Sooo, I'm working on putting together the 775 system I started quite a while ago. The motherboard is an ASUS P5e3 Premium.

Was just watching a YouTube video and apparently this board can do around 705fsb on air!?!?!?!?!

Have an idea of what CPU is in it, but not going to mention anything until I get it going and do a bit of testing.

Have a bunch of high speed DDR3 as I have been collecting it for a while now and snatch it up when I see it for a good price.

cyclone3d wrote on 2025-03-21, 19:28:

True, but what about background processes? You still have other things going on that can be offloaded to other cores. […]
Show full quote

True, but what about background processes? You still have other things going on that can be offloaded to other cores.

May not really matter as long as you have a dual core. The E8600 also has more cache and higher IPC.

I would put the E8500/8600 up against the Q/QX9xxx series and test them at the same clock speeds.

I can test this and post numbers.

Since we are going for absolute performance, I'll test on a DDR3 based board as well so I can push the FSB.

If you want testing on a DDR2 based board, I can do that as well.

Gimme a list of CPUs and a list of games.

What about the GPUs?

Currently I'm relying to Anandtech legacy benchmarks results with Cinebench R10, and comparing their numbers to mine.
My Q6600 easily beats a stock Q6600, but in single threaded performances an E8600 is a whole other level.

But I don't knwo about DDR2 and DDR3. I always thought DDR3 was more of a Vista/7 ram rather than XP. And any latter era XP game can be ran on Windows 7 without issues, so any older game that would require XP for best compatibility should be able to run fast even on DDR2.

ElectroSoldier wrote on 2025-03-21, 18:59:

Most XP games were single threaded so I cant see a Q6600 having any more performance in game than an E8600.

Yup, Benchmarks are also of that opinion when looking at single threaded performances.
Even a Q9650 will get its ass beaten in single thread benchmarks by an E8600.

Then there's also overclocking on top of that. I know I can push the FSB up to 400mhz without having to touch anything else in the system. I don't know how high it could go, but the last strap I have is 400/800. So after that is a whole lots of math on top of stability to figure out ratios and actually working FSB frequencies.

cyclone3d wrote on 2025-03-21, 19:28:

True, but what about background processes? You still have other things going on that can be offloaded to other cores. […]
Show full quote

True, but what about background processes? You still have other things going on that can be offloaded to other cores.

May not really matter as long as you have a dual core. The E8600 also has more cache and higher IPC.

I would put the E8500/8600 up against the Q/QX9xxx series and test them at the same clock speeds.

I can test this and post numbers.

Since we are going for absolute performance, I'll test on a DDR3 based board as well so I can push the FSB.

If you want testing on a DDR2 based board, I can do that as well.

Gimme a list of CPUs and a list of games.

What about the GPUs?

Would be interesting to see how they actually stack to each other.

Common GPU for me. as its about the CPU.

For me when you look at the thread title. Then the first CPU that comes to mind is the E8600 because of the single threaded performance.

Well, the CPU I thought was in this board is not. How about I start out with the fasted Pentium 4 based CPU, the Pentium-D 965 Extreme Edition?

Was having trouble getting it to POST and found out that it doesn't like 8GB DIMMs. Understandable as the official max RAM is 8GB total. It does work with 4GB DIMMs though so I will go with that.

Mondodimotori wrote on 2025-03-21, 11:27:

And here's another big doubt about 775 platforms: I'm currently running 333mhz FSB on the Q6600, no overvolt and nicely stable. […]
Show full quote

agent_x007 wrote on 2025-03-20, 20:39:

Note : You can have multiple straps get to the same FSB/MEM speeds (I think 266 and 400 straps are the same for 1:1 RAM with 1600FSB/800 DRAM).

And here's another big doubt about 775 platforms: I'm currently running 333mhz FSB on the Q6600, no overvolt and nicely stable. I've set the ram (800mhz Corsair kit that can OC up to 2.1V 4-4-4-12) to 667mhz with 4-4-4-12 timings at 1.9volt.

So I have a 1:1 ratio , using the 333/667 strap, but... My mobo also has a strap for 333FSB with 800mhz ram.
Are we sure it's faster to use the 333/667 strap instead of the 333/800 one? Considering that I would OC the ram to 2.1 volt and keep the rated 4-4-4-12 timings.

I haven't seen many differences in Aida64. I actually saw worse latency with 333/667 instead of 333/800.

