@Rayman said:
Ok, seems need to upgrade regulators to 2 A from 500 mA, like Eval/Edge to do the crazy overclocking.
Was definitely pushing the 500 mA past the limit...
I like the AP3445L from Eval. 2 A and adjustable. Will use for both 3.65 and 1.8 V.
TI have new parts that have ferrite bead filter support, which should allow lower noise on the 1v8 rail ?
A bit more current headroom means less power loss and tolerance of brief peaks, especially if over clocking is the prime target.
TPSM82913 17-V VIN, 3-A low-noise and low-ripple buck module with integrated ferrite bead filter compensation
TPSM82916 New 17V VIN, 6A low-noise low-ripple buck module with integrated ferrite bead filter compensation
That's interesting @jmg... By "integrated" ferrite bead, they mean external ferrite bead? Guess the compensation is integrated? I don't really understand that.
Guess it's a way to reduce ripple?
Anyway, not taking any chances now, just copied exact parts from Eval board.
Using them for both 1.8 V VDD and 3.65V VIO.
Will be interesting to see if it's really just higher VDD current that is needed to overclock or if raising VDD really makes a difference.
Just occurred to me that should have replaced barrel jack and 5V regulator with USB-C connector for additional power.
That can provide more current (1.5 A or 3.0 A) and don't need the regulator...
Probably more efficient too...
@Rayman said:
Just occurred to me that should have replaced barrel jack and 5V regulator with USB-C connector for additional power.
That can provide more current (1.5 A or 3.0 A) and don't need the regulator...
Probably more efficient too...
Makes sense. Once you go to USB-C you can start to leverage it's power delivery stuff and avoid the need to regulate higher voltages down on your own board. You may want to find some handy PD capable chip for this. Search for USB trigger board and you'll see a bunch of them.
@rogloh Was looking at PD stuff. That looks aimed at getting higher voltage input via negotiation with charger. It's a way to get super high power. But, maybe not so useful here?
Actually, could be useful for large LCD panel with backlight that is power hungry. Maybe. Seems most take 5 V anyway and I'm sure 1.5 A is enough...
Maybe 20 V would be good for charging 18V battery or something?
I'm not really seeing a use case for higher voltage input...
@Rayman said:
@Wuerfel_21 There are notes in the instructions on how to map Z to D. Just surprised didn't have actual D button...
All the buttons should be mapped to something by default. You only need to remap if it's uncomfortable or strange.
Having a "D" button is stranger than not having one. It's kindof a NeoGeo-ism.
@Rayman said:
@rogloh Was looking at PD stuff. That looks aimed at getting higher voltage input via negotiation with charger. It's a way to get super high power. But, maybe not so useful here?
Actually, could be useful for large LCD panel with backlight that is power hungry. Maybe. Seems most take 5 V anyway and I'm sure 1.5 A is enough...
@Rayman said:
Saw a Black Friday deal might have to get…
That’s only 30W though… enough? Maybe for two wheels? Have to look more…
30W is pretty useful tbh. The surge current is to get things moving (or in stall), so what' be really neat is to control things knowing you have to stay within that constraint.
As an example the parallax servos are rated at 6v nominal, 1.2A stalled, so 30W would cover at least 2 stalled servos even at 8.4V max, and be able to recover
The motors we use in our industrial conveyors are only 90W, so similar to laptop adapters (bricks) nowadays. I would love someone to make a really efficient USB PD variable speed drive that could run a 60W AC induction motor from 100W USB PD.
Yup, a wide solid ground plane is very important at frequency. The situation is not unlike structural engineering for handling wind. But we have the luxury of blanket fills, equivalent of using solid sheets of steel, to cover every case without needing to do any of the maths to resolve for strength to weight efficiency.
I actually just pulled out one of your boards for having a go on the RAM tester.
