How to get low 3D mark scores/many hours hwbot achivement?

[jimba]

So I am trying to attain these time intensive hwbot achievements  "Make 5 3DMark submissions with a very low score.", "Submit 5 Aquamark3 scores that lasted 6 hours to finish." and "Submit 5 SuperPI 32M runs which took over 12 hours to finish.".

I have got one of the 3D mark ones attributed to me already, but its a error on my part (I entered the wrong score) and I cant change myself.

For those who have done it how did you do it?

I guess I like pain.

[Macsbeach98]

Really old hardware think 386 or 486 really old graphics cards trident S3 or something like that.

[Aleslammer]

The 3D achievements can be done by slowing down your hardware, though it does help to use older platforms. Search AM3 in the data base using a max score of say <=500, don't check best, (forget the cut point for 6 hours). Look at the systems that are used in the list, last place of 2 more than 20 days as I remember, the 0 scores are wrapper problems. The 3D marks same basic idea you're trying to be slower than last place, though time not a factor based on how frame rates are calculated. Going to say a score of 250 or less for the early 3D Marks but don't take it as gospel. 

SP can be treated the same way, I've thought about doing this, as I remember SP32 won't go past 24 hours.

[mrpaco]

You have to scrape the bottom of the barrel

[MachineLearning]

As others have said, very old or very shit hardware.
For example: hwbot.org/submission/5235226_machinelearning_superpi___32m_eden_esp_5000_16h_15min_14sec_428ms

Luckily such systems should be relatively cheap. You can probably take a more modern platform and handicap it as much as possible. i.e. thermal throttling + single channel + using swap + horribly unoptomized OS etc. etc.

[austin86]

Here is my take, slow hardware, down clock as much as you can, get the lest ram you can put in the system, down clock it and jack up the timings, disabled l2/l3 and run prime95 and something like memtest g80 in the background.

[jimba]

May 15, 2023 12:23:57 GMT -5 austin86 said:

Here is my take, slow hardware, down clock as much as you can, get the lest ram you can put in the system, down clock it and jack up the timings, disabled l2/l3 and run prime95 and something like memtest g80 in the background.

So something like a single core AM3 or single core 1.5ghz 775 chip or a Pentium 4?

[MachineLearning]

If you want to get creative... Tax the CPU or GPU for a determined amount of time over 12 hours.

Example:
C2D E8500 running low priority sPi 32m, but you also run high priority Linpack Xtreme for enough loops to last >12 & <24 hours. So at the end of the Linpack loops, it flies through sPi and you get your result.
Do 3 loops Linpack with sPi 32m running to see how much it gets slowed down and calculate the required loops.
There are other ways like this, I'm sure. Just giving ideas how you can stage this.

[freeagent]

Bring it down to single channel, that should suck even harder

[antinomy]

For Super Pi - put it on a dead slow SD card and set USB in 1.1 mode. It hurts Pi performance.