theelf wrote on 2024-04-09, 23:42:

Like i said, is pretty clear you did not tested throttle enought, please do, and post again with results

No. This is a fallacious statement. Picking one discrepancy out of the convergence of facts that lead to my conclusion doesn't negate that conclusion. I tested Throttle. Throttle is insufficient. Period. End of discussion.

mockingbird wrote on 2024-04-09, 23:47:

theelf wrote on 2024-04-09, 23:42:

Like i said, is pretty clear you did not tested throttle enought, please do, and post again with results

No. This is a fallacious statement. Picking one discrepancy out of the convergence of facts that lead to my conclusion doesn't negate that conclusion. I tested Throttle. Throttle is insufficient. Period. End of discussion.

Insuficient to what? give me examples i can test in my computers

I did not find yet nothing that fail because cpu speed, and i tested hundred of games, software, etc

Tomorrow i will test your examples and give results back

Throttle Blaster seems interesting. I wonder how well it compares to ACPI and ODCM throttling which aren't a universal solution to slowing down the CPU as some people tout it.

This does look like an intriguing throttling solution.

I wouldn't see it as a wholesale replacement for software throttling, since it does require more setup and the installation of a custom hardware device. From an ease-of-use perspective nothing beats software.

On the flipside, since this uses PWM with an analog encoder it seems a much more granular option than existing software solutions.

I'd be curious to see if it could work with an LGA775 + Pentium 4 combo. Might be a little trickier to rig up given the difference in socket design.

I've been doing some more reading on how Throttleblaster works and how other throttling methods work.

Per the Intel 64 and IA-32 Architecture Software Developer's Manual, they do mention ODCM throttling with respect to Pentium 4 and P6 architecture.

Regarding the Pentium 4 they state:

Modulation of duty cycles is processor model specific. Note that the processors STPCLK# pin is not used here; the stop-clock circuitry is controlled internally.

This was also noted in the CPUSPD tool thread. And it makes sense that this is software controllable via CPUSPD for processors like the Pentium 4, D, and Core 2.

For P6 processors (e.g. Pentium Pro, PII, PIII), they state:

For the P6 family processors, on-demand clock modulation was implemented through the chipset, which controlled clock modulation through the processor’s STPCLK# pin.

Reading up on how Trottleblaster functions, it sounds like it doing exactly that: introducing halt states at specified intervals via the STPCLK# pin. The main difference is that it is being controlled externally via a micro controller and therefore allows for more granular control over the frequency.

From the sounds of it, Throttleblaster is just ODCM throttling done via hardware instead of internally like on a Pentium 4.

With that information, Throttle Blaster may potentially have the same issues as with the other methods (stuttering and/or slow music playback in some situations) but the granularity it offers could mitigate it.

If I remember correctly, for best throttle results without stuttering, CPU frequency must be divided by 8. So for Pentium 4 it's 1600Mhz or 3200Mhz CPU, because Pentium 4 multiplier can't go lower than 12x.

Gmlb256 wrote on 2024-04-26, 13:37:

With that information, Throttle Blaster may potentially have the same issues as with the other methods (stuttering and/or slow music playback in some situations) but the granularity it offers could mitigate it.

This is what I'm wondering.

I'm also thinking it might a bit redundant for processors like the P4 or Core2 which can already be throttled via ODCM using software. Granted the P4 only has 8 steps (Core2 has 16 I think?), so maybe the added granularity of a micro-controller solution would be better. Though in my own experience, I can already throttle my P4 down to 486 speeds with by disabling L1 & L2 cache. That alone solves most of the speed sensitive issues I run into.

I suppose the biggest benefit would be enabling ODCM throttling for processors that don't have it built-in (e.g. AMD).

The Serpent Rider wrote on 2024-04-26, 15:44:

If I remember correctly, for best throttle results without stuttering, CPU frequency must be divided by 8. So for Pentium 4 it's 1600Mhz or 3200Mhz CPU, because Pentium 4 multiplier can't go lower than 12x.

By this I assume then 16x multiplier would be the optimal setting for doing ODCM throttling?

Do you happen to know good games or other applications to test for stuttering?

I'm in the process of testing and benchmarking my P4 setup, so I'd be interested to test this out. I think I've seen Jazz Jackrabbit mentioned as a problematic game, but don't know specifically what the stuttering issue is.

edited: Found a post that suggests the issue with Jazz is related to ACPI throttling, not ODCM.

FWIW, I tried Jazz on my P4 with various ODCM settings and didn't notice any obvious issues. I should note my copy of Jazz is patched so it also runs fine without any throttling.

The Serpent Rider wrote on 2024-04-26, 15:44:

If I remember correctly, for best throttle results without stuttering, CPU frequency must be divided by 8. So for Pentium 4 it's 1600Mhz or 3200Mhz CPU, because Pentium 4 multiplier can't go lower than 12x.

sound stuttering is how I remember Throttling done with ISA DMA/SMM(afaik this is thru ACPI?)- its bad kind of throttling because it uses low frequency resulting in big gaps of no CPU activity. With Throttle Blaster you can use any frequency you want.

Shponglefan wrote on 2024-04-26, 20:07:

By this I assume then 16x multiplier would be the optimal setting for doing ODCM throttling?

Yes.

rasz_pl wrote on 2024-04-27, 03:31:

sound stuttering is how I remember Throttling done with ISA DMA/SMM(afaik this is thru ACPI?)- its bad kind of throttling because it uses low frequency resulting in big gaps of no CPU activity. With Throttle Blaster you can use any frequency you want.

Having done more research on throttling methods, it looks ACPI throttling depends on the chipset implementation.

In the case of Intel ICH5 for example, it uses the STPCLK# pin, same as what the Throttle Blaster is doing:

Within the C0 state, the Intel ® ICH5 can throttle the STPCLK# signal to reduce power consumption. The throttling can be initiated by software or by the THRM# input signal.

But as you say, Throttle Blaster gives direct control over the frequency as opposed to being constrained by a particular chipset implementation.