The RAM tester as-released instant-crashes the board. I have to remove the two coginit(COGEXEC_NEW,@dummy_entry,0) lines that I use to keep all the extra cogs busy to get it to work (or more accurately, the in-development version that didn't have those yet didn't crash and then it started crashing when I added those). I feel that points very decisively at the 1V8 core supply being at the heart of that issue. If you add waitms around those, you can watch it crash in real-time. So it's not a spike problem, it just can't keep the core alive under high load.
This is on the newest one I have, I think. That was always the most crash-prone one. It has the electret mike and the quiet VIO regulators. I also gave it a heatsink at some point. Interestingly, if the dummy cogs are disabled, the RAM test seems to be working in fast mode just fine. So I guess there was never a design problem with the PSRAM section? I just blamed the general instability on the PSRAM.
Wow, this shit is really potent. I went through all your boards and the RAM tester kills all of them, instantly. It's a bloodbath! Sometimes they even get into a super hard crash state where you need to power-cycle, the normal reset doesn't work.
They all work fine with the dummy cogs disabled (test passes for at least 1 minute in fast mode with 14, async, async)
@Rayman said:
Seems to be important that all wires see the same capacitive loading..
True.
That gives me an idea…
If add actual capacitance to each signal, maybe can share 16 bit bus with other things…
That may help with the skew between bits/lanes however it will also likely reduce the upper frequency the board will work at if the IO slew rate is reduced with even more additional capacitance. So you may win and lose at the same time.
Comments
TI have new parts that have ferrite bead filter support, which should allow lower noise on the 1v8 rail ?
A bit more current headroom means less power loss and tolerance of brief peaks, especially if over clocking is the prime target.
TPSM82913 17-V VIN, 3-A low-noise and low-ripple buck module with integrated ferrite bead filter compensation
TPSM82916 New 17V VIN, 6A low-noise low-ripple buck module with integrated ferrite bead filter compensation
That's interesting @jmg... By "integrated" ferrite bead, they mean external ferrite bead? Guess the compensation is integrated? I don't really understand that.
Guess it's a way to reduce ripple?
Anyway, not taking any chances now, just copied exact parts from Eval board.
Using them for both 1.8 V VDD and 3.65V VIO.
Will be interesting to see if it's really just higher VDD current that is needed to overclock or if raising VDD really makes a difference.
Just occurred to me that should have replaced barrel jack and 5V regulator with USB-C connector for additional power.
That can provide more current (1.5 A or 3.0 A) and don't need the regulator...
Probably more efficient too...
The adafruit usb-c connector looks compatible:
https://www.adafruit.com/product/4458
The board side slots are 1.2 mm instead of 1.4 mm but that should be ok…
In stock at mouser
Interesting that there is such a wide variety of footprints out there…
1.2 Guess I'll upgrade and try again...
Ok, yeah, old controller works with latest.
New Sega Genesis works too. Not sure why it doesn't have a D button though...
Metal Slug seems fine, but Last Blade crashes right away. Maybe means first bank is OK, but second bank not so much?
You mean none of the buttons on the pad get mapped to D? On the Sega-style controllers, either Y or Z should be mapped to D.
Makes sense. Once you go to USB-C you can start to leverage it's power delivery stuff and avoid the need to regulate higher voltages down on your own board. You may want to find some handy PD capable chip for this. Search for USB trigger board and you'll see a bunch of them.
@Wuerfel_21 There are notes in the instructions on how to map Z to D. Just surprised didn't have actual D button...
@rogloh Was looking at PD stuff. That looks aimed at getting higher voltage input via negotiation with charger. It's a way to get super high power. But, maybe not so useful here?
Actually, could be useful for large LCD panel with backlight that is power hungry. Maybe. Seems most take 5 V anyway and I'm sure 1.5 A is enough...
Maybe 20 V would be good for charging 18V battery or something?
I'm not really seeing a use case for higher voltage input...
All the buttons should be mapped to something by default. You only need to remap if it's uncomfortable or strange.
Having a "D" button is stranger than not having one. It's kindof a NeoGeo-ism.
