A few finishing touches for AGP system:

- Printed one of those universal io shields. Probably functionally irrelevant, but looks better than no shield at all:

- Managed to find working SF-2281 SSD, switched it to SATA1/SATA150 mode and was able to get rid of IDE-SATA adapter:

It is a shame i do not have another CPU with 2x1MB cache, but it is what it is, this system turned out pretty good... not sure if i am going to use it though. I'll try to see how well HD3850 compares to 8800GT in terms of compatibility and performance in relevant stuff and then... probably will just store this system as a spare.

Repo Man11 wrote on 2025-09-01, 19:26:

I just dusted of my Asus A8V-XE (K8T890) for some testing. With a Silicon Power A55 SSD, a 4200+ etc. I get about 130 MB/S write and 160 MB/s read as measured with ATTO, no issues at all.

I switched to the Opteron 180, the same memory, and a GTS250. Similar to the A8V-X, this board has no options for increasing the voltage on the memory or the CPU, but I was able to push the FSB to 218 with no issue. With the increased FSB the drive speed increased to 170 write and 138 read as measure by CrystalDiskMark. 3d 2001 score is 34,648, 3d 03 is 37,446. Next up will be the A8N-SLI.

Final report on the A8V-XE: I thought I hit the limit with 218 FSB on this board, but then I realized I was using a 450 watt PSU and that could be the limiting factor. So I swapped it for the EVGA 850 I had been using with the A8V-X and was then able to hit 225 FSB as I had on the A8V-X. This bumped the 3D 2001 score to around 36,000 and the 3d 03 score to about 38,000. This also bumped the read/write speeds to 175/140. I tried the Patriot Burst SSD that refused to work with the A8V-X and it worked fine on the A8V-XE though the write speed was a bit lower than with the Silicon Power SSD.

I always use 800W PSUs in my retro builds. Athlon 64, Phenom II and Vishera FX-8370 all use 800W PSU to avoid issues.

Also interesting that it has such a low limit. Gigabyte GA-K8NE (nforce4) was able to run at like 250, same my AM2+ board Gigabyte GA-MA770-UD3. I never tested the upper limit.

Back in 2008, I added an HIS HD 4670 IceQ 1GB DDR3 PCIe video card to my my Opteron 175. It was a delightful upgrade from the x800xl that I bought with the system originally. The 4670 was selling for $65 at the time, was whisper quiet, needed no external power, and handled left 4 dead, counter strike, defense grid, etc like a champ at 1680x1050 resolution. Then Borderlands came out in late 2009 and I needed a new everything, but until then, that was the card that gave my Athlon 64 extra life without requiring a monster power supply. I guess the 10K sata drive helped too, but you know what I mean.

I picked up a used HIS HD 4670 IceQ a couple years ago just because of how much I liked that card. Ended up spending a couple hours pulling hair out of the fan as a labor of love and I put it in an Athlon x2 build.

I imagine it would be pretty cool to find a second and see if I can do a cross fire config, but they might be getting scarce.

AlexZ wrote on Yesterday, 20:26:

I always use 800W PSUs in my retro builds. Athlon 64, Phenom II and Vishera FX-8370 all use 800W PSU to avoid issues.

Also interesting that it has such a low limit. Gigabyte GA-K8NE (nforce4) was able to run at like 250, same my AM2+ board Gigabyte GA-MA770-UD3. I never tested the upper limit.

As previously mentioned, this is with default voltage for both the CPU and the memory. I'm impressed by a 300 MHz overclock at stock core voltage.

Well, that sounds like PSU was not great if it affected overclocking. It really should not. Probably either old and partly faulty, or simply no good to begin with.

Most of the time "quality" of output (filtering, regulation, etc) does not depend on rated power and PSUs operate most efficiently at ~40-60% load, so too much overkill is not really beneficial.

That said finding good quality lower rated power supplies nowadays is not easy, so i generally stick to ~500-600W stuff which still exist.

Meanwhile i've found how to change multiplier and CPU voltage on ASUS A8N32-SLI Deluxe... exploring this BIOS feels like playing one of those old adventure games - cool &quiet has to be disabled, which unlocks an option in completely different place which, when enabled, shows multiplier and Vcore controls which are otherwise completely hidden (not locked -hidden). What confused me before is that there is separate option to bump Vcore by 0.2V which is always accessible - i though that's the only thing which was possible...

Dropped Vcore by 0.1V, left cool&quiet off, this resulted in lower temperatures than with cool&quiet. By default Vcore is slightly higher than it should be on this board - ~1.4V instead of 1.35. Sensor is completely unreliable - jumps all over the place by like 0.2V, ultimately had to use multimeter to confirm it is what it is set to and jumping is just the sensor - not actual voltage. And it is.

Also cut RPM sensor wire for zalman fan i am using for CPU cooler - it reports wildly varying inaccurate results (verified on different boards too), which sometimes causes "CPU fan error" message during POST.