Strap for "333/667" and "333/800" is the same (you just drop DRAM clock).
Strap options (in ASUS boards), and DRAM Frequencies associated with them [FSB = 333/1333 (real/effective)] :
200 :

266 :

333 :

400 :

Also, something to think about with 1:1 being "fastest" strap 😉

(lowest strap = fastest strap, going 200 is hard on Pxx chipsets though - stick to 266 and 333 if you can)
Q : Can you write/show what memory latency AIDA64 shown with your settings ?

EDIT @cyclone3d : To get 4x 2GB/2x 4GB/4x 4GB going, you will need :
1) Higher NB voltage (~1.35V),
2) Manual adjustment of tRFC (based on XMP profile for your memory).

The E8600 is probably the best cpu for soc775 gaming that uses single core engines as it hits an 8% increase over a Q9650 in Passmark Single thread tests (not 10% due to smaller cache and other stuff).
If you plan on doing anything else then the more cores the better (example: I use my Q9650 XP box for games and video re-encoding and is about 50% time versus the E8500 it replaced using Xmedia Recode if you set the threads in preferences)
fwiw: E8500 is only 4% less than the E8600 Passmark in single thread, not much difference....sure some may say Passmark is not a good test but it does give a very good indicator of overall performance.
Both the E8600 and the Q9650 can be found very inexpensive right now.

The vast majority of XP era games that support multi core capped out at dual cores. And one that used more threads were pretty unbalanced where yes cores 3 and 4 ran something. But they were mostly idle so the extra cores didn't help versus a dual core that could sustain a higher clock.

Video encoding on such an old 97W CPU is an interesting choice. A 6W Intel N100 is more than 2x the raw CPU performance but also contains an outstanding media engine in its iGPU which Xmedia Recode can take advantage of. But hey, if you get free electricity, awesome.

Intel core 2 Quad Extreme
4gb of RAM or More
Sata SSD

SScorpio wrote on 2025-03-23, 14:29:

The vast majority of XP era games that support multi core capped out at dual cores. And one that used more threads were pretty unbalanced where yes cores 3 and 4 ran something. But they were mostly idle so the extra cores didn't help versus a dual core that could sustain a higher clock.

That's been my experience as well... at least with the games *I* play (which is a pretty short list, TBH.)
So yeah, for XP era games, I think a fast dual core would be better.
If doing something else on the PC (i.e. browsing the internet, video encoding/decoding and or using some other legacy software that's known to use more CPU cores), then go with more cores.

SScorpio wrote on 2025-03-23, 14:29:

But hey, if you get free electricity, awesome.

No free electricity here... but I do bring the "big rigs" out in the colder months of the year to help with keeping my room warm.
So I'm not really "wasting" electricity in that regard.

And in the summer I do the opposite - I bring out the cooler and more power-efficient stuff out so that I'm not sweating like a pig when I'm using my PC. I don't have A/C here, so that's another reason to bring out the cooler PCs. The "hot" ones, aside from warming up my place too much, also may reach quite high running temperatures, and that's just no good for the hardware in the long term. For the ones that I've built to cope with the heat, they can get quite loud... which too is an annoyance. So I rather avoid all of that and use my PC according to the weather/temperatures of the year. 😁

momaka wrote on 2025-03-23, 14:46:

And in the summer I do the opposite - I bring out the cooler and more power-efficient stuff out so that I'm not sweating like a pig when I'm using my PC. I don't have A/C here, so that's another reason to bring out the cooler PCs. The "hot" ones, aside from warming up my place too much, also may reach quite high running temperatures, and that's just no good for the hardware in the long term. For the ones that I've built to cope with the heat, they can get quite loud... which too is an annoyance. So I rather avoid all of that and use my PC according to the weather/temperatures of the year. 😁

It sounds like there could be a market for WiFi enabled crypto space heaters. If you're paying for the heat, might as well try to break even or strike it rich.

agent_x007 wrote on 2025-03-22, 22:46:

Q : Can you write/show what memory latency AIDA64 shown with your settings ?

Sure, I'm not that versatile in Straps and stuff. In the system thread I state that this is the PC I wished I had in 2008, but couldn't afford. I ended up getting a core2 cheap laptop in 2009 that kept me up until december 2013, when I got the i7 laptop I'm still using. Desktop I wento from a Socket A system in 2002 to an AM4 system in early 2020. So I have very little experinece on desktop platforms from the mid to late 2000s.

Here's the latency with the 333 strap, 667mhz ram at 4-4-4-12, 1.9 volt

If I OC the ram to the vendor suggestion (2.1 volt, 4-4-4-12 at 800mhz), if I recall correctly, I get latency in the low 70 ns. But I loose a bit of speed in all the other benchmarks (reading/writings and copy).
The 333 strap also has 800mhz for ram. The 400 strap only has 800 for ram.