You can still use them to fall back to 5V with higher current limits IIRC. But maybe some of that can be done with the resistors as well if you are willing to take risks, see this:
https://forum.digikey.com/t/simple-way-to-use-usb-type-c-to-get-5v-at-up-to-3a-15w/7016
@rogloh yes, that’s the plan
Was thinking about fuses though…
Might just leave off…
@rogloh actually did just think of use case —> robots
The big parallax robot wheels can take 20v and there are usbc battery banks that can provide 20v @ 1.5 A. That might be something…
Saw a Black Friday deal might have to get…
That’s only 30W though… enough? Maybe for two wheels? Have to look more…
30W is pretty useful tbh. The surge current is to get things moving (or in stall), so what' be really neat is to control things knowing you have to stay within that constraint.
As an example the parallax servos are rated at 6v nominal, 1.2A stalled, so 30W would cover at least 2 stalled servos even at 8.4V max, and be able to recover
The motors we use in our industrial conveyors are only 90W, so similar to laptop adapters (bricks) nowadays. I would love someone to make a really efficient USB PD variable speed drive that could run a 60W AC induction motor from 100W USB PD.
/me kicks, slaps and punches Tubular.
Fine, but I still want to see it exist :-P
I see there are 240W usbc pd power banks now. That’s the max, looks like.
That’s quite a bit of power. Maybe fire hazard…
Got new board but AP3445L was driving me crazy. It refuses to operate above around 3.4 V although data sheet says it does.
1.8 V is fine though.
Fortunately, don’t actually need 3.65 V now just jumpered to 5 v for now.
Was thinking that might actually be the Vdd current and not voltage that was limiting max freq
But, new board shows it is actually Vdd.
Only does 335 MHz at 1.8 V, but can do 370 MHz at 2.0 V. Needs cooling after a few minutes though.
Want to crank up higher but want to test other things before breaking P2…
Yikes, removing ground plane around PSRAM appears to have been a horrible idea. It's all kind of messed up on new board...
How does it compare in terms of performance?
Yup, a wide solid ground plane is very important at frequency. The situation is not unlike structural engineering for handling wind. But we have the luxury of blanket fills, equivalent of using solid sheets of steel, to cover every case without needing to do any of the maths to resolve for strength to weight efficiency.
@rogloh
It doesn’t work at all with chips driver.
MegaYume can run in 4 bit mode, but each 4bit lane seems to need a different setting…. It’s ugly
Have more appreciation for previous version now!
Seems to be important that all wires see the same capacitive loading..
That gives me an idea…
If add actual capacitance to each signal, maybe can share 16 bit bus with other things…
I actually just pulled out one of your boards for having a go on the RAM tester.
The RAM tester as-released instant-crashes the board. I have to remove the two
coginit(COGEXEC_NEW,@dummy_entry,0)lines that I use to keep all the extra cogs busy to get it to work (or more accurately, the in-development version that didn't have those yet didn't crash and then it started crashing when I added those). I feel that points very decisively at the 1V8 core supply being at the heart of that issue. If you addwaitmsaround those, you can watch it crash in real-time. So it's not a spike problem, it just can't keep the core alive under high load.This is on the newest one I have, I think. That was always the most crash-prone one. It has the electret mike and the quiet VIO regulators. I also gave it a heatsink at some point. Interestingly, if the dummy cogs are disabled, the RAM test seems to be working in fast mode just fine. So I guess there was never a design problem with the PSRAM section? I just blamed the general instability on the PSRAM.
Wow, this shit is really potent. I went through all your boards and the RAM tester kills all of them, instantly. It's a bloodbath! Sometimes they even get into a super hard crash state where you need to power-cycle, the normal reset doesn't work.
They all work fine with the dummy cogs disabled (test passes for at least 1 minute in fast mode with 14, async, async)
True.
That may help with the skew between bits/lanes however it will also likely reduce the upper frequency the board will work at if the IO slew rate is reduced with even more additional capacitance. So you may win and lose at the same time.
@rogloh true but there are a lot of things that don’t need extreme overclocking.