I found out what the problem was with the Radeon HD 5750. Somehow some case cables got into the fan, and stopped it from moving. With that out of the way, benchmarking can commence. Except that I don’t have much time for this right now.
I’ve tried doing some things, and it’s a struggle. For one thing, I can’t seem to be able to undervolt or underclock the CPU. Neither Ryzen Master nor the BIOS options ended up doing anything.Well, not completely nothing, the voltage did seem to take more time to read 1.4V and system power when running Prime95 was around 120W instead of 127W. But underclocking didn’t seem to do a thing and voltage tweaking didn’t seem to do any more than that. It’s not that the system was unstable or stopped working, it simply didn’t do a thing.
GPU undervolting did work, although only when I dropped the SOC voltage. But that made quite an obvious change, from around 88W in FurMark at 1.1V to around 82W at 1V. Still, without the ability of saving power by tweaking the CPU I felt that it was easiest to just use the 35W BIOS option.
Given that I had already tested that, I figured I’d move from power testing to something that was on my original test plan: comparing the integrate Vega 8 to the Radeon HD 5750 in my Phenom II PC.
For a start, I discovered that it was possible to change the location where Afterburner displays its text, by running RivaTuner Statistics Server. That helped display CPU and GPU stats without them overlaying the benchmark provided stats.
Unfortunately the GPU percentage showed garbage (a very large number) which overlayed other GPU stats. I tried installing the Raven Ridge drivers from AMD’s site, but that didn’t help, so I returned to the Microsoft supplied driver, which is newer.
Making the power meter more visible took a bit of effort. After several attempts I ended up propping a torch (flashlight) directed at it from the side, and that seemed to highlight the display better.
Syncing and overlaying 4 videos was more difficult than the two I had previously done, and it also turned out that cropping the start of the video after Track Motion is set resets it, which was annoying. The end result still isn’t perfectly frame aligned, but it’s close enough.
I feel bad calling this series ‘The Minimal 2200G Build’, now that I’ve added another 4GB RAM, but I’ll continue with the name for consistency’s stake. 🙂 I do plan to test the 4GB configuration.
My first video is up. It’s a test for creating videos, so isn’t all I wanted to show, but any start is a good start. It shows the system running the Metro: Last Light Redux benchmark at default (low) settings at 1080p, with system power shown as measured by my cheapowatt (that’s my power meter’s nickname).
For those who are interested in making videos, or are interested in telling me how to make them better, here’s how this one was made.
Yeah, we all know that. I installed the new 4GB stick and ran Metro, FurMark and Prime95. Both Metro and FurMark ran about twice as fast, but they also took more power. Metro ran about 95-105W, vs. 80-90W with 4GB.
FurMark + Prime95 started at 125W but then dropped to 105W, same as the 4GB power consumption, and stayed there. FurMark went up from around 12 FPS to about the full speed of 23 FPS, stayed there a while and then dropped back. Now it’s running at 19 FPS. I’m not sure what the deal is there. I can imagine that the drop in power has to do with the CPU getting throttled (not sure why it took time), but the FPS switch is stranger.
I installed Ryzen Master and saw that the CPU cores are running about 3100Mhz (+-25MHz). Ryzen Master claimed that the CPU was taking around 20W. It then crossed my mind that perhaps the power options in the BIOS simply didn’t kick in because power wasn’t high enough with only 4GB of RAM.
So I went to the BIOS and switched the power from 45W to Auto. Power when running FurMark + Prime95 jumped to 135-140W. FurMark didn’t go up in frame rate. In fact, it continues its strange behaviour of jumping between frame rates and staying there fore a while. GPU clock is significantly (around 150MHz) higher, but it makes no difference. I’m assuming that RAM is the bottleneck here. If that’s the case, it means that power use could be improved by dropping the GPU clock when it’s bottlenecked anyway.
Just a little update. I haven’t spent that much time working on the system or the posts, but figured it’s worth making sure you know that I’m not dead yet, and neither is this series of posts.
According to DHL, the A8-9600 should have reached AMD on the 29th of March. But on the 3rd of April I got an e-mail warning me that if the CPU wasn’t returned within 15 days I might lose the privilege of future RMAs. Made me wonder if that penalty is worth it, and I should have kept the A8-9600. Wouldn’t have done it, but considering I’ve never had a CPU fail, there’s some temptation there.
On the 4th of April I got a message that AMD has received the CPU.
I did some testing, running Metro: Last Light Redux, FurMark and Prime95. Metro: Last Light Redux is a much better stress test than the games I ran on the A8-9600, taking 80-90W of system power at the wall, not 60-70W like the rest. Made me even sorrier I forgot to test the A8-9600 with it.
Prime95 took system power was 100W, and Prime95 + FurMark 105W (which cut FurMark’s frame rate in half). I forgot by now how much FurMark took alone. Will retest. Oh, and the system was stable throughout the testing. I’m not using it for everyday use, so can tell if everything else works fine, but for a few hours of testing it was fine.
I tried playing with the power limit option in the BIOS (System Configuration, under Advanced / AMD CBS / NBIO Common Options). There are options for 35W, 45W and 65W, but they did nothing, unfortunately. I’m disappointed, because low power use is something I’m interested in and I want to see how it affects performance. I guess I’ll have to install Ryzen Master and try to achieve results with it.
I said before that the 2200G ran fine with the 18.2.1 drivers installed with the A8-9600, except that MSI Afterburner didn’t show GPU utilisation. When I later booted the PC and ran some tests, I found out that Windows installed a newer Raven Ridge specific driver, 17.40.3735, which is newer than the one AMD has on its site (17.40.3701).
These are all the updates I can think of for now. Well, except that I ordered another 4GB DIMM from Amazon, this time a white one (it was cheaper than the red). Got it today.
I added some images to the BIOS Blues post. At first I didn’t remember how to add custom HTML, which made arranging the images hard (I wanted a table). So I glued them side by side in Paint, and that seemed to produce the effect I wanted.
Once I found the HTML tab (hidden in plain site), I decided I was too lazy to try to change what already looks okay, so left it that way.
Many of the images I have are low quality anyway. I took them with my Galaxy 4 Mini phone, which does a bad job in low light conditions.
I will add images to the It’s Aliiiive post later, and will try to use custom HTML if I think it’s necessary.