I have this one.

I don't understand, that says "168 (W) x 219.3 (D) x 218.3(H) mm" - how do you fit anything in there? How does even the motherboard get in? A motherboard is 300x300mm minimum. And every GPU I've owned is 300mm long. Perhaps I'm misunderstanding - is that a box for your Raspberry PI?

That's an ITX case. Those things are seriously small.
There's an ITX motherboard in it, a short GPU, a normal PSU, and it can take up to three hard drives / SSDs.
And did you really read the specs up to the point Ryzen processors are mentioned?
Just like in Ryzen 7 5700X?

---

- - - - - - - - - -
Greetings, Jens

That's an ITX case. Those things are seriously small.
There's an ITX motherboard in it, a short GPU, a normal PSU, and it can take up to three hard drives / SSDs.

So a tiny motherboard without many connectors, and some kind of specialist GPU which is either slow or expensive. That's a machine for a small office to do word processing on, not for Boinc.

And did you really read the specs up to the point Ryzen processors are mentioned?
Just like in Ryzen 7 5700X?

Well that gave me another laugh. "Supports AMD AM4 Socket Ryzen™ 2000, 3000, 4000 G-Series, 5000 and 5000 G-Series Desktop Processors, up to 65W"
SIXTY FIVE WATTS?! That's not a processor! My Ryzen 9 3900Xt uses 105W.

There is nothing wrong for using a weaker processor to contribute to any projects. I don't see the point of making fun of other people's setups. However, if you want to make fun of that, please fact check first. Your impression with ITX and the power of low-TDP CPU might be a bit out of date.

So a tiny motherboard without many connectors, and some kind of specialist GPU which is either slow or expensive. That's a machine for a small office to do word processing on, not for Boinc.

Running CPDN doesn't need any connectors for external I/O. A lot of ITX cases since 5-10 years ago can use normal full-sized dual-slot GPUs easily, not to mention CPDN doesn't use GPUs, so a powerful GPU may not even be a concern. That's a perfectly valid build for BOINC, so long as thermal is handled properly.

I built ITX before and it's manageable, especially for a 65W CPU. It was quite a lot of fun picking just the right parts and tune thermal for that tiny amount of room. My case was a 10L Node 202 and I had GTX1080 in it along with i7 6700 running 24/7 just fine. I did switch away eventually because I find myself upgrading too often and want multiple GPUs. However, that doesn't mean it wouldn't be perfect for others, especially the folks don't need GPU performance.

Well that gave me another laugh. "Supports AMD AM4 Socket Ryzen™ 2000, 3000, 4000 G-Series, 5000 and 5000 G-Series Desktop Processors, up to 65W"
SIXTY FIVE WATTS?! That's not a processor! My Ryzen 9 3900Xt uses 105W.

65W can do a lot and 5700X has better per-core performance than 3900XT. Unless you have 128GB of memory to saturate all your threads for the current model, 3900XT would likely lose to 5700X for CPDN when running same number of tasks. Even for full out multi-thread performance where 3900XT still has an edge, 5700X is better in power efficiency.

Good that you are commenting now. In a few weeks when non-X 7000 series are available, those 65W TDP CPUs will not only destroy 3900XT, but also destroy the entire 5000 series based on the 65W testing of 7950X compared to 5950X. That's partially because of the newer process node, but there are also fundamental reasons due to how silicon scaling works. For most voltage ranges, frequency scales linearly with voltage and dynamic power scales with square of voltage. For the same processor, to push for that last 10% of performance through higher frequency, it's going to suffer a cubic growth of power, burning 30%+ more. That's just how the physics works. 65W is not far from a sweat spot for efficiency for the size of consumer CPUs.

There is nothing wrong for using a weaker processor to contribute to any projects. I don't see the point of making fun of other people's setups.

I have some old 4 core CPUs too, I'm not making fun, but I'd never actually buy something like that, if I was given it I'd use it. But I'd put it in a bigger case, or take the side off and connect GPUs via USB risers.

However, if you want to make fun of that, please fact check first. Your impression with ITX and the power of low-TDP CPU might be a bit out of date.

AMD wouldn't be making 105W TDP CPUs if they didn't go faster.

