7 15 tasks and a gpu task at a time but only at 2.?ghz. I also use it as a back-up project because they respect the zero resource share, ie only send me the tasks I need not hundreds as I wait for my main project to fill the cache again. |
![]() Send message Joined: 18 Nov 22 Posts: 84 Credit: 640,530,847 RAC: 9 ![]() ![]() |
depends on your personal goals and if you make milkyway a priority or not. I don't think the project has anything to worry about. Plenty of people will happily build systems focused on MW and continue to buy the high performing FP64 cards, either older models like TitanV or Radeon VII, or spend more for even better performing cards like newer Nvidia Tesla and AMD Instinct cards. ![]() ![]() |
![]() Send message Joined: 3 Dec 22 Posts: 1 Credit: 1,311,662 RAC: 0 ![]() ![]() |
Hi,it's me who only crunch M@H on CPU. I have three devices, and they do not have independent graphics cards. One is a laptop and the two are desktop. |
![]() ![]() Send message Joined: 13 Dec 12 Posts: 101 Credit: 1,782,758,310 RAC: 15 ![]() ![]() ![]() |
Agreed it depends on those who wish to target Milkyway@Home over other projects that don't require DP. Not sure about people (ie - the masses) buying Tesla's & Instinct cards. I guess time will tell. I can't dedicate my NVIDIA cards to M@H though. The credits are just too low and those cards for me personally are better used elsewhere. Great discussion. Thank you ![]() ![]() |
![]() Send message Joined: 19 Jul 10 Posts: 618 Credit: 19,254,980 RAC: 50 ![]() ![]() |
Does anybody crunch Milkyway on their CPU? I did in the past, I joined Milkyway before I had any GPU, which was able to crunch anything at all and I don't mind doing it in the future, at least if I don't find a better place for my CPUs like for example right now WCG's MCM. I prefer to use my CPUs for work that can't be done by GPUs, but in fact there are not that many projects out there doing "real science" according to my requirements, so it's not very unlikely, that I might run Milkyway on CPU again. Milkyway actually has n-Body, which can't be done on GPUs, but they still don't offer the possibility to limit CPU-tasks to n-Body only while doing Separation on GPU and a mix of the single core Separation and MT n-Body is quite a PITA. ![]() |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
Agreed it depends on those who wish to target Milkyway@Home over other projects that don't require DP. I have a question for you...you say you want to crunch for projects that give alot of credit for their tasks and you DO have a bit over 21 Billion credits already so my question is what are you going to do with them? As you can see from my own signature I'm working on my second Trillion credits but I STILL can't buy a cup of coffee with them at Starbucks or a Dunkin Donut store!! YES credits are nice but it can't be the sole reason for crunching someplace. Seti did research on credits one time and came up with a few things that projects must consider when giving out credits for their tasks, they concluded that giving out more credits increases the likelihood of more users coming to your project which also brings more cheaters so your anti-cheating systems have to be robust or alot of the work done is suspect. Another thing all those users coming to your projects is that the network can be overwhelmed, just ask World Community Grid about that right now, and can have a negative effect on your project for a long time. And giving out more credits than other projects for your tasks increases the need for a more robust supply of tasks to meet the new demand. In short giving out more credits is a two edged sword for projects whole like MiklyWay operate with an ever changing group of college students at the helm. ![]() |
Send message Joined: 15 Nov 21 Posts: 4 Credit: 5,135,128 RAC: 1,057 ![]() ![]() |
I have a HPC that I use for work, gaming and other leisure activities. I often have BOINC running in the background as I have excess CPU/GPU power that I'm more than happy to use for crunching scientific aspirations. I also have an Asus Tinkerboard (SOC Raspberry Pi device) and a server that I use as a home lab. Again I crunch with these with the excess CPU cycles. I've also installed BOINC on my mum's computer as she doesn't need four cores for playing solitaire and reading emails. My point is that I crunch with excess cycles. If I was trying to make money I'd connect with Gridcoin or just mine Monero. I do this more as a hobby or offhand task, I don't go out of my way to rack up an energy bill. This is fun and while it's gamified and has competitions it doesn't make sense to crunch as if it's your job and the world depends on it. If you are only in it for the points there are plenty of other projects that you can focus on. I prefer space based BOINC projects. Usually these BOINC projects are under funded and require volunteers to maintain software code and hardware, not just crunching WU. If they had money they sure wouldn't be using BOINC. Look at SkynetProject or Microbiome Immunity Project. Both got large government grants and access to super computers so they shut down their use of BOINC. You are more than welcome to assist with optimisation of their code so that it will work better with new nVidia cards. BOINC projects always always want assistance from professionals https://github.com/Milkyway-at-home Wow that was long and I could have kept going... |
![]() Send message Joined: 18 Nov 22 Posts: 84 Credit: 640,530,847 RAC: 9 ![]() ![]() |
You are more than welcome to assist with optimisation of their code so that it will work better with new nVidia cards. BOINC projects always always want assistance from professionals the applications work fairly well with nvidia cards already. unless the project wants to reduce their precision (ie stop using FP64 almost exclusively in their code and switching to FP32) then there isnt much to be done. you could eek out another 20-30% with some nvidia specific optimization on existing cards. but not much more can be done if the project remains on double precision. I have to assume they have a good reason for it, and they weren't happy with the precision using only FP32. the problem with newer nvidia GPUs isn't the application, it's the limits in FP64 performance put on the consumer level (and most professional level) GPUs, either through software or physical hardware nerfing of the FP64 capabilities. the more recent nvidia product stack only has good FP64 at the very top of their enterprise product stack (A100, H100, etc). AMD in recent years is starting to do the same. they are still nerfing FP64 on their consumer cards, to a lesser extent than nvidia but still enough that older cards are more worthwhile to use for MW. and again, Line nvidia, you only get truly great FP64 from AMD these days at the top of their enterprise/AI product stack (Instinct line, MI100, MI210, etc) ![]() ![]() |
Send message Joined: 28 May 17 Posts: 76 Credit: 4,398,269,182 RAC: 1,016 ![]() ![]() |
That's why I wonder if this project will slowly produce less results. ![]() Seems that 2020 was the uptick year. 2008 - 2009 seen growth 2010 seen a major growth 2011 - 2016 seen decline 2017 - 2019 seen good growth 2020 has seen the best growth of the project 2021 - 2022 being the 2nd and 3rd best years of the project might have seen a slight decline due to server issues. Keep in mind, the 280X came out in 2013 which was for the longest time one of the best performing gaming cards for Milkyway, yet we still seen a decline from the year before that and the following 3 years seen a decline as well. The Titan V, which was released in 2017 was the first growth of the project since the 2008 - 2010 years. The Radeon VII was launched in the beginning of 2019 and only saw minimal growth from 2018. I think the only thing that will hold MW back would be the server hardware for now (unless the upgrades have taken place already) |
Send message Joined: 28 May 17 Posts: 76 Credit: 4,398,269,182 RAC: 1,016 ![]() ![]() |
https://www.teamanandtech.org/images/boinc/MW_GRAPH.jpg Seems the SSL cert is broken. Click link to see the chart until I can get the cert fixed. |
©2024 Astroinformatics Group