PSA: Clean your blocks

[Zerileous]

Well everyone, I'm a bit embarrassed to say that I let loop maintenance get away from me. I was running some TimeSpy Extreme for grins and I noticed my CPU hitting 100C. Now I always knew (or thought) this CPU to run a little hot, but that was a bit of a shock. So I tore down and at first the loop looked pretty clean. Some cloudy coolant as I pulled tubes etc, but overall not bad condition. Then I found some gunk in one of the fittings. I tore open the CPU waterblock and was appalled to find a gritty and sticky substance blocking all of my pressure plate (sorry in my haste to eradicate this slime, I forgot to take pics). I'm wondering if it wasn't some plasticizer, there also seemed to be some metallic particles.

I'm honestly guessing some of this was from when I initially built the loop, since I never had the best CPU temps (I always blamed the 300W GPU sharing the loop, even though physics doesn't work that way). In April when I upgraded to a 3700x, I was not thrilled with the temps, I should have torn down the block then, but I never suspected since I cleaned everything (I thought) really well and I only use cryofuel clear coolant.

I also of course had some moderate dust build up which I'm sure wasn't helping, but I think the biggest thing was the restriction in the loop (in terms of overall performance) and of course the block itself. I am also running (stock) prime95 smallFFT at 76C. That's a substantial improvement from the upper 80s when I first installed the CPU.

TimeSpy saw 73C on the CPU, also the GPU hostpsot temp fell from 90c to 79c. There is a small amount of the CRUD in the GPU block but I didn't want to tear it down, since I don't have extra thermal pads, it doesn't seem to be severely obstructing flow, and I am hoping to replace with a 3800 or 6800XT within 6 months (so I can clean it then).

This improved cooling caused a massive improvement to my TimeSpy Extreme score from 3029 to 3462 (for anyone trying to compare, it is not a valid score, I did not bother to disable the driver modified tessellation or freesync). That's 14% for the cost of a few hours work! All of this with the fans barely breaking 1000rpm, when before they would often max out under a heavy game. I've tried all sorts of things, adding radiators, adding fans, playing with airflow, and all it took as a screwdriver and some crest 3DWite.

 

[RJARRRPCGP]

You're lucky to get good RAM speeds with all 4 slots! First off TMK, the Ryzen 7s don't support quad channel. TMK, you need a Ryzen Threadripper or an Epyc for that.

For goodness sakes, it looks like we can easily get 2x16 GB DDR4 sticks now.

And with my Ryzen 7 3700X, the temps seem fine with just an air cooler.

 

[Zerileous]

Not sure what TMK means, however I began my system with a 16GB kit but needed more for DCS world, especially multiplayer or VR. I never had any issues going to 4 sticks (even with a 2600X), but I believe that may have to do with the board using T-topology instead of daisy chain. Also the fact that it is B-die rated for 3200C14 does not hurt.

 

[RJARRRPCGP]

TMK=To my knowledge

 

[Ben333]

I'm pretty sure that even Ryzen 5 \ 7 chips can benefit from four sticks vs two. Someone please correct me if wrong. It isn't the difference of dual-channel vs quad, but there is a term for it, and it offers measurable improvement.

Here is an article explaining it in some capacity: https://www.techspot.com/article/2140-ryzen-5000-memory-performance/

I don't know which generations of CPUs this applies to, but should be helpful for the benchmark results.

 

[Janus67]

I'm pretty sure that even Ryzen 5 \ 7 chips can benefit from four sticks vs two. Someone please correct me if wrong. It isn't the difference of dual-channel vs quad, but there is a term for it, and it offers measurable improvement.

Here is an article explaining it in some capacity: https://www.techspot.com/article/2140-ryzen-5000-memory-performance/

I don't know which generations of CPUs this applies to, but should be helpful for the benchmark results.

It's about ranks, not channels. And is dependent on the work load as to if it is a noticeable improvement or not.

 

[Zerileous]

Yes but I believe that has only been noted in Ryzen 5000 / Zen 3.

 

[Woomack]

It was here for years but improvements depend on many variables like for example if software can use it. It's single or dual rank memory so nothing new and also memory controller in Ryzen 5000 hasn't really changed. Suddenly people make a lot of noise about it because someone pointed that out in the web. Maybe it's because barely anyone was testing memory before and how it affects the performance. Or maybe because all seen only 8GB Samsung B modules as every forum is flooded in why all should buy only Samsung B. It's like barely anyone test anything but all repost things they see in the web. Also, most these, let's say, respected websites make pretty bad memory tests and reviews.
I already mentioned in another thread that in my tests it's more like 0-5% improvement, depends on the test/game, display resolution etc. Of course I couldn't test everything so the improvement can be higher. In games it helps much more at low display resolution but it's still not something worth spending 2x more money for RAM capacity that we won't ever use.
In short, 2x16GB or 4x8GB seems optimal nowadays if anyone invests in a new gaming PC but if it's a budget build then 2x8GB is more than enough.