Running CPDN doesn't need any connectors for external I/O.

I assume everyone runs more than just one project. GPU powered projects get a lot more calculations done. The CPU projects just get whatever power is left over. I've got one machine with FOUR graphics cards in it.

A lot of ITX cases since 5-10 years ago can use normal full-sized dual-slot GPUs easily,

Not that one, as I said it's 20cm and the GPUs are 30cm.

I built ITX before and it's manageable, especially for a 65W CPU. It was quite a lot of fun picking just the right parts and tune thermal for that tiny amount of room.

The only advantage is space, unless you want to run it in the back of your car, I see no point.

65W can do a lot

And 105W can do more.

and 5700X has better per-core performance than 3900XT. Unless you have 128GB of memory to saturate all your threads for the current model, 3900XT would likely lose to 5700X for CPDN when running same number of tasks. Even for full out multi-thread performance where 3900XT still has an edge, 5700X is better in power efficiency.

I buy whatever RAM I need. My two Ryzen 9 3900XT machines will take 128GB, I've only got 64GB in each just now, but it's not enough for LHC, so I'll be getting more. And if it's low on RAM, I simply put smaller tasks from another project on the other cores. CPDN is not the only project in existance.

Good that you are commenting now. In a few weeks when non-X 7000 series are available, those 65W TDP CPUs will not only destroy 3900XT, but also destroy the entire 5000 series based on the 65W testing of 7950X compared to 5950X. That's partially because of the newer process node, but there are also fundamental reasons due to how silicon scaling works. For most voltage ranges, frequency scales linearly with voltage and dynamic power scales with square of voltage. For the same processor, to push for that last 10% of performance through higher frequency, it's going to suffer a cubic growth of power, burning 30%+ more. That's just how the physics works. 65W is not far from a sweat spot for efficiency for the size of consumer CPUs.

I buy what gives the most computing power per £. The last two times I've looked (a year apart), the Ryzen 9 3900XT was the best. I include the cost of RAM, CPU, motherboard, etc. So more power from the same board and RAM means it's worth it.

I assume everyone runs more than just one project. GPU powered projects get a lot more calculations done. The CPU projects just get whatever power is left over. I've got one machine with FOUR graphics cards in it.

That's not necessary a correct assumption though. Definitely don't assume everyone cares about any BOINC projects even if they run multiple. For example, I have not found a single earth science project using GPU and I don't care about any of those math projects. I am not going to waste my power looking for some random numbers, even though that could be really meaningful for others. At least for now, there are astronomy and astrophysics projects I care a bit about, but I really don't know what I am going to do with the GPU if they run out of work.

I buy what gives the most computing power per £. The last two times I've looked (a year apart), the Ryzen 9 3900XT was the best. I include the cost of RAM, CPU, motherboard, etc. So more power from the same board and RAM means it's worth it.

I assume you run projects 24/7 like most of us. If you also include the power cost, assuming it's not free, you might suddenly see why power efficiency matters for performance per unit cost. The concept is known as TCO (total cost of ownership). For example, unfortunately for me, every 100W 24/7 translates to almost $300 per year now and that easily justifies upgrades more aggressively. 3900XT is a worse deal for me than a down-volted 5950X for example. You will likely get more bang for the buck if you down-volt your 3900XT once you factor in power. On the other hand, the best performance per cost is actually the server CPUs and boards because many of them are designed for compute density. Though you also need to carefully pick parts because those boards are even bigger and the cooling can end up being a jet engine if one is not careful.

Anyway, whatever works best for you that's great. Perhaps your power is free and I wish I could have that too. Just realize that others are probably also picking whatever works best for them.

That's not necessary a correct assumption though. Definitely don't assume everyone cares about any BOINC projects even if they run multiple. For example, I have not found a single earth science project using GPU and I don't care about any of those math projects. I am not going to waste my power looking for some random numbers, even though that could be really meaningful for others. At least for now, there are astronomy and astrophysics projects I care a bit about, but I really don't know what I am going to do with the GPU if they run out of work.

I do Milkyway and Einstein on GPU, astrophysics is interesting (ok I'm biased, I have a degree in it). I agree the maths stuff seems pointless. But presumably you're interested in curing cancer etc? If so use Folding at Home. It's not Boinc, but just as easy to set up. Tell it to use GPU only, and leave the CPUs for Boinc here.

