AlexZ wrote on 2026-03-24, 15:21:

Ok so let's be more specific, please add links to those web archives/forums with relevant data.

There are people here that used Turions on their desktop 754 mobos, you can use the search button and find a lot of compatibility tests here on Vogons, just as desktop cpu replacements, not overclocking them, (if you want you can easily reach 220 or 233 fsb with them, ml-34 will hit the 2~2.1ghz and a very mild OC, very safe) but more than enough for a compatibility list, this is quite easy lore, there is more on the web. I don't know which is the problem honestly, it is something I said?

EDIT:
If you want another source just look at the records on HWBOT, motherboard data are there in the description, this is another compatibility source and a testimony of what you can achieve pushing that silicon.
This is for ML-34
https://hwbot.org/benchmarks/cpu_frequency/su … issions/2264004

These links may be of interest to anyone wishing to run Turion 64 on s754 board:

• Athlon64, Turion users: Why do you choose those processors?
• https://silentpcreview.com/amd-turion-6 ... -desktop/
• https://web.archive.org/web/20070920233328/ht … h/turion-e.html - motherboard compatibility list

Especially the last link is important. It turns out some great AGP boards like Asus K8N-E are compatible. My Gigabyte GA-K8NE isn't listed, but its first BIOS is from dec 2005, so there is hope. From VIA boards I have MSI K8T Neo V.

MT-34 and ML-34 are the only two CPUs practically available as AMD CPUs were used in cheap laptops only. Good luck finding MT-37 / ML-37 or better.

The problem is also possible clock speeds

1800Mhz CPU, 200Mhz base, 9x multiplier, HT 800, 400 memory (divisor 9) - base clock
2200Mhz CPU, 245Mhz base, 9x multiplier, HT 980 (4x) / 735 (3x), 400 memory (divisor 11) - too high HT speed, won't be stable, with 3x multi HT speed is too low
2200Mhz CPU, 259Mhz base, 8.5x multiplier, HT 777 (3x), 400 memory (divisor 11) - HT speed slightly lower, will be probably slightly slower than Clawhammer 2.2Ghz <-- good pick
2270Mhz CPU, 267Mhz base, 8.5x multiplier, HT 801 (3x), 412 memory (divisor 11) - slight memory OC
2403Mhz CPU, 267Mhz base, 9x multiplier, HT 801 (3x), 437 memory (divisor 11) - edge of stability for CPU, high memory OC <-- top performance

When I ran my Clawhammer 3400+ as 3700+ temporarily for 3d mark, I ran it using 266Mhz base and 9x multiplier (last option). It gives correct HT speed and memory speed so it is the best option. I wouldn't recommend doing it with Clawhammer as the CPU already has very high power consumption as is and getting a bit more performance is not worth the risk of damaging an old board. An OCed CPU with higher voltage will draw more power.

It may be an interesting choice for someone looking for a Clawhammer 3400+, which are not cheap anymore. Get a MT-34 / ML-34 and run it at 2.2Ghz with 259Mhz base clock. All Turions should run at 2.2Ghz without insane voltage and consume less power than a Clawhammer. You also get SSE3.

It seems that the easier to find CPU's have a very low clock speed making them unsuitable for a high end socket 754 system. Perhaps they are better suited for a low power Windows 9X system!

I did a little test on MSI K8T Neo V (VIA K8T800 chipset) and Turion 64 ML-34. Arctic Freezer 64 pro was used directly on die, mounted on the original plastic frame. Pressure could be higher but is good enough for the test. For permanent usage probably use 0.3-0.5mm copper shim. Without thermal paste CPU temperature was about 65'C in BIOS setup, with paste it went down to 29'C. With prime95 running CPU temperature was 40'C.

CPU runs fine at 2200Mhz, 245Mhz base, 9x multiplier, HT 735 (3x), 400 memory (divisor 11). AIDA64 shows memory latency 70ns, which is a bit higher than what is achievable on Clawhammer 3400+ and more expensive boards (about 66ns). L1/L2 cache latency and bandwidth is the same as on Clawhammer 3400+. I would expect it to be very slightly slower than the Clawhammer due to lower HT speed and slightly higher memory latency. It should be possible to improve latency if one pays for a better board.

