Question AMD HD 4290 (on-board graphics) performance with different RAM speeds, UMA/Sideport

[mikeymikec]

I built a PC for a customer about ten years ago and recently upgraded it to Windows 10. Due to some stability issues that were caused by additional memory modules that I added at t he same time as the upgrade, and given that the customer needs this PC for his business so I'm sure that he would far prefer that no more faffing around with stability issues happens, I'm inclined to replace the 2x2GB modules in there with 2x4GB (rather than adding to what's there and potentially running into similar issues again).

The board in question is an ASUS M4A89GTD PRO/USB3 and he's using the on-board graphics (AMD HD 4290). His use for it is basically office stuff, but I'm mindful of how applications such as modern web browsers but also LibreOffice and Thunderbird will utilise graphics hardware acceleration techniques these days, so here are the options that I can think of:

1 - Get plain DDR3-1600 RAM. He's upgrading from DDR3-1333 so the graphics might get a bit of a performance boost.
2 - Get fancier DDR3-1600 RAM with tighter timings, so a greater performance boost?

The memory controller on the Phenom II supports a max of DDR3-1600 and going further (AFAIK) is OC territory, so I scrubbed DDR3-1866 from the list of options.

Also, the BIOS allows configuration of UMA, Sideport (on-board soldered dedicated 128MB DDR3-1333), and UMA+Sideport. Considering that GPUs love higher video RAM speeds, I'm wondering whether setting it to UMA only to take advantage of the faster RAM is better even with the trade-offs of having to go through the CPU's memory controller.

I'll set the UMA frame buffer size to the max of 512MB once the extra memory is in.

At the end of the day, the customer isn't complaining about possibly graphics-related performance issues, I'm just seeking max performance given the opportunity to make what could be a prudent investment.

Thoughts?

 

[Magic Carpet]

I’d get a cheap 95w Thuban (if you have less than 6 cores), one stick of 8gb ram and a discrete gpu, likely Kepler type, gt710. Windows 10 works great on a similar computer that I have (1035T, 8GB + GT710).

Thuban likes tighter timings, 1600 CL9 and below. Dual gives about 10% speed in some tasks. I run single for simplicity. Extra cores will give you the best performance upgrade.

 

[Arkaign]

Dang that's a lot of years for an AM3 era board. While you've got a convenient chance, eyeball the caps. I've had so many boards from that era whether S775 or AM2/3+ with the electrolytic caps start bulging and becoming unstable over the years. It sounds fine, but if you see any bulging caps or discoloration along them maybe source some replacements, theyr

 

[mikeymikec]

Dang that's a lot of years for an AM3 era board. While you've got a convenient chance, eyeball the caps. I've had so many boards from that era whether S775 or AM2/3+ with the electrolytic caps start bulging and becoming unstable over the years. It sounds fine, but if you see any bulging caps or discoloration along them maybe source some replacements, theyr

I'll bear that in mind certainly for another AM3 board belonging to a customer that's been ailing recently, thanks.

 

[mikeymikec]

With regard to the OP, I have a spare PC at some that's a virtually identical spec to the customer's PC (in particular, same board), and thought I'd put my theories to the test.

I used the Heaven 2009 benchmark in DX9 and OpenGL modes, textures set to low, res set to 1080p and 720p. Faster RAM with tighter timings (DDR3-1600 HyperX Fury) made zero difference to benchmark performance which was a surprise, and disabling the DDR3-1333 sideport when DDR3-1600 RAM was installed resulted in a slight performance reduction which was semi-surprising.

If anyone thinks of any other easy benchmarks I could try over this weekend, I'll do my best to give them a shot.

I assume that faster RAM with tighter timings will help a smidgen during general app usage, but based on what I've seen so far I won't be suggesting the upgrade to any other AM3-using customers I have.

 

[Magic Carpet]

The difference isn't huge to begin with. I'd just get more memory and call it a day. Try using Handbrake as a benchmark instead.

 

[VirtualLarry]

If they're still using an AMD/ATI chipset-IGP, best performance increase that you will actually notice, is moving up to a GT 710 / 730 / 1030, the latter-most I recommend at this point, because the 710 / 730 are now "legacy", the same as the HD 4290.

Edit: If their OS can be easily migrated, consider moving them to B560 / G6400 / 6405 / UHD 610 / DDR4-3200 ?

 

[mikeymikec]

If they're still using an AMD/ATI chipset-IGP, best performance increase that you will actually notice, is moving up to a GT 710 / 730 / 1030, the latter-most I recommend at this point, because the 710 / 730 are now "legacy", the same as the HD 4290.

Edit: If their OS can be easily migrated, consider moving them to B560 / G6400 / 6405 / UHD 610 / DDR4-3200 ?

They've just invested in an upgrade from 7 to 10 along with an SSD, so making the best use of what they have would be more logical at this point. This whole topic came up simply because upgrading from 4 to 8GB RAM was part of the original plan and since my initial use of additional modules didn't go to plan I'm replacing the lot.

 

[NTMBK]

The difference will be so minimal as to not be worth worrying about. Go with the normal 1600, don't worry about timings, and leave the sideport enabled. If the customer really cared about GPU performance, they would ask you to add a cheap modern GPU. If they are happy with the integrated graphics then it doesn't matter.

 

[mikeymikec]

Something funny I discovered - I switched back to the original memory (no frills 4GB DDR3-1333) in the test system, set the UMA back to 256MB (because losing half a gig of 4GB to graphics seems excessive), ran the Heaven benchmark again (720p, DX9, low q textures), and ended up with better results (a few fps) than with 512MB UMA!

Overall it gives me the impression that this benchmark / settings wasn't a great choice for my purpose.

 

[mikeymikec]

Addendum to my last post - I played around a bit more, using bog standard 1333 RAM this time. All the tests I ran were DX9, I tried 720p and 1080p, varying the texture quality on each run, and not once did increasing the UMA from 256 to 512 actually help performance (no UMA vs UMA 256: latter is better, to be expected in sanesville).

I suppose if I wanted to explicitly test whether the increased UMA will ever have any positive effect on graphics performance then I'd need a synthetic benchmark that theoretically does something non-GPU intensive but GPU memory intensive? I just wonder if there's a bug in the UMA implementation for this GPU.