Installed linux mint on the second SSD on this system in order to do TRIM, do backups without having to boot from thumb drive and transfer data over network. Second NIC this board has (pci-e marvel one) turned out to be quite handy - no need to mess with buggy nvidia one and its drivers. It works surprisingly well - apart from simple tasks i've mentioned it is possible to actually use firefox and browse the web including videos and such quite comfortably. This CPU being able to decode 1080P youtube with no hardware acceleration actually surprised me - many low end modern-ish systems struggle with this. 2GB of memory are not much, but SSD swap + vm.swappiness=99 along with memory management modern linux has work surprisingly well - 20-30 browser tabs before things start visibly slowing down. It is curious to see such an old system being able to perform modern tasks competently.

SSDs only work at sata1/sata150 speeds, but NCQ actually works and performance is quite good - i do not think it will be worth messing around to make it sata2 - will likely make no difference.

It was a Corsair CX 450M power supply - not great, not terrible, fine for stock speeds but a 450 watt PSU is the minimum recommended for a GTS250. The Opteron 180 has a 110 watt TDP and I would imagine that rises quickly when overclocking.

Repo Man11 wrote on Yesterday, 22:45:

It was a Corsair CX 450M power supply - not great, not terrible, fine for stock speeds but a 450 watt PSU is the minimum recommended for a GTS250. The Opteron 180 has a 110 watt TDP and I would imagine that rises quickly when overclocking.

Weird. Should not really be happening. This is a decent power supply with practically all rated power available on 12V.

No way those GTS250 + Opteron 180 would pull more than 300-350W, which leaves a healthy headroom...

Meanwhile now that the system is more or less set up the way i want i've swapped "aliexpress special" SSD i was using during setup with 860pro i had laying around from my main system, in which it has been replaced by nvme drive some time ago. And yeah, it definitely makes a difference, sata1 and everything:

Curiously NCQ does work (this is with nvidia sata drivers), but performance benefits are not as large as they are on more modern hardware.

Of course i did not do it for performance reasons - it was fast enough as is, i wanted reliability - i hate sketchy storage and powering a system on to play a game only to have to replace dead drive and restore backup. But performance is also nice to have and shows that whole idea that "any SSD will be bottlenecked by old sata anyway" is not really correct. Linear speeds are limited by old sata, random ones, which matter the most, are not. Decent SSD like this should also handle lack of TRIM much better.

My only regret is that i did not buy a dozen of this SSDs back when they were still made - they are great, nothing like this exists anymore and likely never will given sata is on its way out...

Repo Man11 wrote on Yesterday, 20:52:

As previously mentioned, this is with default voltage for both the CPU and the memory. I'm impressed by a 300 MHz overclock at stock core voltage.

Are you sure it's stock though? It was a thing back then, before CPU manufacturers implemented all this stuff themselves - silently overvolt, overclock and present it as "our motherboard is faster and overclocks better" .

On A8N32-SLI Deluxe Vcore is ~1.4-1.45V by default (depending on other settings like if cool and quiet is enabled), whole 0.1V above rated. And it is ~0.05V above what you set if you configure it manually. And the sensor shows junk and never shows voltage that high. So without poking at the board with multimeter there is no way to know the CPU is significantly overvolted by default.

I usually use SSDs with higher capacity. All SSDs are at 99% health.

Athlon 64 got 1TB Samsung EVO 860 (in IDE mode)
Phenom 2 X4 955 got 1TB Samsung EVO 850 (Sata 2, ahci, Windows XP) and 480GB Intel (server model, will be replaced by 2TB for Windows 7)
Vishera FX-8370 got 980GB Intel (server model, running in Sata 3 mode, ahci. Will be replaced by 2TB for Windows 10).

500GB would have been sufficient for Windows XP, but 1TB EVO 860 doesn't run properly on AMD SATA controllers present in AM2+/AM3+ so it has to be on nForce 4 chipset. SSDs got shifted and some ended up with different OS than intended. I could probably swap SSDs in Phenom 2 to have Windows XP on 480GB and Windows 7 on 1TB but I have found 1TB to be still insufficient.

AlexZ wrote on Today, 18:09:

I usually use SSDs with higher capacity. All SSDs are at 99% health. […]
Show full quote

I usually use SSDs with higher capacity. All SSDs are at 99% health.

Athlon 64 got 1TB Samsung EVO 860 (in IDE mode)
Phenom 2 X4 955 got 1TB Samsung EVO 850 (Sata 2, ahci, Windows XP) and 480GB Intel (server model, will be replaced by 2TB for Windows 7)
Vishera FX-8370 got 980GB Intel (server model, running in Sata 3 mode, ahci. Will be replaced by 2TB for Windows 10).

500GB would have been sufficient for Windows XP, but 1TB EVO 860 doesn't run properly on AMD SATA controllers present in AM2+/AM3+ so it has to be on nForce 4 chipset. SSDs got shifted and some ended up with different OS than intended. I could probably swap SSDs in Phenom 2 to have Windows XP on 480GB and Windows 7 on 1TB but I have found 1TB to be still insufficient.