I assume you run projects 24/7 like most of us. If you also include the power cost, assuming it's not free, you might suddenly see why power efficiency matters for performance per unit cost. The concept is known as TCO (total cost of ownership). For example, unfortunately for me, every 100W 24/7 translates to almost $300 per year now and that easily justifies upgrades more aggressively. 3900XT is a worse deal for me than a down-volted 5950X for example. You will likely get more bang for the buck if you down-volt your 3900XT once you factor in power. On the other hand, the best performance per cost is actually the server CPUs and boards because many of them are designed for compute density. Though you also need to carefully pick parts because those boards are even bigger and the cooling can end up being a jet engine if one is not careful.

I picked up 4 old xeons for virtually nothing on Ebay. 2 huge motherboards, no cases, just sat on a bookshelf. I cable tied some old 110 volt boxer fans (6 inch steel blades, don't put your finger in one!) above the boards as they expect a draught of air across them. There are chips on the board with no heatsinks that get too hot to touch! I'm in the UK so have 240V, but two 110V fans in series works ok.

Anyway, whatever works best for you that's great. Perhaps your power is free and I wish I could have that too. Just realize that others are probably also picking whatever works best for them.

Power also makes heat, so in winter it's basically free. I breed tropical parrots, so they appreciate the heat. You can collect gridcoins to offset the power usage a bit. Or plug it into your neighbour's garage :-) No I don't do that, he's elderly and doesn't need things like that happening!

If the government had the sense to build billions of wind farms, we could be getting power at 5 cents a kWh.

But presumably you're interested in curing cancer etc? If so use Folding at Home. It's not Boinc, but just as easy to set up. Tell it to use GPU only, and leave the CPUs for Boinc here.

I used to run FAH and WCG has sporadic GPU WUs too. However, these days I generally stay away from them mostly because medical research is a highly profitable field that don't really need volunteer contribution IMO. Even for results coming out of university research, they are too often realized in a way not truly beneficial for public health. I'd rather donate my compute power to things that seemingly hard to get investment but still looks interesting or important to me. That's purely a personal preference, just like the whole what project one wants to run falls into personal preference. I won't make assumption for others.

Power also makes heat, so in winter it's basically free.

Even though my place is not cold enough to require heaters in winter, this I totally agree. If one is going to turn on heaters anyway, they certainly should turn on all computers first and efficiency doesn't matter in that case.

I used to run FAH

I stopped because I can find enough things to do in Boinc. It's easier to have everything in Boinc, all controlled by Boinctasks, easier to divide up the processing power. I have a lot of machines to control.

and WCG has sporadic GPU WUs too.

Those seem to have disappeared unfortunately, they only have CPU cancer tasks just now, so I'm running those.

However, these days I generally stay away from them mostly because medical research is a highly profitable field that don't really need volunteer contribution IMO.

I don't look at it that way. The more research done the better. If I can make more research get done then I'll do it. I don't care if it costs a different person money.

Actually, why should medical research be profitable? That means when you get ill you have to pay for the medicine. Do research on Boinc for the free stuff, like Rosetta.

And why is earth science not profitable? With all the governments making a big deal over climate change, you'd think there would be loads of money in it.

Since I refuse to use Linux, my contribution here is on hold until they issue virtualbox tasks. Although I've just received a retread from someone who took 3 months to do 2.5 days of CPU time then crashed it.

Even though my place is not cold enough to require heaters in winter, this I totally agree.

My parrots are happiest at 25C.

If one is going to turn on heaters anyway, they certainly should turn on all computers first and efficiency doesn't matter in that case.

You can use a heat pump to get 300% efficiency, but those things cost a fortune to buy and they wear out. I've decided to just use mine for air conditioning (since it busted it's valve after only 2 years so it can only run one way or the other until I swap the air hoses over). I used to have gas heating, but the pipes and radiators are a hassle and expense.

