Talk:Dynamic frequency scaling

Merger proposal

The following discussion is closed. Please do not modify it. Subsequent comments should be made in a new section.

The result was merge into Dynamic frequency scaling

CPU throttling seems to be the same thing. Suggest it be merged into dynamic frequency scaling. Would flesh this page out with links to some examples, and improve that by giving the theory. Zodon (talk) 01:05, 17 March 2008 (UTC)

Yes, it's the same concept, but this (Dynamic frequency scaling) is the correct name and that one is not. Raul654 (talk) 01:51, 17 March 2008 (UTC)

The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.

Re: Proposed merger from voltage and frequency scaling

I think the merger should go the other way. Rather than having small articles on Frequency scaling, Dynamic Frequency scaling and Voltage scaling, seems clearer to merge them all into one article that covers Frequency and voltage scaling, both dynamic and static. Which would contain merger of the aforementioned articles.

1. ) There is considerable overlap in content between the Dynamic voltage scaling and dynamic frequency scaling articles. (Both use the same equation, etc.).
2. ) Frequency and voltage scaling interact - frequently change both at once, rather than just one or just the other.
3. ) At present, the interaction between voltage and frequency scaling is not well covered, it would be nice to have better coverage of that. If they are kept separate then the problem becomes where to put the interaction.
4. ) It isn't clear that there is enough else to be said about each topic to grow the separate articles into more fully fleshed article. (But I am not a hardware guru, so willing to be persuaded otherwise.)

I concur that there are too many articles on this subject, but think that one article could reasonably cover all of these topics: Voltage and frequency scaling, Frequency scaling, Dynamic frequency scaling, Dynamic voltage scaling, and undervolting and overvolting (parts of voltage scaling). As far as I can tell the reasonable place to put such an article would seem to be Frequency and voltage scaling.

Other related articles:

• Underclocking is logically part of frequency scaling, status is debatable, could be left separate, could be merged.
• Overclocking again, logically part of frequency scaling, but is a large enough article to stand on its own.
• Power management - related, but also laps over into ACPI, hibernation, etc.
• Low-power electronics - covers other aspects beyond voltage and frequency.

Those are some of my thoughts on the merge, interested to hear other perspectives. Thanks. Zodon (talk) 02:30, 28 March 2008 (UTC)

Frequency scaling and dynamic frequency scaling are two very different beasts. The former is a technique used to increase performance in next-generation processors, the latter is a power conservation technique. Ditto voltage scaling, which (while done for the same purpose as dynamic frequency scaling) does so in very different ways. Beyond that, however, none of those other articles should exist except arguably overclocking, which is (IMO) well known and distinct enough from dynamic frequency scaling to merit its own article. (Also, overvolting, I think, should redirect to overclocking.) Raul654 (talk) 06:59, 28 March 2008 (UTC)

I agree with this sentiment; DVS and DFS are usually considered together by today's microprocessor designers as the one naturally leads to the other (scale the voltage down and the maximum operating frequency lowers; scale the frequency down and you can lower the voltage), leading to even greater power reduction. So a single DVFS article makes more sense. I would also agree that "overclocking" and "overvolting" are a different thing, although clearly the theory behind "overvolting" comes from the same basic science. Ptoboley (talk) 09:25, 2 May 2008 (UTC)

Voltage scaling is always done with frequency scaling, and frequency scaling is almost always done with voltage scaling. In the VLSI research community they are considered a single technique. Quanticles (talk) 19:38, 20 November 2008 (UTC)

Other reasons to throttle back

The article fails to mention that there are other reasons why you'd want to throttle back CPU performance. Sometimes older software (e.g. 3D games) won't run properly unless the CPU is throttled. I remember Unreal (released in 1998) having a lot of weird issues when played on my AMD Phenom II X4 965 on Linux via WINE when it was running at the stock 3.4GHz, but ran normally after I throttled it back to 800MHz. I'm going to post this both here and in the underclocking article. 24.131.27.231 (talk) 14:30, 23 April 2011 (UTC)

Units?

