Tzk wrote on 2026-07-15, 04:48:

Sadly software tools can only read the SPD of the ram sticks. There's no way to identify the chips that way. However ram chips g […]
Show full quote

Ozzuneoj wrote on 2026-07-14, 18:29:

This is going to sound like a total noob question, but I'm asking because my DDR test system is not currently ready to hook up... what is the best program to identify the chips used on memory that has glued-on heat spreaders? Is there anything better than CPU-Z for something as old as an Nforce 2 system?

I have lots of memory with head spreaders, and I really don't want to tear them apart to see what chips they use. I have 5 OCZ PC3200 "1024MB EL Dual Ch Platinum Edition 2-3-2-5", a Crucial Ballistix of some kind (looks like I wrote on it 512MB DDR400 2-2-2-6 2.8v ... not sure when I did this, 🤣) as well as various Corsair XMS, SuperTalent and Patriot memory.

EDIT: Were there any overclocking-friendly DDR chips that were in the smaller, more square-shaped (DDR2-like) BGA package? I have come across some of those as well, but they have minimal markings. On one "AMPO PC3200/1GB" module the chips are mostly blank except for a tiny little bullseye symbol and "DDR DXT".

Sadly software tools can only read the SPD of the ram sticks. There's no way to identify the chips that way. However ram chips got very specific metal dots or fins on the shorter edge of the plastic package. So if you can see the short edge below the heatspreaders, you can at least guess what they are. Winbond for example does have two lengthy metal tabs. Samsung 256mbit rev. F chips (TCCC/TCCD) got six tiny metal dots ( . .. .. . ). Micron has a round cutout. Other chips mostly have four dots, but their position varies.

There are BGA chips - for example there's Samsung TCCC (TSOP package) and ZCCC (BGA package). Or Infineon BT-6 (TSOP) and BC-6 (BGA). The only "good" BGA chips i know about is Infineon CF-5 (basically 512mbit CE-6 in BGA package). If you got those, try testing it. Just beware that some of these OEM sticks were made cheap and may have 128Mx4 organization, which will only work on Athlon 64 but not 462 or 423/478/479.

Besides that you can check old ram chip lists or the above mentioned ddr chip thread on hardwareluxx forums. Your Ballistix got their chips on the white label. If it's for example named: BL12864AL1065.16FD3, then it's 5B D chips. If it's named ....16TG then it's 5B G My DDR400 Ballistix (2x512mb 5B G) are able to run 2-2-2-5 2.8V at 200Mhz and 2.5-2-2-9 250Mhz 2.8V on Nforce 2. Setting Tras to 9 is key for stability at 250MHz on my system.

Regarding your mentioned sticks: There are two huge lists for Corsair XMS which you can check. One is the DDR Inquiry List on Corsair forums and the other was compiled by stunned_guy on Hardwareluxx. I took both lists and made them searcheable here . I also found a bit of information for GEIL and Twinmos (see last link). You can also check the DDR1 databases ( Link #1 for 512mb and #2 for 1gb) on hardwareluxx, those are mostly accurate but may contain smaller errors.

In the end it's trial, error and a lot of testing. Having a good 939 system at hand to pre-test the rams is a great idea.

Thanks for the info! I kept reading about the little metal dots and tabs but it never occurred to me that you could see these under the edge of the heatspreader. That helps tremendously! Looks like none of my OCZ Platinum is using Winbond. They all just have four dots, though some are closer together or further apart than others. If there are example pictures anywhere of all the dot patterns that would help me ID them a bit better. There are a few at the Hardwareluxx page linked earlier and I would say the OCZ ones most closely resemble either Hynix or Aeneon\Infineon\Qimonda, but without a more comprehensive list of chips and example pictures it's hard to say for sure.

Out of all of my DDR, I managed to find one single stick that definitely uses Winbond chips that are relabeled as "Liberty". It has a very generic green heat spreader with no labels or other information on it.

The metal tabs at the ends and the two dots on the top are unmistakable. Sadly, again, it is just a single one. If only I had a pair of them...

Regarding testing them on a 939 system, I have an Nforce 3 Ultra system with an X2 4200+ that I can use. What does the 939 system get me with regard to memory testing vs just using my Nforce 2? It's been so long since I messed with overclocking on 939... I can't remember what kind of flexibility I get with setting different memory speeds.