Yeah, i agree that higher capacity is generally better. I like to separate boot drive from storage drive though, as well as boot drives for different OSs. This may be old-fashioned, but i like the convenience of being able to backup/restore it relatively quickly and things generally being more independent from each other. Can do that with partitions, but there is always higher risk of screw-ups here.

For XP 256GB is by far more than enough, i even have all the games i wanted installed there and it has like 65GB used.

I have a ton of cheap small SSDs for this purpose, very convenient swapping them around as needed etc. I do not have enough money to buy as many large SSDs, especially good ones 😀

I temporarily have 320GB 2.5 inch HDD for data in this system, i'll probably find something larger and replace it at some point. Do not want 3.5 one because of noise, so may be i'll end up using separate SSD too...

Okay A8N SLI Premium: the first hurdle was that it refused to POST with the GTS 250 that I used with the A8V-XE, so I tried my GTX 260 and that worked fine. There is a newer BIOS, so when I get around to it I'll update it and try the GTS 250 again just to see if I can make it work more than anything, the GTX is fine. It's been a very long time since I last fooled with an Nforce anything, so that has taken some getting used to. As I drank my morning coffee while starting the benchmarks, I was initially puzzled that the MB/s of the SSD wasn't improving as I overclocked the FSB, then I remembered "Oh yeah, PCI lock." With the Silicon Power SSD it achieves 142/114, which is fine. As I worked my way up to the same FSB that I've hit with the other three 939 boards, suddenly the multiplier dropped from twelve to ten, and the HT link was dropped to a multiplier of three. This surprised me because Google tells me this CPU is locked, but it obviously isn't (this is actually good news, but it was an annoying way to find that out). So then had to manually set the multiplier and the HT link.

The CPU voltage could be slightly high on the other two Asus boards, I didn't check closely (the Gigabyte board allows you to increase the core voltage by 10 percent so I set it to that and don't recall the net increase, but I have a screenshot of CPU-Z reporting it to be 1.350); I suppose I should say "Default voltage." Now that I look at CPU-Z screenshots, the A8V-XE has the core voltage at 1.4! With this motherboard, CPU-Z reports the core voltage as varying from 1.344 to 1.376. At this, it runs stable at 2.7 GHz, the HT is 1125, and the memory is @ 225 MHz. As I'm not using the same video card it isn't an apples to apples comparison, but it seems very similar.

Doom 3 Timedemo 1 gave me 60 FPS with the settings at high quality, 1024x768, and V sync enabled, which is almost identical to the A8V-XE. 3D 2001 score is just under 38,000 and the 3D 03 core is around 37,500. There is still some performance to be had with this board with an unlocked multiplier, but I'll admit that I'm surprised by the consistency with these boards.

Repo Man11 wrote on Today, 19:33:

Okay A8N SLI Premium: the first hurdle was that it refused to POST with the GTS 250 that I used with the A8V-XE, so I tried my GTX 260 and that worked fine. There is a newer BIOS, so when I get around to it I'll update it and try the GTS 250 again just to see if I can make it work more than anything, the GTX is fine. It's been a very long time since I last fooled with an Nforce anything, so that has taken some getting used to. As I drank my morning coffee while starting the benchmarks, I was initially puzzled that the MB/s of the SSD wasn't improving as I overclocked the FSB, then I remembered "Oh yeah, PCI lock." With the Silicon Power SSD it achieves 142/114, which is fine. As I worked my way up to the same FSB that I've hit with the other three 939 boards, suddenly the multiplier dropped from twelve to ten, and the HT link was dropped to a multiplier of three. This surprised me because Google tells me this CPU is locked, but it obviously isn't (this is actually good news, but it was an annoying way to find that out). So then had to manually set the multiplier and the HT link.

This CPUs are locked in a sense that you can not increase the multiplier above nominal value. You can, however, reduce it. The same on S939 and on AM2.

Repo Man11 wrote on Today, 19:33:

The CPU voltage could be slightly high on the other two Asus boards, I didn't check closely (the Gigabyte board allows you to increase the core voltage by 10 percent so I set it to that and don't recall the net increase, but I have a screenshot of CPU-Z reporting it to be 1.350); I suppose I should say "Default voltage." Now that I look at CPU-Z screenshots, the A8V-XE has the core voltage at 1.4! With this motherboard, CPU-Z reports the core voltage as varying from 1.344 to 1.376. At this, it runs stable at 2.7 GHz, the HT is 1125, and the memory is @ 225 MHz. As I'm not using the same video card it isn't an apples to apples comparison, but it seems very similar.

There is no reason to trust the sensor. At best it is inaccurate. At worst it will intentionally show what you want to see - ~nominal voltage. I've also seen it jump around just as you say on asus board and either readings are nowhere close to what voltage is actually there.