I run the Ryzen 7 5700X coupled with a GTX 1660.
I wanted an ITX built that I can tweak for low power consumption while having as many CPU threads in the system as possible.
The Asrock will only take CPUs up to 65 Watts, so the 5700X is the most powerful CPU I can plug into it.
The Palit GeForce GTX 1660 StormX is a relatively short (less than 20cm) GPU that needs 120 Watts at most and can nicely configured to use a lot less.
While the system is tweaked for low power consumption it can easily cope with both my personal needs and crunching or different Boinc projects.

---

- - - - - - - - - -
Greetings, Jens

The Palit GeForce GTX 1660 StormX is a relatively short (less than 20cm) GPU that needs 120 Watts at most and can nicely configured to use a lot less.

All GPUs have that don't they? I can move the power limit from -50% to +50%. I have mine on +50% :-) Even if you can't do that, they all let you drop the clock to half.

The Palit GeForce GTX 1660 StormX is a relatively short (less than 20cm) GPU that needs 120 Watts at most and can nicely configured to use a lot less.

All GPUs have that don't they? I can move the power limit from -50% to +50%. I have mine on +50% :-) Even if you can't do that, they all let you drop the clock to half.

Most should.
I actually have a 1050Ti (Palit StormX as well) that runs at exactly 75 Watts which it gets out of the slot.
If you wanted to make it faster you'd have to undervolt it, to get more amperes flowing through it while running cooler than standard.
I also had an MSI card with the same specs that you could overclock in the same way.
But that's fairly unusual with nowaday's cards, I think.

---

- - - - - - - - - -
Greetings, Jens

Most should.
I actually have a 1050Ti (Palit StormX as well) that runs at exactly 75 Watts which it gets out of the slot.
If you wanted to make it faster you'd have to undervolt it, to get more amperes flowing through it while running cooler than standard.
I also had an MSI card with the same specs that you could overclock in the same way.
But that's fairly unusual with nowaday's cards, I think.

I think undervolting it would prevent it from running fast. People who overclock cards raise the voltage. You need more volts to charge individual bits in the processor faster. Does it have a PCI-Express power connector on the top you could add extra current to? Looking at photos of that model on the internet, I see they don't. Perhaps you could add one - although I never bother overclocking, all you get is a shorter lifespan and crashes and mistakes. The only thing I do is increase the power limit from normal to +50%. That doesn't seem to bother them as much as increasing the clock speed. In fact when they get old and flaky, I often lower the clock speed to keep them going. I did melt the power connector quite badly on an R9 Nano at +50%. The wires are now directly soldered on!

Most should.
I actually have a 1050Ti (Palit StormX as well) that runs at exactly 75 Watts which it gets out of the slot.
If you wanted to make it faster you'd have to undervolt it, to get more amperes flowing through it while running cooler than standard.
I also had an MSI card with the same specs that you could overclock in the same way.
But that's fairly unusual with nowaday's cards, I think.

I think undervolting it would prevent it from running fast. People who overclock cards raise the voltage. You need more volts to charge individual bits in the processor faster. Does it have a PCI-Express power connector on the top you could add extra current to? Looking at photos of that model on the internet, I see they don't. Perhaps you could add one - although I never bother overclocking, all you get is a shorter lifespan and crashes and mistakes. The only thing I do is increase the power limit from normal to +50%. That doesn't seem to bother them as much as increasing the clock speed. In fact when they get old and flaky, I often lower the clock speed to keep them going. I did melt the power connector quite badly on an R9 Nano at +50%. The wires are now directly soldered on!

The fun thing is they don't have that connector, and they will get exactly 75 Watts out of the slot. No more, no less. So if you lower Volt, mathematics say Ampere has to go up because Watts are fixed.
Those cards aren't intended to be clockable in any way, but with this trick it can be run faster and cooler (less voltage, less temperature).

By increaing the power limit I trashed my 980Ti during a GPUGrid challenge. Since then I usually limit power to something at the lower end of what the card takes.

---

- - - - - - - - - -
Greetings, Jens

The fun thing is they don't have that connector, and they will get exactly 75 Watts out of the slot. No more, no less. So if you lower Volt, mathematics say Ampere has to go up because Watts are fixed.

Actually no. When you undervolt, you're changing the voltage on the chip itself, which is about 1 volt. The input to it from the motherboard will remain at 12 volts. You're just changing the voltage regulator setting on the card itself.