SScorpio wrote on 2025-03-23, 14:29:

Video encoding on such an old 97W CPU is an interesting choice. A 6W Intel N100 is more than 2x the raw CPU performance but also contains an outstanding media engine in its iGPU which Xmedia Recode can take advantage of. But hey, if you get free electricity, awesome.

There's no reason to get an N100 when Ryzens exist.

Horun wrote on 2025-03-23, 01:56:

The E8600 is probably the best cpu for soc775 gaming that uses single core engines as it hits an 8% increase over a Q9650 in Pas […]
Show full quote

The E8600 is probably the best cpu for soc775 gaming that uses single core engines as it hits an 8% increase over a Q9650 in Passmark Single thread tests (not 10% due to smaller cache and other stuff).
If you plan on doing anything else then the more cores the better (example: I use my Q9650 XP box for games and video re-encoding and is about 50% time versus the E8500 it replaced using Xmedia Recode if you set the threads in preferences)
fwiw: E8500 is only 4% less than the E8600 Passmark in single thread, not much difference....sure some may say Passmark is not a good test but it does give a very good indicator of overall performance.
Both the E8600 and the Q9650 can be found very inexpensive right now.

Yeah, I'm only interested in games. Production I'm a masochist, because I wanna install Win 2000 on my Socket A system and see what an Athlon 1.4 can do with AutoCAD R14 and Photoshop, trying and opening drawing files that will bring even a new i7 to its knees.
But Windows XP? This system is there for pure gaming. I may use the planned Win7 box for more production, but I also have a ready Win10 box on AM4. That ryzen 3600 will still steam through everything I throw at him.

True that Q9650 and, especially, the E8600, can be found for relatively cheap. That's why I was asking. My dream build in 2008 had listed a Q9650 but, after playing games with the Q6600, I started wondering if I'm underestimating how CPU limited were the games of the times. We see lots of grifters on the internet complaining about "lazy modern devs that won't optimize their games", and then I found out that Most Wanted 2005, an extremely reveered game, appears to be CPU limited on a OCd Q6600. Even COD4, that should run flawlessly on a HD6970, at 1024*768 2x AA, it can't keep up the 85fps to match the monitor's refresh rate. It can't be the GPU, so it must either be the CPU or the RAM. Not sure it's the RAM. (but even on my Athlon and Pentium III system I found several instances of CPU limitation in games that, in recommended requirements, claimed to run on a 400mhz CPU. Maybe they meant "recommended for booting the game")

SScorpio wrote on 2025-03-23, 14:29:

The vast majority of XP era games that support multi core capped out at dual cores. And one that used more threads were pretty unbalanced where yes cores 3 and 4 ran something. But they were mostly idle so the extra cores didn't help versus a dual core that could sustain a higher clock.

Video encoding on such an old 97W CPU is an interesting choice. A 6W Intel N100 is more than 2x the raw CPU performance but also contains an outstanding media engine in its iGPU which Xmedia Recode can take advantage of. But hey, if you get free electricity, awesome.

This is another thing. Core parking. I've read someone talking about "background tasks", but will the OS park those background tasks to the least used cores when playing a game?

Mondodimotori wrote on 2025-03-23, 20:12:

Sure, I'm not that versatile in Straps and stuff. In the system thread I state that this is the PC I wished I had in 2008, but c […]
Show full quote

Sure, I'm not that versatile in Straps and stuff. In the system thread I state that this is the PC I wished I had in 2008, but couldn't afford. I ended up getting a core2 cheap laptop in 2009 that kept me up until december 2013, when I got the i7 laptop I'm still using. Desktop I wento from a Socket A system in 2002 to an AM4 system in early 2020. So I have very little experinece on desktop platforms from the mid to late 2000s.

Here's the latency with the 333 strap, 667mhz ram at 4-4-4-12, 1.9 volt

The attachment ramlatency.jpg is no longer available

If I OC the ram to the vendor suggestion (2.1 volt, 4-4-4-12 at 800mhz), if I recall correctly, I get latency in the low 70 ns. But I loose a bit of speed in all the other benchmarks (reading/writings and copy).
The 333 strap also has 800mhz for ram. The 400 strap only has 800 for ram.

For ease of use, use this option :

^You can click start to test everything, or double click on each option to test only it.

Not tweaking straps will leave some performance on table, but shouldn't matter too much (depends on FPS, higher FPS = more memory intensive). Personally, I'd try 266 strap with "833" DRAM speed on (1)333 FSB.

