Overclocking discussion

Riverstorm

I've been testing different cases lately too. I was doing some testing with thermal paste and tape with different cases in another post but it might do some good here. I just pasted them all together so if they don't make sense just let me know. The two cases I was mainly testing were the Flirc 2 and BIQU. Mentioning the Wicked Aluminum but that one is quite expensive but I hear great things about it.

I've been testing the thermal paste vs tape we talked about a while ago on a Zero. They both seem to do well. I don't think paste is a clear knock out but maybe a preference and definitely an alternative if you don't have one or the other handy.

It's fairly rock solid when cool but when heated up it's a bit slippery but seems to sit fine. It's easy to break the cohesion if bumped when cool which would be a CPU killer on a real PC or a degradation in cooling but doesn't seem to effect the Pi to much when heated back up.

It can be a bit messy and a bit of trial and error getting the correct amount but it does work well. The tape I have been using is easy to pull off in a single piece. I've had some heatsinks where it is gooey and messy to remove though once the heatsink was removed. Inside a larger unit [arcade style unit] I would do paste but if it's a small case that's portable being shoved in a bag, pocket, etc. I would probably use tape.

If I get some time I will sit down do more exact temp difference.

I was a bit disappointed in the Flirc 2 case for cooling. Aesthetically it's my preference [for looks] but I just ordered a BIQU from what I read it cools better due to 3 points of contact [main SoC, Ethernet & memory] chips with a solid aluminum column vs. the Flirc which covers only the main chip and it's a hollow column not to mention the rubber top covers the hollow column [hole] which I would say holds in more heat.

That nice looking "Wicked Aluminum" case I hear cools incredibly well with it's 3 chip contacts and paste. It's solid aluminum and screws together tightly so no slippage but when the BIQU is similar and about 6 times cheaper I'll give it a shot even if a few degrees warmer. Basically I am hoping to get rid of the fan and go all passive cooling overclocked.

The original tape that comes with the Flirc is thick. It's cut smaller by quite a bit from fully fitting the main SoC, not sure why. So I tried cutting a proper size piece from stock I had but it's to thin to fill the gap between column and chip. I read Artic (not Artic Silver) makes a thermal tape of different thicknesses.

The BIQU's columns are thinner/smaller than the on board chips so they only cover a portion of the chip (save some aluminum by the manufacturer) but it's solid aluminum. Also you can't use the case mounting holes on the back of the case as even low profile pan head screws would come in contact with the board and short out on the bottom. They cut one big hole to cover all ports instead of individual like the Flirc

The Flirc has nice attention to detail cutting around each port. The two tone case and rubber top are nice looking and feel but not geared as well toward cooling. Plastic bottom with holes.

The Wicked Aluminum did all things right the first time but at a ridiculous cost as the new B+ with an IHS (looks like it only covers the top of the SoC instead of enclosing it from the pics) almost makes it void now as I am sure the B+ will not even fit that case. The Flirc might be ok as the gap is fairly large and paste isn't even an option with the Flirc so the IHS might narrow it enough.

I read it's beneficial to take the time to sand the contact points to remove the paint on the BIQU for bonus cooling points. Keeping it level is key like lapping a CPU. I should have it this weekend to mess around with. Well I ordered it Amazon Prime anyway as we keep a subscription but Prime hardly means anything anymore. It seems very select on what arrives in two days.

Yes I've seen thermal compound "crust" when properly applied (thin enough) and screwed down after enough time but usually a few years. I think AS5 recommends around a certain number of cycles of heating and cooling to cure and maximize cooling.

It does look like it would work but without it being clamped down in some fashion it seems like it would loosen up between cycles or seasons. I know here we have extreme dry in the winter to extreme humidity in the summer and that makes a big difference. The freezing temps in the winter cause "frost heaving" pushing up solid concrete, footings and building foundations can be moved unless you keep them below the frost line but most contractors in this type of climate know better. Probably not quite the same thing but doors are more snug in the summer when full of moisture and loose in the winter when dry, windows, furniture, etc. That humidity can expand or lack of can crack and dry things out pretty badly.

