I don't think you have to go to a much later GPU before they were munching off the 12V and you've got 100s of watts of that untouched.

Photos? Seems nice. Mainly I wonder how much of a speed bottleneck conventional PCI places on your USB 3.0 card.

SW-SSG wrote:

Photos? Seems nice. Mainly I wonder how much of a speed bottleneck conventional PCI places on your USB 3.0 card.

Well it is and it isn't a bottleneck, PCI does just over a gigabit per sec while 3.0 is 5Gbit, however, a lot of the first crop of non-enthusiast stuff only does about a gigabit, like mainstream USB 3.0 flash drives, USB external drives with SATA platter drives, the drives have high burst but still for "ordinary" platter drives 120 or 130 MBps sustained is mainstream performance. Then it's still 4x faster than USB 2.0 high speed, so useful improvement.

I do have photos, which I will upload in the next day or so 😀

The speeds on the PCI interface should be as BitWrangler states, which is 130MB/s. There is the potential for many devices to be connected to the PCI daisy chain, but only 1 device will be used at any given time on average, so it works. For fun, I will do some CrystalDiskMark benchmarks to see how well it performs using a RAMDisk for maximum throughput, which should be fun 😁

On the topic of the GPU.. during this build I saw PhilsComputerLab's video on Socket A (462) power usage testing, along with a few other threads here on the boards. I got so focused on the fact that "Socket 462 draws most of its power from the 5V rail", and how even if a 4-pin CPU connector exists, it still might draw over the 5V. xbitlabs had some great power consummption data on the 9xxx era GPUs. I became determined to find modern PSU that would run a CPU/GPU combo well on Socket 462.

Some questions:

• So BitWrangler, are you saying a later model graphics card like a Radeon HD 3850 AGP card would draw from the 12V rail?
• Which era of Radeon GPUs started to use the 12V rail? and do the cards draw power from the 5V rail through the AGP Socket up to a point, and then use the molex/6pin power connector over the 12V rail for the remainder of its power needs?
• Does one need to plug the GPU into the PSU on a specific port to ensure the GPU draws from the 12V rail?

At the moment I am assuming a bit of a bottleneck with the 9600 XT, as people like to talk about the 9800 Pro often when discussing the more performant Athlon XP CPU's. Although, I am enjoying the 9600 XT (paid $13) and it's power efficient RV360 chipset.

Sorry, couldn't find any reliable references, think there used to be, but being "out of date" now seem to have dropped off the web... however embedded in this older online calculator are values for older stuff https://outervision.com/power-supply-calculator

Also yes most socket A were 5V thirsty... until NF2 boards came out then they were using 12V like P4 boards had been doing for a few years by then.

The outervision psu calculator is inaccurate, at least for the Athlon XP era.

It shows 158 W Load Wattage if I enter:

CPU: Athlon XP 2800+ 2083 MHz Barton
Memory: 2x1GB DDR
Storage: 1x SATA 5.4K RPM
1x kb, 1x mouse
no case fans, no gpu, dvd drive, no addon cards

If I actually put the full config in in, it yields 237 W. This isn't even in the stratosphere of the 112W max utilization I am getting from the build at load during gaming.

Side note: The same exact overall build was previously (for a brief time) on a Thermaltake TR2-430W with 22A on 5V rail and the system still ran just fine in full load gaming. This PSU has something closer to 65% efficiency and a tendancy for poor noise ripple when a rail is strained. A killowatt was reading the build at 125-128W utilization (130W max on 3.3+5V on this PSU). Even still, I could game on this PSU for hours on end without any instability issues.

I am still digging for more info on the topic of 12v rail gpu's. I would love to identify some cards and pick one up in the next few weeks.

http://www.tomshardware.com/reviews/corsair-r … upply,4303.html
The label in the photo shows max combined 150W.
Their 3.3V reg is max 25A = 82.5W
Their 5V reg is max 25A = 125W

Look at PSU reviews from 2001-2005
Found this PSU model from a 2005 article.
https://www.newegg.com/Product/Product.aspx?I … N82E16817151021
200-400W was typical for socket A. Rails on this model are lower than your model. It should be fine. Or keep finding PSU models of that era and compare.

Hey Auzner, I am fully confident in the PSU I have chosen, and it wasn't in question prior to creating the thread. The Corsair RM750X is power efficient and well built, and pairs nicely with the components I have. In-fact, its part of the reason I was showing off this build, as its a great option for those looking to pair up Socket 462 hardware with modern PSU features. 😀

Discussions about the 12V rail were mainly about being able to add a slightly more powerful GPU to unfurl more of the CPU's power.

BitWrangler posted the outervision psu calculator link in realtion to finding GPU's that draw off the 12V rail and calculating their power draw rather than trying to correct anything with my PSU choice, and my follow up was just showing that there was a discrepancy with my load at wall numbers vs what the calculator was showing. So it is hard to use the outervision calc to get accurate numbers.

Well 9600XT matches socket A era just fine.