• BIOS detects the CPU correctly as AMD Turion(tm) 64 Mobile Technology ML-34
• BIOS setup doesn't allow changing CPU voltage, 1.45V is applied automatically, one shouldn't use higher voltage anyway
• CPU FSB Clock was set to 245
• CPU multiplier 8.5x is unavailable in BIOS setup, therefore I used 9x
• DRAM clock 166 Mhz was used in BIOS for 11 memory divisor. Higher clock results in too low divisor. DRAM clock 133 results in divisor 14 which is too high.
• Async Frequency Select was set to 66Mhz since we want 66Mhz AGP clock.
• LDT to AGP Lokar Frequency was set to 600Mhz. This is hypertransport speed and it gives us 3x multiplier.
• CAS latency was left on auto
• Cool'n'Quiet Support was disabled
• at 2.2Ghz idle Windows XP temperature is 29'C

It passed 30 minutes run of prime95 (L1/L2 stress), 3d mark 2000, 2001 & 2003.

CPU also boots Windows XP at 2400Mhz using DRAM clock 133 in BIOS, but real DRAM clock is too low (171Mhz with divisor 14). Stability was not tested at this speed. It didn't boot with DRAM clock 166 in BIOS as divisor 11 is too low for memory used.

AlexZ wrote on 2026-04-08, 18:45:

I did a little test on MSI K8T Neo V (VIA K8T800 chipset) and Turion 64 ML-34. Arctic Freezer 64 pro was used directly on die, m […]
Show full quote

I did a little test on MSI K8T Neo V (VIA K8T800 chipset) and Turion 64 ML-34. Arctic Freezer 64 pro was used directly on die, mounted on the original plastic frame. Pressure could be higher but is good enough for the test. For permanent usage probably use 0.3-0.5mm copper shim. Without thermal paste CPU temperature was about 65'C in BIOS setup, with paste it went down to 29'C. With prime95 running CPU temperature was 40'C.

CPU runs fine at 2200Mhz, 245Mhz base, 9x multiplier, HT 735 (3x), 400 memory (divisor 11). AIDA64 shows memory latency 70ns, which is a bit higher than what is achievable on Clawhammer 3400+ and more expensive boards (about 66ns). L1/L2 cache latency and bandwidth is the same as on Clawhammer 3400+. I would expect it to be very slightly slower than the Clawhammer due to lower HT speed and slightly higher memory latency. It should be possible to improve latency if one pays for a better board.

• BIOS detects the CPU correctly as AMD Turion(tm) 64 Mobile Technology ML-34
• BIOS setup doesn't allow changing CPU voltage, 1.45V is applied automatically, one shouldn't use higher voltage anyway
• CPU FSB Clock was set to 245
• CPU multiplier 8.5x is unavailable in BIOS setup, therefore I used 9x
• DRAM clock 166 Mhz was used in BIOS for 11 memory divisor. Higher clock results in too low divisor. DRAM clock 133 results in divisor 14 which is too high.
• Async Frequency Select was set to 66Mhz since we want 66Mhz PCIe clock.
• LDT to AGP Lokar Frequency was set to 600Mhz. This is hypertransport speed and it gives us 3x multiplier.
• CAS latency was left on auto
• Cool'n'Quiet Support was disabled
• at 2.2Ghz idle Windows XP temperature is 29'C

It passed 30 minutes run of prime95 (L1/L2 stress), 3d mark 2000, 2001 & 2003.

CPU also boots Windows XP at 2400Mhz using DRAM clock 133 in BIOS, but real DRAM clock is too low (171Mhz with divisor 14). Stability was not tested at this speed. It didn't boot with DRAM clock 166 in BIOS as divisor 11 is too low for memory used.

Via KT800 was a great chipset, but not the best OC candidate, as you already know, nevertheless with a nforce3/4 board you can easily beat those numbers. There are also boards that let's you adjust the VCore, even when you use a mobile CPU. The biggest performance hit is on the memory divider, but to address that you need very specific ICs on your DDR memory modules.
With the right combination I score ~55ns memory latency in AIDA64 in Win98SE and the system is perfectly stable.
You get those performances at almost the same TDP of a stock frequency desktop clawhammer.

I will test also Asus K8N-E (nForce3) and Gigabyte GA-K8NE (nForce4). We will see whether memory divider 12 can be used there and whether CPU multiplier 8.5x works. I will verify how high my ML-34 will go. If it goes well I may get a few more of them to test the ceiling. Typically, one would find the frequency when the CPU fails to boot OS and just go down by about 200Mhz not to be on the edge of stability.

Good DDR400 sticks are no longer available for a decent price, people selling them usually know what they have.

AlexZ wrote on Yesterday, 06:43:

I will test also Asus K8N-E (nForce3) and Gigabyte GA-K8NE (nForce4). We will see whether memory divider 12 can be used there and whether CPU multiplier 8.5x works. I will verify how high my ML-34 will go. If it goes well I may get a few more of them to test the ceiling. Typically, one would find the frequency when the CPU fails to boot OS and just go down by about 200Mhz not to be on the edge of stability.