The viscosity of compound changes when it's hot too making it more suspectable to moving. I did tin the SoC and just put a dab in the center. which misses the extreme corners if you look at the spread pattern. I remember going through that stage of applying and reapplying paste to a CPU to find the maximum cooling for your PC. You can get them (the corners) if you don't mind some excess squeezing out the sides but some of the metal based ones are electrically conductive and would require cleanup due to shorting. With some of the dual core CPUs Artic Silver recommends a thin line down the center but there's not enough room on the Pi. I dislike it when it squeezes out the sides and stains the motherboard CPU socket around the chip.

A good "clamp" might gain you a few degrees. Either way I agree I think paste or a good quality tape are going to keep the cooling well within spec.

Just to follow up on the BIQU case. It does cool much better than the Flirc 2.

The columns are much slimmer than I thought from looking at pics. They also have "dimples" in the contact surface. Not sure what the reasoning is even though it's a molded case. Two dimples in the main SoC column and one in the ethernet(?) chip. The underside memory(?) chip had no dimples but the surface was convex and required sanding it down.

The contact surface on the columns sanded easy revealing the metal. The underside chip took a bit more work as it was convex and also tight to work inside the lip of the case. The top columns are open and easy to work with. I used a fairly high grit sand paper as my brother does auto body but 300 grit or so from the hardware store works. I used a small block to keep the sanding level.

I did a "cold press" with thermal paste and the columns hit near dead center. I used Shin-EtsuMicroSi vs AS5 as the viscosity is almost clay like-ish and it worked nicely to fill the dimples and set well.

I was testing in an ambient of around 75F/24C. With the BIQU I leveled off around 56-57C. The Flirc 2 was around 74-75C. That's with all settings overclocked and a voltage bump.

Another noticeable drawback of the BIQU case is no visible lights as they are completely enclosed inside the case.

I like the BIQU as it cools easily in the 50s with no fans or extras. It's quiet and it's a decent case with a decent finish for around $12 US which is a good bit cheaper then the Flirc 2 even though I like the Flirc 2 look better.

Something I found interesting or I didn't think about really. Is the when I removed the Pi from the Flirc it was "stuck" it took about 10 minutes of gentle prying around the edges of the board to get the tape to break loose. There's no easy way to pull them apart as the Pi is slightly coutersunk into the top.

The other thing I noticed the tape was roughed up around the edges but smooth and flat in the center, basically untouched. I came to the conclusion that not all the SoC's are completely level so that chalks one up to paste being a better option in some scenarios. Plus with paste it [cases] comes apart much easier.

quicksilver

@riverstorm said in Overclocking discussion:

I was testing in an ambient of around 75F/24C. With the BIQU I leveled off around 56-57C. The Flirc 2 was around 74-75C. That's with all settings overclocked and a voltage bump.

Are those the temps at idle? Or while running a program? If so, which program. Im curious how your numbers stack up against my Kuma 9k case. Great info btw.

Riverstorm

@quicksilver said in Overclocking discussion:

Are those the temps at idle? Or while running a program? If so, which program. Im curious how your numbers stack up against my Kuma 9k case. Great info btw.

They were running a game. More of a real world test. ;) I set the Governor to performance so it stayed pegged at 100% for 4 to about 6 hours. The game was Dungeons & Dragons: Shadow over Mystara (ROM=ddsomu.zip) in lr-mame2003 and also a sdbench script (it's pasted in a Wiki somewhere).

If you have another test you like to benchmark with just let me know so it's more apples to apples that might be better. I hope to get a chance this weekend to do some overclock testing.

What do you think of this guys script? It looks like a nice battery of tests here. I found it trying to find that sdbench stress test I used. It [sdbench] seems to focus more on reading/writing and memory than graphics though.

What are you full overclock settings? I can't seem to find a link to the Kuma 9k when Googling. If it's passive and cools better I'll give it a whirl.

Alturis

I would be curious to hear of any numbers used for the Pi3B+

I was overclocking my Pi3B at about 1400 before but now that the baseline is 1400 with a B+ I would love to hear a suggestion for a stable overclock value for it.

