Future of the CPU board?


yep, Todays overclocking steps are bigger than this..lol wasnt the base clock for a IBM clone 4.7MHz?? those were the days...
yep 4.77MHz Apparently marketing never rounded up back then.
 
And the Switch runs the CPU @ 1GHz, even while having a fan. Not the max 1.9GHz or so that the Tegra theoreticaly supports.
 
And the Switch runs the CPU @ 1GHz, even while having a fan. Not the max 1.9GHz or so that the Tegra theoreticaly supports.

1GHz is also a magic number for the OMAP5.
Up until 1GHz, heat and power usage is pretty linear. From 1GHz to 1.5GHz, power usage and heat production goes up exponentially.
So 1GHz is probably a pretty good max if you want a power saving device.
 
1GHz is also a magic number for the OMAP5.
Up until 1GHz, heat and power usage is pretty linear. From 1GHz to 1.5GHz, power usage and heat production goes up exponentially.
So 1GHz is probably a pretty good max if you want a power saving device.
1 GHz sounds like good default for pyra CPU, with optional turbo mode upto 1.5/1.7 whichever is safe.
 
  • Like
Reactions: rSl
Yeah going from 33MHz to a whopping 66MHz :D

Translation for the youngsters : going from 0.033 GHz to 0.066 GHz :D (a number that doesnt even reaches in to the realms of significance nowadays...)

That would have been on a 486 maybe? By that point in time the 'Turbo' button was a marketing gimmick that resulted in people running with it turned off to, "not wear the computer".

The original 'Turbo' button who's purpose was to slow the CPU down to 4.77 Mhz. Nobody would want a computer with a 'slow' button. But - a fair number of original PC computer games used the CPU cycles as the clock. If the CPU was running at 8 or 10 Mhz, those games instantly became both hilarious and unplayable.

So, by those traditions, a 'Turbo button' is not to make it go faster, but to slow it down via some form of underclocking.
[doublepost=1558021656,1558021442][/doublepost]
1 GHz sounds like good default for pyra CPU, with optional turbo mode upto 1.5/1.7 whichever is safe.

Actually - the reverse is likely true. The default should be 'ungoverned' and subject to thermal throttling to keep it in check. What you are referencing as 'Turbo' would actually just be 'flat out' - i.e. no influence from the 'Turbo Modifier'.

The Turbo Button's actual purpose isn't to make it go faster - it is to engage a set of circumstances to make it run slower but consistent. So, unlike the TV show Knight Rider, the turbo button being engaged deliberately makes things slower and more consistent.
 
You can see the fan in the Nintendo Switch at 2:11
Ok, I didn't knew that. Is the fan loud? ^^"
And the Switch runs the CPU @ 1GHz, even while having a fan. Not the max 1.9GHz or so that the Tegra theoreticaly supports.
The Switch can squeeze (compared) impressive graphics out of these 1GHz, thanks to the dedicated GPU unit. The Pyra also has a GPU part in the SoC but I guess not THAT kind of GPU that the Tegra has.
1GHz is also a magic number for the OMAP5.
Up until 1GHz, heat and power usage is pretty linear. From 1GHz to 1.5GHz, power usage and heat production goes up exponentially.
So 1GHz is probably a pretty good max if you want a power saving device.
I guess it depends on the power management software if and when a "turbo" is activated? Like mentioned already, short spikes up to 1.5GHz for short demanding stuff like rendering a webpage or menu operations or whatever. Gaming and constant load situations are maybe locked then at 1Ghz. At least I hope the Pyra get's an clever power management software.
 
Provided your 486 was fitted with its standard anodised light blue heatsink, it ran no less consistently in turbo 66MHz mode than in 386-compatibility 33MHz mode. I'm assuming PC gaming wasn't a thing in the original PC or AT 286 era, but took off during the 386 era, so back in the early days you'd just be impressed that your copy of Lotus 123 calculated its cells faster.

Even at the same clock rate a 486 would be a little faster than a 386, primarly due to the on-board cache, but I guess it was close enough to force a lot of games to slow down and become playable again.
 