Good DDR400 sticks are no longer available for a decent price, people selling them usually know what they have.

They not always know what they have, even on ebay; back then as you know, before Corsair, Crucial, Patriot, Gskill, Kingston started to hand bin the ICs of specific DDR memory manufacturers, before they knew what certain ddr memory ICs could achieve with a particular K8 family core they used those memory ICs even in their value lines of memory modules, if you know what to look for, you can still get amazing OC ddr modules cheap (labeled as common DDR400 modules)
Another advice, that maybe it's not needed: even if the CPU boots fine and looks like it is working fine, you should always test the K8 CPUs memory controller, sometimes it might look it is working fine but if you run intesive superpi runs and verify the results at the end, you could find that the memory controller overclock is not stable in the long run and it throws errors, this method is often better than prime95.

I did a another test on Asus K8N-E (nForce3 250Gb chipset) and Turion 64 ML-34. Arctic Freezer 64 pro was used directly on die, mounted on the original plastic frame with steel backplate. This time heatsink pressure is sufficient probably due to steel backplate and slightly different plastic frame. No copper shim is needed for long term usage.

CPU runs fine at 2200Mhz, 259Mhz base, 8.5x multiplier, HT 777 (3x), 400 memory (divisor 11). AIDA64 shows memory latency 66ns, which is equivalent to Clawhammer 3400+. We got 4ns less than MSI K8T Neo V. It is probably possible to go lower by tightening memory timings, this was not done due to time constraints.

• even latest BIOS doesn't like 2 memory sticks and Memclock Value 400. It automatically lowers it to 333 even if nothing is OCed. With one stick it allows 400. We can still reach DDR 400 by using 333 and higher CPU FSB Frequency though.
• BIOS detects the CPU correctly as AMD Turion(tm) 64 Mobile Technology ML-34
• BIOS setup allows changing CPU voltage (FID/VID Change set to Manual), I used 1.4V. Beware, when switching to manual voltage BIOS suggests 1.5V.
• CPU FSB Frequency was set to 259. Maximum stable FSB is 275. 277 doesn't boot to BIOS at all.
• CPU multiplier 8.5x is available in BIOS setup
• DRAM clock 333 was used in BIOS for 11 memory divisor. Higher clock results in too low divisor. DRAM clock 266 results in divisor 14 which is too high.
• Hypertransport multiplier was set to 3x
• CAS latency was left on auto
• Cool'n'Quiet Support was disabled
• With prime95 running, CPU temperature was 37'C due to less CPU voltage.
• Chipset runs very hot, I would highly recommend mounting a little fan.

The following memory divisors are possible for Turion 64 ML-34 based on CPU multiplier and Memclock Value in BIOS:

• 9x CPU multiplier - memory divisor 11 (Memclock Value 333) or 14 (Memclock Value 266)
• 8.5x CPU multiplier - memory divisor 11 (Memclock Value 333) or 13 (Memclock Value 266)
• 8x CPU multiplier - memory divisor 10 (Memclock Value 333) or 12 (Memclock Value 266)

8x CPU multiplier is impractical as we are limited by CPU FSB 275. Real memory speed is also too low as a result. I updated previous post with suggested frequencies as memory divisor 12 isn't practical. This complicates running at 2.4Ghz as OCing memory is unavoidable.

It passed 30 minutes run of prime95 (L1/L2 stress) at 2.4Ghz (ddr speed set to 266 conservatively in BIOS). CPU is prime95 stable at 2475Mhz (275 x 9) and 1.4V too. I didn't test lower voltage. It is the maximum frequency Turion 64 ML-34 can achieve on this board due to CPU FSB limit and 9x CPU multiplier. Turion 64 ML-37 would be needed to go higher, but those are unavailable. True ceiling thus cannot be determined.

The best options for Asus K8N-E and Turion 64 ML-34 are:

• 2200Mhz CPU, 259Mhz base, 8.5x multiplier, HT 777 (3x), 400 memory (divisor 11, Memclock Value 333 in BIOS) <-- good pick, will be equal to Clawhammer 3400+, but with SSE3 and lower TDP
• 2270Mhz CPU, 267Mhz base, 8.5x multiplier, HT 801 (3x), 412 memory (divisor 11, Memclock Value 333 in BIOS) - slight memory OC
• 2403Mhz CPU, 267Mhz base, 9x multiplier, HT 801 (3x), 437 memory (divisor 11, Memclock Value 333 in BIOS) - high memory OC <-- top performance, will beat Clawhammer 3700+

At DDR 406, CL2 could no longer be used, CL2.5 was needed. The system could boot Windows XP at DDR 437@2.7V and relaxed timings, but was not stable. The other two options are probably not worth the time and risk.

