The Serpent Rider wrote on Yesterday, 01:40:

AlexZ wrote on 2025-09-01, 15:45:

To improve write performance we would probably have to use 2 memory modules only to allow Command Rate 1T.

No. That's not how that works with DDR2. Command Rate 1T is generally not achievable on high memory clock rates. Plus it doesn't add much improvement.

Many of the tests were done at DDR2 800. Lowering frequency further to achieve CR1 is not meaningful, hence my suggestion to try 2 memory modules instead. If it doesn't work, no big deal.

The Serpent Rider wrote on Yesterday, 01:40:

L3 cache bandwidth scales much less with additional cores as it is shared.

L3 does not scale with cores at all. They all share the same pool of cache.

Also such L3 cache results for Phenom II are painfully slow. Even for default settings. And it's important to overclock NB frequency to get overall better results for memory and L3 cache.

Back then I had Phenom II X6 1100T in a running system and probably observed very slight improvement in L3 bandwidth. Your own screenshots shows it as well. The CPU is now sitting in shelf as it is not worth it.

I checked again my Phenom II X4 and L3 bandwidth bandwidth/latency does appear to scale a bit with CPU frequency. I would expect more cores to have a similar effect. Performance deteriorates more rapidly as we approach NB frequency.

I do not recommend OCing old systems in general to extend their life. I also do have Phenom II 965, 975 and Phenom II X6 1100T which could run in my AM2+ board but I decided not to.

AlexZ wrote on Yesterday, 20:50:

I checked again my Phenom II X4 and L3 bandwidth bandwidth/latency does appear to scale a bit with CPU frequency. I would expect more cores to have a similar effect. Performance deteriorates more rapidly as we approach NB frequency.

a) Windows 10 (I assume) does something horrible to Phenom II.
b) Gigabyte motherboard does something fishy with performance settings. Stock NB has over 40gb/s on read and copy with normal readings.

Many of the tests were done at DDR2 800. Lowering frequency further to achieve CR1 is not meaningful, hence my suggestion to try 2 memory modules instead.

Even 2 modules won't help. CR1 will kinda work if modules are single rank, but that's a loss of bank interleaving.

Back then I had Phenom II X6 1100T in a running system and probably observed very slight improvement in L3 bandwidth. Your own screenshots shows it as well.

That's mostly within a margin of error. In fact, Thuban has slightly slower access time.

Just wondering that are there that many XP games that don't work with a modern PC? I'd imagine most would work, at least with a virtual machine if necessary.

Mike_ wrote on Today, 07:31:

Just wondering that are there that many XP games that don't work with a modern PC?

its more like those were win95 games that worked up to XP, and most can be fixed by dropping directdraw shim library like
https://github.com/narzoul/DDrawCompat
https://github.com/DxWnd/DxWnd.reloaded
https://github.com/hifi/cnc-ddraw
https://github.com/jarikomppa/dxwrapper
https://github.com/ToadKing/ddhack
https://github.com/excitoon/ddraw
+ DirectSound3D/EAX being cut off with Vista, but there are patches/workarounds for this too

The Serpent Rider wrote on Today, 04:35:

AlexZ wrote on Yesterday, 20:50:

I checked again my Phenom II X4 and L3 bandwidth bandwidth/latency does appear to scale a bit with CPU frequency. I would expect more cores to have a similar effect. Performance deteriorates more rapidly as we approach NB frequency.

a) Windows 10 (I assume) does something horrible to Phenom II.
b) Gigabyte motherboard does something fishy with performance settings. Stock NB has over 40gb/s on read and copy with normal readings.

I used Windows XP. What version of AIDA64 did you use? I used 6.0. It is from 2019 so should be capable of accurate readings.

rasz_pl wrote on Today, 08:09:

its more like those were win95 games that worked up to XP, and most can be fixed by dropping directdraw shim library like https: […]
Show full quote

Mike_ wrote on Today, 07:31:

Just wondering that are there that many XP games that don't work with a modern PC?

its more like those were win95 games that worked up to XP, and most can be fixed by dropping directdraw shim library like
https://github.com/narzoul/DDrawCompat
https://github.com/DxWnd/DxWnd.reloaded
https://github.com/hifi/cnc-ddraw
https://github.com/jarikomppa/dxwrapper
https://github.com/ToadKing/ddhack
https://github.com/excitoon/ddraw
+ DirectSound3D/EAX being cut off with Vista, but there are patches/workarounds for this too

I see, but this makes me wonder even more that what's the point of building an XP retro computer apart from just tinkering with old-ish hardware. I mean, hardware from that era is already quite modern, it's usually 64bit even if the OS isn't and mostly uses PCIe like modern PCs. There are no weird quirks unlike with DOS.

Mike_ wrote on Today, 09:10:

I see, but this makes me wonder even more that what's the point of building an XP retro computer apart from just tinkering with old-ish hardware.

I've mentioned this elsewhere, but a WinXP rig can be nice if you own a bunch of physical games from that era. All the weird DRM crap works on WinXP while being (rightfully so) disabled on modern Windows due to security concerns. Obviously, this can be circumvented by using fan made fixes and such. But with WinXP, you just pop in a disc, install the game, apply the latest official patch (if you feel so inclined) and play. No need to mess around with wrappers or extra fixes, everything just works.

Basically, WinXP simplifies things, allowing you to play games without spending a bunch of time researching how to get them running on modern Windows. To be fair, GOG and (to a lesser extent) Steam now include many of those fan made fixes in their releases of older games. But if you have an actual physical copy, you often need to fix it up by yourself if you want to run it on modern Windows.

I didn't realize there were such problems with Safedisc and other DRM crap, that explains it.