I have a flirc2 case for it. Had to shave the heat sink "paste" pad down to about 25% thickness to get the B+ to fit properly in the case.

Riverstorm

@alturis said in Overclocking discussion:

Had to shave the heat sink "paste" pad down to about 25% thickness to get the B+ to fit properly in the case.

That's interesting mine fit without modification. There must be some variance in the manufacturing process or maybe they ship different thermal tape to what's on hand.

Alturis

Yeah, actually the fit is still not exactly perfect. It's pressing down pretty tightly on it compared to how it is with my Pi3B causing an ever-so-subtle bend in the board.

quicksilver

@riverstorm https://www.amazon.com/SNES-inspired-Raspberry-Case-functioning/dp/B079T7RDLX

Comes with a custom heatsink and thermal paste

If you really want to test the cooling capabilities of a case, try running cpuburn-a53. This will test your cooling and if your power supply is adequate. If you can run it without throttling, your cooling is very good.

quicksilver

@alturis I have heard of people getting the 3b+ up past 1500mhz. The problem is that people claim all sorts of numbers with no proof of stability. For example my pi3 will boot at 1475mhz but is only stable at about 1350mhz. Anything past that and eventually I will get a crash if I run a particularly taxing emulator.

Riverstorm

@alturis said in Overclocking discussion:

causing and ever-so-subtle bend in the board.

There's a certain amount of flex to solder and circuit boards but at some point I think the stress causes micro fractures in the solder and creates issues.

In the past I've baked 2 graphics cards back to life and just recently a completely dead LG TV circuit board. 350 degrees for 10 minutes is basically like a solder reflow and corrects those micro-fractures. Cheaper than buying a solder reflow oven but probably not the best for the 'ol oven so to make the wife happy we run it through a cleaning cycle after which takes roughly 3 hours.

@riverstorm https://www.amazon.com/SNES-inspired-Raspberry-Case-functioning/dp/B079T7RDLX
Comes with a custom heatsink and thermal paste.

Oh that's a nice heatsink not sure the BIQU case can beat that one. The aluminum casing might help it along vs. plastic but you can't beat the aesthetics.

I have heard of people getting the 3b+ up past 1500mhz. The problem is that people claim all sorts of numbers with no proof of stability. For example my pi3 will boot at 1475mhz but is only stable at about 1350mhz.

Yeah even liquid cooling can only take you so far before you fry the silicone.

Alturis

@riverstorm said in Overclocking discussion:

In the past I've baked 2 graphics cards back to life and just recently a completely

Well at 35 bucks for an entirely new Pi3B+ I am not overly concerned. ;)

dirthurts

@Alturis I have mine at 1500 MHz right now, but i can push it a bit further. I'm, however, keeping my over volt locked at 2 to avoid producing too much heat and consuming more power. Like others, I have no proof, but I'm not sure why anyone would make it up really.
It seems quite feasible that a good chip could get close to 1.6 GHz with a little luck and proper cooling.

I'm currently using the Flirc gen 2 and can't make it throttle under any conditions, so it's providing plenty of cooling.

Riverstorm

@alturis said in Overclocking discussion:

Well at 35 bucks for an entirely new Pi3B+ I am not overly concerned. ;)

Sorry to clarify I meant actual PC video cards like a GTX 980 Ti that gets you into that upper midrange around $400 or so for the video card only. I figure a quick bake is worth a few minutes. The TV was completely dead and it was a known issue with LG TV's so I ripped it apart pulled the board and sure enough that was the issue. Also a standoff was missing so I went to the local hardware store and bough a threaded nylon standoff, cut it down to size and it set the board rock solid. The TV works well again. Surprisingly (or not) poor manufacturing techniques are more and more common. My main stay has always been PC computers that I build from scratch. It's nice as it allows you to choose the exact parts you want.

I'm currently using the Flirc gen 2 and can't make it throttle under any conditions, so it's providing plenty of cooling.

