Raspberry Pi boosts Pi 5 performance with SDRAM tuning

So, I saw this link and immediately thought, &quot;I bet the discussion will be about how we should use N100s, instead.&quot; I wasn&#x27;t disappointed. Even on dedicated Pi forums, you see that happening.<p>I guess I understand this point of view if you were trying to use Pis to experiment with Kubernetes or something. You&#x27;d have built your own (desktop) PC or a personal server rack for that kind of thing, years ago. But for the vast majority of typical uses (Home Assistant, VPNs, etc.) a Pi is going to be <i>way</i> more than you need. It will sit there and silently and reliably run, for years at a time, powered by a USB cable. I know mine have.<p>Why would I consider replacing those with a bigger box, fan noise and a power brick? Maybe I&#x27;m missing something?

Interesting, but why bother. Pi&#x27;s were great when they were cheap, but one can get a better performing Intel N100 box for a similar price and power consumption now.<p>The only reason I can think of to use a Pi is GPIO.

Just tried the SDRAM tweak and got myself a respectable 10% speedup with LLMs. Hat tip.

Points of further interest:<p>= TFA also mentions that the R.Pi 4 can be tweaked [1]<p>= an R.Pi 500 is in development [2]<p>[1] _ &quot;The tweaks can also give the Pi 4 a boost, but the Pi 5 improves more dramatically&quot;<p>[2] _ &quot;Pi 500 Geekbench results starting in September—all seemingly with these tweaks in place already!&quot;

&gt;Changes may roll out in a firmware update for all Pi 5 and Pi 4 users soon.<p>Where can I sign up to not miss the new official firmware release?<p>Edit: well, I&#x27;ll just do the firmware upgrade now, why wait...

Good to read they&#x27;re doing this for the older CM4 too!

&gt; You can actually monitor the temperature of the sdram and it reports if refresh at half or quarter the rate can be done. That allows the overhead due to refresh to be reduced by a half or a quarter which does improve benchmark results.<p>I wonder why this is not available on PCs. tREFI tuning has a big impact on performance, and having an automated guided optimization based on temperature would be nice.

Available as latest in apt now <a href="https:&#x2F;&#x2F;github.com&#x2F;raspberrypi&#x2F;rpi-eeprom&#x2F;blob&#x2F;master&#x2F;firmware-2712&#x2F;release-notes.md">https:&#x2F;&#x2F;github.com&#x2F;raspberrypi&#x2F;rpi-eeprom&#x2F;blob&#x2F;master&#x2F;firmwa...</a>

&gt; That allows the overhead due to refresh to be reduced by a half or a quarter which does improve benchmark results.<p>Does this increase susceptibility to rowhammer?

about &quot;why not use x over rpi&quot; argument, this article itself is a testament to the first-party and community support across the stack that rpi enjoys.<p>for x86 or other arm platforms, these changes comes once in a blue moon. but i do wonder how much further improvements could be achieved from other unexplored optimizations.

Wow... I haven&#x27;t heard of someone using refresh timing to speed up a system outside of retro hardware in ages. I guess I&#x27;d just assumed that after we&#x27;d transitioned to DDR that refresh cycles were just such a small part of overall cycle time that it didn&#x27;t matter. Goes to show what I know!

Raspberry pi 4 has NUMA? The same NUMA thing large servers with 1Tb of RAM have?<p>It looks like it does <a href="https:&#x2F;&#x2F;github.com&#x2F;raspberrypi&#x2F;firmware&#x2F;issues&#x2F;1854#issuecomment-2265721079">https:&#x2F;&#x2F;github.com&#x2F;raspberrypi&#x2F;firmware&#x2F;issues&#x2F;1854#issuecom...</a>

I&#x27;d like to see the numbers around the power usage added too, I see the Pis and audrinos personally as a low wattage device that I don&#x27;t need to worry about the cost, I&#x27;m sure this overcloclocking isn&#x27;t too much but would be nice to see to weight up viability

Find 30% is quite a feat

This is really interesting, thank you for sharing this

Really interesting, thanks for sharing!