This is BH-5, these sticks pop up all the time on Ebay.

https://www.ebay.com/itm/324257375160?_skw=IN … ABk9SR_KfhIHtZw

https://www.ebay.com/itm/406836401072?_skw=In … %3ABFBMntKPge1n

supercordo wrote on 2026-07-15, 23:16:

This is BH-5, these sticks pop up all the time on Ebay.

https://www.ebay.com/itm/324257375160?_skw=IN … ABk9SR_KfhIHtZw

https://www.ebay.com/itm/406836401072?_skw=In … %3ABFBMntKPge1n

Except for the timings in BIOS and no boot with CL3, how can I distinct BH5 from CH5?

They look the same. A minor difference is the production date, BH was made till about week 20-25 2003, ch-5 is newer.
However there’s also UTT BH and UTT CH which basically is untested bull chips. These are usually relabeled and were sold in 2005. so you can’t tell BH/CH from production date alone.

The best and most reliable test is still trying to boot CL3.

Regarding the Infineon sticks: those got a non-zero digit right in front of the chip revision (9AT-6) for relabeled chips. I know of 4 7 and 9 as digits.

Ozzuneoj wrote on 2026-07-15, 15:54:

Regarding testing them on a 939 system, I have an Nforce 3 Ultra system with an X2 4200+ that I can use. What does the 939 system get me with regard to memory testing vs just using my Nforce 2? It's been so long since I messed with overclocking on 939... I can't remember what kind of flexibility I get with setting different memory speeds.

One less variable while testing. You’re usually not running into rem controller or chipset limits on 939. you can test directly on socket 462, just be aware that the board (not not the ram) might limit your OC.

I usually bin my ram on 939 to find the best stick of a lot and then try these on NF2. I got about 40 sticks of Infoneon AT-6 though. So binning is a must.

[quote=I usually bin my ram on 939 to find the best stick of a lot and then try these on NF2. I got about 40 sticks of Infoneon AT-6 though. So binning is a must.
[/quote]

That's a good idea. I never thought about that. Something to do this weekend.

Anyone know why there is such a big gap in FPS with "Game 4 - Nature" between my 7600GT and my X1950 Pro / digitalbath's 7600GT ? It's killing my total score.

My 7600 GT

My X1950 Pro

digitalbath's 7600 GT

Maybe your RAM timings and fsb. Also, what Mhz do you have your AGP slot running at? Do you have AGP days writes on or off? What is your AGP aperture set at?

Your 7600GT is also not overclocked as much.

I'm going to throw my hat in the ring with the top T-Bird score so far. I know the CPU can do more if I have more voltage available. Going to see what I can do with my NF7 2.0 board.

CPU cooler is a Thermaltake Silent Tower with lapped base and dual upgraded 92mm fans.

Edit: Redid the screenshot to show the info on the primary window of 3DMark

cyclone3d wrote on 2026-07-18, 04:58:

Maybe your RAM timings and fsb. Also, what Mhz do you have your AGP slot running at? Do you have AGP days writes on or off? What is your AGP aperture set at?

Your 7600GT is also not overclocked as much.

AGP is running at 66MHz. I don't think AGP aperture matters too much since it's really just a PCIe card with a AGP bridge. There isn't too much difference in the overclock compared to digitalbath. If you compare the 7600GT to the X1950 Pro, the 7600GT dominates in every test except for "Game 4 - Nature". Very weird

zuldan wrote on 2026-07-18, 06:01:

cyclone3d wrote on 2026-07-18, 04:58:

Maybe your RAM timings and fsb. Also, what Mhz do you have your AGP slot running at? Do you have AGP days writes on or off? What is your AGP aperture set at?

Your 7600GT is also not overclocked as much.

AGP is running at 66MHz. I don't think AGP aperture matters too much since it's really just a PCIe card with a AGP bridge. There isn't too much difference in the overclock compared to digitalbath. If you compare the 7600GT to the X1950 Pro, the 7600GT dominates in every test except for "Game 4 - Nature". Very weird

I think the 30MHz more FSB and the resulting 30MHz more memory clock are making a big difference together with CL2 vs your CL2.5 - just overall.
I think it's quite interesting how well the 7000 series geforces performe in 3Dmark01SE.

Nature is the most gpu limited test in 01 SE while car chase low is the most cpu limited. Maybe the 7600 driver got less cpu overhead and the 1950 is faster and thus gets a higher nature score?

Besides that 01SE scales very well with FSB and low timings on athlon xp, especially with fast GPUs.

A while ago I tested this on Nforce 2 with a 9800 Pro. Going from 2200 to 2500Mhz on the cpu scaled as much as pushing the fsb and timings from 200 to 260 at equal cpu clocks.

200x11 = 2200Mhz 2.5-3-3-8 13365
200x12.5 = 2500Mhz 2.5-3-3-8 14381
262x9.5 = 2497Mhz 2-2-2-6 15202

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working, I'll go back to 150x15.5 to get 2325 and see where I get there. ITMT, I think I will play with the GPU clocks.