@dirthurts So you have your overvoltage set to 2 and your able to get your arm_freq to 1500? What are your other settings? Have you modified the Flirc 2 case at all? I have a Flirc 2 here next to me and under load I was in the 70s Celcius with and ambient of about 75F/24C and my arm was only at 1260. It did not cool well at all. You must be in that 5% on top of the bell curve of production variance. I have 4 Pi 3's and all of them are average. Three from Element 14 and one from Adafruit.

quicksilver

@riverstorm I think he is referring to a pi 3b+

quicksilver

I posted this in the other thread but wanted to post it here as well for posterity (nobody will see it over there):

I have found that quake 3 is a good resource for testing overclocks. I run all the automated tests that are recommended first like memtester and stress and if I get a passing grade then I test with quake 3 . Often times clock speeds that pass the automated tests fail miserably when put to a real world test like quake 3. Quake 3 works the GPU/CPU and RAM pretty hard. I found it to be more stressful than the individual tests and in most cases if your overclock is unstable the game will crash within a few minutes. However to verify my overclocks stability I will usually let the game run in first person spectator mode for 4-5 hrs at least. I have had quake 3 freeze after 1-2 hours if the overclock is unstable. So thorough testing is definitely needed when overclocking. I doubt most people test thoroughly enough and are running around with unstable overclocks.

the freeware version of quake 3 should work fine. I believe it's listed as quake3 under optional packages. Easiest way to test is to load a skirmish, load a bunch of bots (8 seems a good number). Set kill count and time limit to 0. Once you get into the game press ~ key to open the console. Type /team spectator

Then press enter. Press ~ again to close the console. You are now spectating. Press the fire button and it will place you in a first person view of one of the bots. Now the game will run until you stop it or it crashes. Make sure you have good cooling because it will (as you noted) get your pi very warm. I am using a kintaro 9k case that has an awesome heatsink. For me the temp will climb to 69C and hover there indefinitely. Before i had that case I had a small heatsink and my pi would overheat within about 15 mins of quake 3.

Riverstorm

@quicksilver said in Overclocking discussion:

iverstorm I think he is referring to a pi 3b+

Ah, ok. So the 3b+ does fit in the Flirc 2? I was wondering if the IHS (making a bit higher) would fit under the heatsink column in the case.

Thanks for the post with the Q3 instructions. Having multiple bots seemed key. With 1 I had no issues so I bumped it up to the recommended 8 and there's so much graphical activity on the screen I see why it's taxing all components. It's been years since I played Q3 I swear it was slower back then, must be getting old.

quicksilver

@riverstorm I think I heard that a pi 3b+ will fit in a flirc 2 case, although I have no personal experience with it.

quicksilver

Just watched a video ETA Prime put out, looks like he got his Pi 3b+ to run Conker's Bad Fur Day . Looks like a very playable framerate too. His overclock settings are a bit wonky though.

Riverstorm

@quicksilver said in Overclocking discussion:

Just watched a video ETA Prime put out, looks like he got his Pi 3b+ to run Conker's Bad Fur Day .

I am not quite as up on consoles games. I do have the classics mainly. One of the challenges for me on the N64 is trying to get the d-pad (hat) working. I tried using lr-mupen64plus. Everything seems to work fine except making changes for saving/loading. For incrementing/deincrementing my save slot I can press hotkey+right (or left) which is defined in all/retroarch.cfg and that works fine. To save/load I have hotkey+up (or down) defined in all/retroarch.cfg but that doesn't work. What works is defined in the mupen64plus.cfg.

It has in that config file "J0B8/B4" which works fine but I wanted to change it to "J0B8H0V4". Looking at the docs on mupen64plus. It's Joystick 0 button 8+button 4. So that longer string with the H is Joystick 0 button 8+hat 0 direction down (V is value and 4 is down). It's pretty straight forward but the emulator doesn't seem to recognize the H variable defined on the mupen64 docs page.

On topic before doing any testing I have the Cana Kit power supply it's black (US plug) but on their site I see they have a white one (universal plug) that says official are they the same? I can't really find the answer/difference between the two but they both show as certified working with all Pi versions.

quicksilver

@riverstorm What controller are you using?

I have the canakit black plug too. I would assume its the same as the white one, if the specs are the same.

Riverstorm

It looks to be the same specs https://www.canakit.com/official-raspberry-pi-power-supply.html