By increaing the power limit I trashed my 980Ti during a GPUGrid challenge. Since then I usually limit power to something at the lower end of what the card takes.

Not had a problem here, but most of my cards are old crap anyway, I wouldn't know if it was me that broke them. My good card I play games on is not that old. But I've run it 24/7 at 150% power for a year without breaking it (apart from melting the power connector).

xii5ku wrote:

Hardware requirements for current "OpenIFS 43r3 Perturbed Surface" work:

The following items need to be taken into account, in descending order of concern:
1. Upload bandwidth of your internet link.
2. Disk space.
3. RAM capacity.
99. CPU. This one doesn't really matter, except of course that CPU core count has got an influence on how many tasks you may want to run in parallel, and that core count × core speed influences how many tasks you can complete per day at most. One or both of these factors (concurrent tasks, average rate of task completions) influence the sizing of items 1…3 in this list.

Note, I bolded one word in the first sentence after the fact.

This priority order has been criticised here. I admit that my perspective is somewhat biased, as I am owning several computers with relatively high core count and high computational throughput and am used to be able to fully utilize them. (Although that's not always trivial to accomplish, because many BOINC projects are focused on low core count/ low throughput hosts.)

However, given how the current "OpenIFS 43r3 Perturbed Surface" campaign is going so far, my priority list is – empirically; refer to thread 9167, thread 9178 – indeed quite generally applicable.

Although that's not always trivial to accomplish, because many BOINC projects are focused on low core count/ low throughput hosts.

What on earth do you mean by that? I've never known a project not give out enough work to fill my 126 cores.

I guess you are used to get by with very shallow work buffers, and/or rarely deal with periods of heavily congested project servers.¹

Edit: The solution to this is obvious (partition one wide host into several narrow hosts) and its implementation is straightforward — but arguably not something which one should call trivial.

Edit 2: ¹) Those are the common problems. In addition there are a few special cases, e.g. attachment of a new host to CPDN: Guess how long it takes until there are e.g. the first 64 tasks assigned to a host, given the 3636s request_delay and small number of tasks assigned per work request.

Edit 3: This is my first and last off-topic post in this thread, which already suffers from a low SNR.

Edit 4: "Easy to solve - set a bigger buffer" – Wrong. Goes to show that the poster knows little about quota of tasks in progress, of the significance of scheduler request rates, and more.

When I use a project with heavily loaded servers, or not much work available, that will leave a small computer without work just as easily as a large amount of big computers.

Easy to solve - set a bigger buffer (which I prefer not to as it makes it harder for me to change over to another project quickly, and it means the project gets the work back later), or do more projects. There are plenty of projects which interest me. Anything to do with physics or biology. At the moment my GPUs are on Milkyway and Einstein, and my CPUs are on World Community Grid, Rosetta, and Asteroids. The last three are often out of work, but between all three there's always plenty.

I cannot really run all cores on Boinc tasks. I do not have air conditioning. I can control the fan speeds by BIOS settings. But the maximum I can run the fans at, before their noise gets really annoying, lets me run 8 or so cores in summer and I am currently running 11. The maximum number of cores for Boinc tasks is easily set in the Boinc-manager. And I can set the number of concurrent tasks for a given project in the app_config.xml file in the appropriate directory.

I'm surprised. CPUs generate so little heat compared with GPUs. I have 12 GPUs and 100 CPU cores. Most are in my garage, and I just leave windows open. No, the glass ones not the MS ones! One day I'll join the garage to the house so I can let the heat drift through in winter.

As for noise, just install a bigger heatsink and fan (or even water cooling). My 130W Ryzen 9 CPUs make very little noise running flat out. Two 6 inch fans on a 6x6x6 inch heatsink. Hence slow quiet fans.

A power supply can also put off ALOT of heat especially as the number of cores in use goes up, getting a Gold or above level psu will drastically cut the heat put out by it as well.

A power supply can also put off ALOT of heat especially as the number of cores in use goes up, getting a Gold or above level psu will drastically cut the heat put out by it as well.

Corsair supplies are very efficient, they run under body temperature unless they're loaded to over 50%. None of mine generate any heat, even with a 130W Ryzen. They only gave out heat when I had a GPU on them too.