If an AGP card needs an additional molex for power that's easy to mod from 12V. It's whatever card you can find or want to pay for. Mix anything with a newer design.
A DC-DC buck converter as a second power supply to use a modern one's larger 12A rail would work.
It sits inbetween the PSU and the power connector to the video card.
PSU --> Molex to input 12V --> DC-DC buck step-down adj pot to 5v ---> Molex 5V rail --> AGP card w/ vreg using the 5V rail
If the vreg on the card uses 12V then you don't need this. Ask the seller for the model numbers on the ICs near the molex connector on the card and find the datasheets.
In this case the 9600XT could be either way since the part says 5 or 12V input.
http://archive.techarp.com/showarticle3a3a.ht … rtno=111&pgno=7
https://www.intersil.com/content/dam/Intersil … sl6/isl6522.pdf

To be sure, V+A meters are also cheap. Splice it into the molex and measure the card's draw on both 5V and 12V (order 2) to see where it's lob-sided.

I was researching all day today and you are right Auzner. One would have to bridge something between the PSU and the Power Connector to actually get the readings needed to determine which cards would work well, aside from just guessing and hoping you are all good. Power is distributed across all rails in varying amounts. I probobly won't setup the hardware to test cards like this, but I am compiling some info from other sources to zero in on some potential cards to try out. Interestingly enough, the 9600 XT draws 45% of its power from the 3.3V rail.

I do think I am hitting a CPU bottleneck though on UT2004, and I have read multiple sites say the game is notorious for being a CPU bottleneck. Its hard to prove since all benchmarks for Athlon XP era machines are on UT2003. So slipping in a more powerful GPU might result in no gains for this specific game. I have found 1 link that shows Athlon XP CPUs being tested against both UT2003 and UT2004, so there is a difference between them perfromance wise even though they contain a lot of the same content.

There are some higher CPUs to get for the socket over a 2800 barton. Had to do some research, but I remember hearing the XP-M's OC really well. More CPUs will let you get apples to apples on that test. Can you underclock the one you have and rerun? Graphing multiple clocks vs the game might show a growing bottleneck. Try slower ram too?

TB - Athlon XP 3100+
Barton - Athlon XP 3200+
Sempron - 3000 or 3300+
Barton - Athlon XP-M 3000+
Athlon XP-M 2500 or 2600+

http://techreport.com/review/6341/overclockin … -2500-processor

http://www.ebay.com/itm/AMD-Athlon-XP-M-AXMH2 … boAAOSwl75ZpvWF

Well, if I buy too many spare parts, then I am tempted to build more machines 😉 So I try and do a lot of research and spec things right before buying.

I bought the Athlon XP 2800+ Barton core because it matched the FSB of my motherboard, and was the most power per TDP watt upgrade over the machines original Sempron 1500+. I was trying to work under a certain 5V power envelope and the 2800+ fit the build just right. Traditionally I do not overclock, and doing so would put more strain on rails that I am already trying to keep balanced to run a modern PSU.

Instead of testing more CPU's, I went another direction and bought a new graphics card: a Radeon X800 Pro for $25. From a power perspective it sits well within the system, and pushes more juice over the 12V rail, however during gameplay... no change in performance. Only 1-2 frames on the various time demos I ran on UT2004. Also, Quake 4, which I was using to really stress the machine, showed no change in performance either. I think the X800 fits better with an Athlon 64.

So that means the Athlon XP 2800+ is the bottleneck, but that is just fine. Whats important is the 2800+ Barton and the 9600 XT are a great match, and that I am getting the most performance power from the Socket 462 build.

Heh it's like, darn, I need an optical drive, and there's one in a box of parts on craigslist for $15, wow, there's most of a slot 1 system here, just need another PSU... oh cool, here's a PSU in a box of parts for $15 ....

Have you considered an X1600PRO AGP for this system?

I did not, as the X1600 Pro uses 41 watts, and the 9600 XT uses 22.72 watts.

The X800 Pro that I picked up for testing uses 48.17 watts pushing 28.59 watts over the 12V rail, and it outperforms the X1600 Pro.

The X800 Pro yielded little to no improvement in performance over the 9600 XT. So the X1600 Pro would do no better.

an x800 should do better than a 9600xt, the Barton's could push up to a 6800 NU/x800 without an issue, the bottleneck started with the x800Pro/6800GT. You've got something else going on if they preform the same

Finding benchmark examples to use as a basis for comparison has been a challenge.

The best I could find is the following build:

http://www.overclockersclub.com/reviews/shutt … &gid=1&pid=1400

Its an Athlon XP 2500+ overclocked to 3200+ speeds. As best I can tell (as it is never explicitly mentioned in the article), there is a 9800 Pro in the system. It also has faster DDR400 ram in a dual-channel configuration, and an nForce chipset. It yields 76fps in a UT2003 bot match.

My Athlon XP 2800+ (stock), DDR333 in single channel, 9600 XT, SiS 741GX/963L chipset, running UT2004 @ 1024x768, Normal Settings yields 55fps average in a botmatch on Deck17.

BitWrangler wrote:

Heh it's like, darn, I need an optical drive, and there's one in a box of parts on craigslist for $15, wow, there's most of a slot 1 system here, just need another PSU... oh cool, here's a PSU in a box of parts for $15 ....

🤣 that could go in a sig

But yeah SDRAM vs DDR400, quite a bus difference. DDR DC was a big leap in memory. Newer games will make use of that

not for Socket 462. Even with a Nforce 2 ultra the difference was 5% at best.