Hacker News Reader

Comment by Causality1 | Hacker News Reader

I don't doubt that it's better for everyone that Intel and AMD are getting closer to the theoretical max performance of chips right out of the box, but I do miss the days when you could work up the nerve to start playing with the voltage and frequency settings and get a free performance boost well after your purchase date.

View on HN

>but I do miss the days when you could work up the nerve to start playing with the voltage and frequency settings and get a free performance boost well after your purchase date.

On the flip side, I'm glad that I can get the full performance of the chip without having to spend hours tinkering with it and/or worrying about silent data corruption.

I for one hated gambling on overclocks that could brick my expensive hardware. I’m glad the equipment being sold today comes at max performance out of the box and that the reputable company making it will support it.

Regarding the max-performance-out-of-the-box, unfortunately you're mistaken. Part of the reason people could often get insane overclocks out of their hardware was that AMD and Intel would periodically produce a higher than expected number of processors that could handle a greater clock speeds. And rather than cutting prices on those parts they'd simply configure and sell them as a lower-spec'd part - so back in the day AXIA core Athlons may have been sold at 1GHz, but would often comfortably hit 1.33GHz and beyond[0]. This still happens to this day[1], nothing's different about that.

Whether you're comfortable going down this unsupported/warranty-breaking route is another story, but it's certainly not the case that all processors nowadays are sold at their peak possible performance out of the box.

[0] = https://hexus.net/tech/reviews/cooling/186-axia-overclocking

[1] = https://www.anandtech.com/show/16857/overclocking-with-intel...

I embarrassed myself recently by trying to get the performance of my AMD Zen 3 CPU beyond what precision boost overdrive and the default XMP profile of my memory were able to achieve. The single threaded performance never came close to it with the kind of old school static overclock + overvolt I was running with my previous Skylake Intel chip.

I did end up getting some promising results by using the curve optimizer feature of PBO which allows you to change the voltage/clock curve on a per-core basis but while it was stable at higher clocks, it did have some trouble while idling or performing some tasks that weren't running the CPU 100%. I had to admit that I was doing all of this more for the excitement of seeing high numbers and not for the actual practical benefits. Testing for stability with non-demanding tasks just wasn't exciting enough to justify continuing.

Well, to satisfy your impulse to optimize your hardware you can look into undervolting. Getting lower temps/noise/running cost. In case of the AMD RX (4|5)(7|8)0 even higher performance, since they can hold their boost because they don't hit their thermal threshhold. For laptops it prolongs battery life.

Not possible on Intel chips any more due to Plundervolt.

<edit>Thank You for bringing Plundervolt [0] to my attention</edit>

They removed the possibility to change voltage on my chip from their eXtreme Tuning Utility in an "update". Had to scour the web for an older version. <edit>Luckily i use Windows around 3% of the time.</edit> Won't buy Intel in the future.

[0] https://plundervolt.com/

I have my Ryzen 9 5950X overclocked to 4.1 GHz from a factory default of 3.4 GHz so I'd say those days are still here. Just needed an AIO liquid cooler.

I heard that AIO liquid coolers are not necessary too. I saw some YT videos saying that air coolers were equal or better in performance.

For those who really want to live on the edge and take risks, I heard liquid metal was the what all the "cool kids" were into these days. Just be warned though, like a cool tattoo, some people regretted that decision.

I saw those videos too... a decade ago. Things have really changed. I could cite some benchmarks showing that 2x120mm Watercoolers these days trade blows with high end air, and 2x140mm or 3x120mm usually outperforms air, but consider the following:

1: Air is essentially limited to using two smallish radiators and a couple of ~140mm fans in the airpath of eachother with limited placement options. Water-cooling radiators allow you to run obnoxious amounts of fans with big heavy radiators and very flexible placement options.

2: Cars used to be air cooled and companies eventually gave up and started using radiators because they simply dissipate heat better, despite being extra expense and another point of failure.

As component power draw is going up, and a1o radiators are getting economies of scale, water-cooling has become far more mainstream and easily dominates the >$100 cooling market. We're even seeing watercooling in the datacenter space, Google tensor chips are watercooled.

Updated with references.

What matters for thermal performance is thermal resistance at the cold plate. You'll find that the best air coolers end where mid-range AIO coolers begin in this metric.

The NH-D15 sits pretty at about 0.11 degC/W but the Arctic Freezer II line is all sub 0.10 degC/W.

Air coolers: https://images.anandtech.com/doci/16898/TRvsSPL.png

AIO coolers: https://images.anandtech.com/doci/16427/TRSP.png

How hot does that get at full load? I was worried about having so many cores at factory if not overvolted. For example, you can find plenty of people freaking out about 5800X temps. On the other hand, people claim the high temps are fine.

Just by enabling PBO my 5950x goes automatically to 4.2-4.3 GHz all-core and 4.8-5.0 GHz single-core.

They will never get close to the theoretical max performance. That's just bad business. You have to cut down the performance if you want to your market segmentation to survive.

Reply on news.ycombinator.com