Based on my observations I would recommend to go with the first option, Memory OC is problematic and requires lot of tinkering with timings and memtesting. That limits CPU frequency to 2.2Ghz due to memory divisor 11. To make use of memory divisor 12, we would need CPU frequency 2.4Ghz and CPU FSB 300, far beyond the max stable value 275. To make use of memory divisor 13, we would need CPU frequency 2.6Ghz and CPU FSB 306 which is far beyond the max stable value 275.

Turion 64 ML-34 is a good replacement for Clawhammer 3400+, supports SSE3 and has lower TDP. It is widely available and cheap.

Turion 64 ML-37 unlocks multipliers up to 10x and another interesting option becomes possible:

2403Mhz CPU, 253Mhz base, 9.5x multiplier, HT 758 (3x), 400 memory (divisor 12, Memclock Value 333 in BIOS) - replacement of Clawhammer 3700+

Memory timings and sub timings are paramount to get stable oc with athlon64, with Newark 3400+, which is 2.2 GHz stock, I can clock stable up to 2933, with ml-37 which is a 2 GHz stock, I can clock stable at 2666, with ml-34 I clock stable at 2400, all of them Vcore 1.450V, AGP at 66mhz, PCI at 33mhz, htt X3 at 800mhz. Nforce3 250

These are my Memory Timings with memory at DDR533 frequency (1:1):

CAS Latency (Tcl): [3]
RAS# active time (Tras): [6T]
RAS# to CAS# delay (Trcd): [3T]
Row Precharge Time (Trp): [3T]
Row active to row delay (Trrd): [2T]
Row Cycle time (Trc): [9T]
Row Refresh cyc time (Trfc): [11T]
Read-to-Write time (Trwt): [2T]
Write Recovery Time (Twr): [3 bus clock]
Write-to-Read Delay (Twtr): [2 bus clock]
Refresh Rate (Tref): [200 MHz , 15.6 us]
Read Preamble value: [5.5ns]
Async Latency value: [7ns]

As you noticed a single Memory module is way more stable than 2 modules due to the fact that the CPU memory controller has to maintain signal integrity at such a high clock.
As you have already stated you should never if possible use auto for memory timings settings, that is going to limit you a lot, because the Mobo is going to use the default manufacturer SPD timings of the ram module which are not going to work at different frequencies than stock ones.

Many s754 boards do not allow to set those timings at all. MSI K8T Neo V (VIA K8T800) allows to set CAS Latency only, the rest is set automatically and will most likely be wrong for memory OC. Asus K8N-E (nForce3) allows setting only the top 4 you listed + Command Rate. Asus K8N4-E (nForce4) has the same options as Asus K8N-E according to manual. Asrock K8NF3-VSTA (nForce3) has slightly more memory options. Gigabyte GA-K8NE (nForce4) is a hidden gem, manual doesn't list these options at all but many of them are in BIOS after pressing ctrl+F1.

On Asus K8N-E I used CAS Latency=3, RAS# to CAS# delay=4, Row Precharge Time=4, RAS# active time=8, voltage set to 2.7V with 2 GEIL DDR400 512MB memory sticks and it wasn't stable. My best memory stick is OCZ Platinum 1GB, but I have only one.

1GB RAM is ok for nForce3 system for dual booting Win98/WinXP, but not enough for nForce4 system intended for WinXP only. Late games need really a lot of memory and with 2GB RAM swap file can be disabled. With SSD it makes it a fast system. nForce4 can run many late games after 2005 as long as they are not FPS, racing or flight sims. It can run strategy games and sports games well.

With 1GB RAM and nForce3 one can argue CPU clock 2.4-2.6Ghz isn't necessary at all, neither is memory OC or even SSE3, it is an AGP system and games until 2004 will run fine. From 2005 and later, AGP card is not sufficient for max resolution and details. ML-34 at 2.2Ghz is a great fit for both nForce3 and nForce4. ML-37 is a great fit for nForce4.

Abit NV8/nforce4 has extensive timings that you can set; however the main problem i see is the limitation to 1 memory module only! 1gb of RAM simply don't cut it even in early Windows XP era games (2000-2004).

nd22 wrote on Today, 07:25:

Abit NV8/nforce4 has extensive timings that you can set; however the main problem i see is the limitation to 1 memory module only! 1gb of RAM simply don't cut it even in early Windows XP era games (2000-2004).