I did not have any issues (with my e8500 or the q9650) with Call of Duty 4 MW or WAW which recommend a dual core 2.4ghz or better, those games do utilize dual cores.
My only downfall is the nV 8800 GT but have kept it since all the parts are from same 2007-2008 era, there are better vid cards...

GemCookie wrote on 2025-03-23, 20:20:

There's no reason to get an N100 when Ryzens exist.

For general purpose? Sure. But AMD is a very distant third when it comes to media engines. Intel comes out on top for inexpensive and extremely lower power draw, NVIDIA matches it in capabilities, but requires discrete cards and uses much more power. And AMD doesn't have feature parity with some things just being broken. AMD is a little better in Windows, but their Linux media support has been lacking for years.

Mondodimotori wrote on 2025-03-23, 20:23:

This is another thing. Core parking. I've read someone talking about "background tasks", but will the OS park those background tasks to the least used cores when playing a game?

I don't believe it does by default. But you can mess with core assignments to make that happen. The XP scheduler was very early days and basic.

I used to have it worked out as far as latency goes where I could calculate what the latency would be at different cas settings vs mhz.

I've got XP installed on my test system. Just have to get the rest of the drivers installed before I can start doing testing.

agent_x007 wrote on 2025-03-23, 20:35:

For ease of use, use this option :

I'm currently running an older version of AIDA64, the latest one will crash the OCd system when opening it. Sometimes it crashes the system even without OC.
I don't remember if my version had that option.

agent_x007 wrote on 2025-03-23, 20:35:

Not tweaking straps will leave some performance on table, but shouldn't matter too much (depends on FPS, higher FPS = more memory intensive). Personally, I'd try 266 strap with "833" DRAM speed on (1)333 FSB.

I'm not reallt that interested in optimal performances, squeezing every last drop of frequency. It's always 15+ years old hardware, better not stress it too much.

Horun wrote on 2025-03-23, 22:05:

I did not have any issues (with my e8500 or the q9650) with Call of Duty 4 MW or WAW which recommend a dual core 2.4ghz or better, those games do utilize dual cores.
My only downfall is the nV 8800 GT but have kept it since all the parts are from same 2007-2008 era, there are better vid cards...

I've got myself a much faster GPU. The XP system came with an HD4850, wich was decent enough at 1024x768, but I wanted something similar to a 560ti. I ended up finding an auction for a Sapphire HD 6970, and snatching it for 21€ shipping included. Now I can easily max out games up to 2008 even at 1280x1024, and even add a sprinkle of anti aliasing. Unfortunately some higher details settings for objects and rendering distances will still hamper performances, and it's not the GPU bottlenecking, because I can see it's usage dropping to less than 80%. I guessed it must be a CPU bottleneck. After all those recommended settings weren't there for rocksolid 60 fps at max settings. It's not like today where, if you meet recommended settings, 60 fps at 1080p and higher are a no brainer.

And that's why I wanna max out the gaming potential of the CPU too. Then I can OC it to a 400MHz FSB and run it 1:1 with the 800MHz Corsair ram kit I got. It's a pretty tight ram kit.

Mondodimotori wrote on 2025-03-24, 19:53:

I've got myself a much faster GPU. The XP system came with an HD4850, wich was decent enough at 1024x768, but I wanted something similar to a 560ti. I ended up finding an auction for a Sapphire HD 6970, and snatching it for 21€ shipping included. Now I can easily max out games up to 2008 even at 1280x1024, and even add a sprinkle of anti aliasing. Unfortunately some higher details settings for objects and rendering distances will still hamper performances, and it's not the GPU bottlenecking, because I can see it's usage dropping to less than 80%. I guessed it must be a CPU bottleneck. After all those recommended settings weren't there for rocksolid 60 fps at max settings. It's not like today where, if you meet recommended settings, 60 fps at 1080p and higher are a no brainer.

And that's why I wanna max out the gaming potential of the CPU too. Then I can OC it to a 400MHz FSB and run it 1:1 with the 800MHz Corsair ram kit I got. It's a pretty tight ram kit.

Nehalem was out for two years by the time that GPU was released. And Sandybridge was release 2.5 months after that GPU. 2008 games were also a 1 to 1.5 years into Vista, with 7 coming out the next year. LGA 775 can make a good XP machine. But if you want to max games from Vista, 3-4 year old hardware was no longer top of the line. It's not uncommon for games to release where even the best CPU and GPU can't max it. But two years later new hardware is out that finally lets you hit 60fps at maxed settings.