The phrase "frequency (as a unitless quantity)" seems wrong. Power is energy per unit time so doesn't f provide the /s ? --Chuunen Baka (talk • contribs) 16:31, 19 October 2011 (UTC)

term "Dynamic frequency scaling" used for wifi networks too!

It seems you may find online articles using the term "Dynamic frequency scaling" on wifi networks too! A BGN wifi router can dynamically scale between B and G speeds, and among them have several subdivisions in speed. When a wireless (mobile) device accesses a router at a close distance, quite often, the wifi router will assign a 54Mbps pathway between both devices. When the signal to noise ratio increases between them, like when the distance increases between the router and wireless mobile device, or when a 2,4Ghz interferrence device (like a microwave) is radiating interferrence, the noise might overwhelm the actual usable data on the wireless connection. To prevent a connection to be lost between both devices, the ROUTER generally assigns a lower bandwidth to the pathway. A lower bandwidth would increase signal perception, as well as range slightly, at the cost of lower transfer speeds. For most wireless modems this is not a real issue, as most wireless modems are connected to the internet through a 10Mbit connection. Eventhough their wireless speeds are much faster than their actual DSL/cable connection with the internet service provider, they don't actually make use of most of this gained bandwith, as the device is still throttled by the ISP (ISP is the bottleneck). In other words, most wireless routers might reduce their wifi connection from 54Mbits to 11Mbits, while still provide the same download speeds as before.

I presume this article needs to be expanded to include a wifi section of "Dynamic frequency scaling". — Preceding unsigned comment added by 98.77.246.242 (talk) 07:48, 9 January 2012 (UTC)

Manual?

What about manual adjustments of cpu clock? -96.233.19.191 (talk) 20:58, 11 July 2014 (UTC)

Article (or at least parts of it) is out of date

Under "Performance impact" the following is stated:

"One major difference between the two is that in modern PC systems overclocking is mostly done over the Front Side Bus (mainly because the multiplier is normally locked), but dynamic frequency scaling is done with the multiplier. Moreover, overclocking is often static, while dynamic frequency scaling is always dynamic. Software can often incorporate overclocked frequencies into the frequency scaling algorithm, if the chip degradation risks are allowable."

This hasn't been the case for some time now. While manual overclocking and changing the base clock (what would have been referred as front side bus years ago) are still possible neither of these is currently the recommended method. Manual overclocking can limit the ability of the cpu to automatically adjust its frequency/voltage as required currently and adjusting the base clock can lead to various instability problems because it doesn't only affect the cpu clock but also things like the pcie and other frequencies which you generally want to keep running at what they were designed to run at.

Today the most common and 'best' (in most cases) way of overclocking is adjusting the configuration of the cpu's built in algorithms for adjusting it's frequency voltage. On AMD Zen cpus that would be things like ppt/tdc/edc limits, applying negative voltage offsets using curve optimizer, adjusting the max cpu boost speed and possibly adjusting the pbo scalar. If these values are properly adjusted and enough cooling/power is provided, the cpu will boost to higher frequencies (both on multicore and single core work loads), stay there longer and with enough luck (aka getting a chip that is stable even with high undervolt settings) use less power/produce less heat. And when the cpu load is low it will still lower it's frequency/voltage/turn off unused cores normally. I'm not sure what the equivalent settings on Intel platforms are, but afaik it works pretty similarly.

Memory optimization is pretty popular too, especially on cpus that respond well to increases in memory bandwidth/decreases in memory latency, but since this is a comment about the cpu overclocking statements being out of date I won't comment further about this here.

I would update the part I commented on myself, but since I'm not as familiar with the Intel side of thins as I am with the AMD side I think it's best if someone who is familiar with the Intel side of things writes some more details about it and then we can combine the comments and use them to update the article. 193.95.219.31 (talk) 18:50, 30 April 2024 (UTC)