You can use more than one modules with more relaxed timings, that was what I meant. You can set all the timings when you reach the OS using utilities if your bios does not allow you to do so.
Btw for XP gaming you can go core duo or core quad, which is cheaper and better. That's the reason why I use this system on win98se.

nd22 wrote on Today, 07:25:

Abit NV8/nforce4 has extensive timings that you can set; however the main problem i see is the limitation to 1 memory module only! 1gb of RAM simply don't cut it even in early Windows XP era games (2000-2004).

I wonder if Turion 64 ML-34 works on Abit NV8. They are very cheap on ebay, just 10 EUR + 10 EUR postage.

I think most s754 users will not be interested in memory OC or tinkering with timings, therefore I suggested options that don't OC memory or HT and should be safe even long term.

I will test it on Gigabyte GA-K8NE next. I had Gigabyte GA-K8NS and I think it had same memory timing options as Gigabyte GA-K8NE. I checked and Gigabyte GA-K8NE has all memory timing options Madowax listed. Gigabyte GA-K8NS and Asrock K8NF3-VSTA were the best s754 AGP boards I had.

AlexZ wrote on Today, 07:44:

I wonder if Turion 64 ML-34 works on Abit NV8. They are very cheap on ebay, just 10 EUR + 10 EUR postage. […]
Show full quote

nd22 wrote on Today, 07:25:

Abit NV8/nforce4 has extensive timings that you can set; however the main problem i see is the limitation to 1 memory module only! 1gb of RAM simply don't cut it even in early Windows XP era games (2000-2004).

I wonder if Turion 64 ML-34 works on Abit NV8. They are very cheap on ebay, just 10 EUR + 10 EUR postage.

I think most s754 users will not be interested in memory OC or tinkering with timings, therefore I suggested options that don't OC memory or HT and should be safe even long term.

I will test it on Gigabyte GA-K8NE next. I had Gigabyte GA-K8NS and I think it had same memory timing options as Gigabyte GA-K8NE. I checked and Gigabyte GA-K8NE has all memory timing options Madowax listed. Gigabyte GA-K8NS and Asrock K8NF3-VSTA were the best s754 AGP boards I had.

As we already said you can just select a memory divider close to stock memory frequency and you will have a very capable and cheap win98/early xp platform. If you want motherboard longevity, you should never OC HTT bus, that is what cooks the chipset on nforce3/4, memory is just a bit more complicated but it is not dangerous for the motherboard per se.
You can change the whole range of memory timings directly in the OS with A64Tweaker Utility.

You boot at a safe memory clock and then you tweak it inside the os to your perfect performance profile:

XtremeSystems Forums - WPCREDIT tweaks for the A64

This is going to work even on those mobos that do not have extended memory timings in the bios.
BTW DFI and Chaintech have extended timings in their nforce3/4 boards bioses.
Do not get lured into the Command Rate 1T race, if you reach a 33% wider memory bus you don't need 1T, it won't be stable at very high memory frequency even if it looks like so at boot.

You can't freely select memory divisor. It is selected indirectly through combination of CPU multiplier and Memclock Value in BIOS (called MEMCLK Frequency in A64 Tweaker). Some BIOSes display MEMCLK after multiplying by 2. I provided the list of possible divisors in one of previous posts for ML-34 and calculated preferred divisor for ML-37.

I agree about MT OC and avoided suggesting it. nForce3/4 chipsets come without active cooler but a little fan needs to be added as they run very hot. Probably many overclockers killed them with heat.

Tweaking memory timings in Windows is insufficient as they need to be thoroughly memtested. Memtest86+ needs to be run for many hours and needs access to almost whole memory. I have seen cases where there was just one failure after like 6 hours. When tweaking memory timings, I don't boot into Windows XP as it could get corrupted but just boot Memtest86+ after tweaking 1-2 memory timings. Each change needs a quick test of 1-2h. Many aggresive timings fail within a few minutes. I would typically re-test final timings with 12-14h Memtest86+ run.

I suggest to run superpi in windows to achieve a very good memory testing, you should always verify the superpi calculation results at the end, it is comparable to memtest and you can run it for the amount of time you want. I'm sure you have a stock os install image ready to restore if you end up corrupting the OS. If you have those settings in your bios, you can go on with memtest at boot.

You can select memory dividers by converting the resulting stock memory frequency to the new frequency after fsb overclock. It is just a proportional calculation, if I'm not mistaken. I know you listed those that's why I suggested to tweak memory timings to achieve a 221.66 mhz memory bus and 266 MHz CPU fsb. (Ddr333 divider)
You are basically custom tuning a car engine, so there should be no compromise in the effort you are putting in this, if you find it entertaining. 😉