The CPU of the GPD Win1 has a feature called "Turbo-boost". It pushes the CPU clock higher for short bursts.
It might be fine with some uses, like web browsing where you'll need that boost very occasionally and for very short durations. The idea is that these "boosts" will be followed by periods of low activity when your CPU will cool down.

But when you're playing intensive games, you don't have these periods of low activity. So what happens is that the Turbo-boost starts, the CPU heats up and has to throttle down to something lower than the base clock. Generally speaking, disabling Turbo-boost gets better performances since the CPU doesn't heat as much and can stay longer at a good performance. It's better to be consistently at 100% than to be at 150% for 2 minutes and then have to be at 75% for the rest of the time.
 
Yes, modern turbo boost is more real than turbo mode on a 486 used to be, which was actually a slow down for compatibility. Modern turbo modes are designed to run the CPU out of its designated thermal envelope for short periods, although in the land of desktop computers this has resulted in massive aftermarket coolers that can keep even a boosting CPU relatively cool. But in any kind of laptop or handheld you're pretty much limited to the manufacturer's spec cooler.

For games it's more important to know what it can consistently pull and stick to that, because sudden drops in framerate are likely even more distracting than just general slowness.

It remains to be seen how our Pyra's perform with their ARM cortex-A15 CPU cores. I assume the DBP system can specify clock speed to the things it includes like the old PND system could, but early software might specify that more wrongly as we all develop a better understanding of the thermal performance of this setup over time.
 
Provided your 486 was fitted with its standard anodised light blue heatsink, it ran no less consistently in turbo 66MHz mode than in 386-compatibility 33MHz mode. I'm assuming PC gaming wasn't a thing in the original PC or AT 286 era, but took off during the 386 era, so back in the early days you'd just be impressed that your copy of Lotus 123 calculated its cells faster.

Even at the same clock rate a 486 would be a little faster than a 386, primarly due to the on-board cache, but I guess it was close enough to force a lot of games to slow down and become playable again.

Actually, the, 'Turbo Button', originated in the XT clone era. It's existence was born form gaming needs to slow down XT clones and AT (286) computers so that games written for the PC era could be made playable. The IBM PC 5150 and XT 5160 ran 8088 CPUs at 4.77Mhz. Games used the CPU cycles as their clock.

PC gaming was a thing clear back in the days of DOS. Any game on this list that was released prior to about 1985 would have had to 'slow down' on later hardware. The 'Turbo Button' wasn't meant to speed up a slow computer - it was meant to be able to slow down a fast(er) one.
https://en.wikipedia.org/wiki/Index_of_DOS_games_(A)
Note that is one page of a 27 page list.
 
Thinking about Pyra throttling, it sounds like control system theory has a part to play.

There needs to be something which throttles the CPU in response to rising temperature, but still allows it to run fast in short bursts. If it's too slow to react, the CPU temperature will overshoot beyond safe thresholds (or a hardware-level cutout will be triggered). If it reacts too rapidly, the whole system will start 'ringing'.

For example, as soon as the CPU temperature passes 80°C, the CPU is throttled to 100MHz to cool it down as quickly as possible. Then, as soon as the temperature falls below 70°C, the clock runs back up to 1.7GHz and it all heats up again. You'd end up with cycles of high performance, interspersed with periods of very poor performance, which would make the whole system horrible to work with.

There must be a critical function relating temperature to clock speed such that power-hungry applications cause the CPU to heat up to a maximum safe temperature as quickly as possible, and then stay there. I wonder what it is.
 
Yes, I remember ED saying the current throttling measure is too harsh on the prototypes at present, although I think that was maybe a month or two back. He was looking for anyone with knowledge of this stuff for help I think.

There was also this posted to LWN earlier today, but it's behind their subscriber wall at present. I assume it's more about the sort of workloads running on servers; games generally run everything in a small handful of processes, so the scheduler can't control how much time they spend running the GPU or anything like that.
 
ARM have announced the availability of A77 designs to licensees (or at the very least, they've 'launched' the A77 whatever that actually means). The information focuses on peak throughput rather than what you actually get if you have a hard power limit, other than vague words to that effect.

Nobody's making cores out of this just yet, let alone any that would actually sell parts to ED, but it might be interesting for people as an indication of future developments.
 
Back
Top