Headlines - 2x Bartons @2346MHz (138x17), 2-3-3-6, HD3850 512 @ stock
3DMark01SE - 16246
Cinebench Multi - 537
Cinebench Single - 292
SuperPi 1M - 54.937s
CPUMark99 - 195

geordiepingu wrote on 2026-07-18, 08:37:

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working, I'll go back to 150x15. […]
Show full quote

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working, I'll go back to 150x15.5 to get 2325 and see where I get there. ITMT, I think I will play with the GPU clocks.

Headlines - 2x Bartons @2346MHz (138x17), 2-3-3-6, HD3850 512 @ stock
3DMark01SE - 16246
Cinebench Multi - 537
Cinebench Single - 292
SuperPi 1M - 54.937s
CPUMark99 - 195

That dual setup is absolutely gorgeous.

geordiepingu wrote on 2026-07-18, 08:37:

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working.

Have you tried changing the multi in windows? My dual Tyan can adjust these on the fly using crystal cpu Id.

Tzk wrote on 2026-07-18, 14:49:

geordiepingu wrote on 2026-07-18, 08:37:

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working.

Have you tried changing the multi in windows? My dual Tyan can adjust these on the fly using crystal cpu Id.

yes, unfortunately, the board's multiplier gen is missing a bit, so I can only get 12.5x without soldering.

geordiepingu wrote on 2026-07-18, 16:32:

Tzk wrote on 2026-07-18, 14:49:

geordiepingu wrote on 2026-07-18, 08:37:

EDIT - Seems to be where I am stable at the moment. If I can find a socket wire mod to get 15.5 working.

Have you tried changing the multi in windows? My dual Tyan can adjust these on the fly using crystal cpu Id.

yes, unfortunately, the board's multiplier gen is missing a bit, so I can only get 12.5x without soldering.

You can use a single strand wire mod in the socket. No soldering so it is easily reversible.

https://web.archive.org/web/20150330212146/ht … _en/amd_pinmod/

I wrote in my build thread why I am soldering rather than dropping wire into a socket Dual Athlon MP Rig Build and Repair | K7D Master in a Chieftec Dragon

Tl;dr concerns around pin fragility and it is quicker than unmounting and remounting 3x waterblocks each time. Doesn’t take long, about 10 mins to do both sockets, less if it’s just a couple of wires. Also easily reversible, if you heat the wire up it just sticks to the tip of the iron.

tehsiggi wrote on 2026-07-18, 06:27:

I think the 30MHz more FSB and the resulting 30MHz more memory clock are making a big difference together with CL2 vs your CL2.5 - just overall.
I think it's quite interesting how well the 7000 series geforces performe in 3Dmark01SE.

I need to get my Winbond memory working so I can get out of this low FSB hole I’m stuck in 🥹

Tzk wrote on 2026-07-18, 08:30:

Nature is the most gpu limited test in 01 SE while car chase low is the most cpu limited. Maybe the 7600 driver got less cpu ove […]
Show full quote

Nature is the most gpu limited test in 01 SE while car chase low is the most cpu limited. Maybe the 7600 driver got less cpu overhead and the 1950 is faster and thus gets a higher nature score?

Besides that 01SE scales very well with FSB and low timings on athlon xp, especially with fast GPUs.

A while ago I tested this on Nforce 2 with a 9800 Pro. Going from 2200 to 2500Mhz on the cpu scaled as much as pushing the fsb and timings from 200 to 260 at equal cpu clocks.

200x11 = 2200Mhz 2.5-3-3-8 13365
200x12.5 = 2500Mhz 2.5-3-3-8 14381
262x9.5 = 2497Mhz 2-2-2-6 15202

I’ll try different drivers versions to see if I can get any sort of improvement. Interesting to see the CPU scaling on a NF2 🤔

@Zuldan
No tricks on my end for the 3DMark2001 runs, except for a negative (or was it positive?) LOD setting in RivaTuner.

I wonder if there is a better driver version for the 7600GT. Detonator 44.03 is known to have good results, but this version should be too old for the 76GT.

Winbond FTW! Low latency is the way to go here. In terms of performance, the difference between 2,5-3-3-8 and 2-2-2-5 is somewhere at 25MHz FSB speed. That means, you have to use a 25MHz higher FSB Speed with CL2,5 to equal the CL2 performance.

The unobtainium has been obtained...as soon as I saw supercordo's prized possession I knew I had to have it. Look what arrived today! I managed to get both WB2 and a WB3

WB2

WB3

The full retro collection. WB2 / WB3 / WB4