No the problem occurs even at base clock 2Ghz for ML-37. Difference in behaviour at 2.0, 2.2, 2.4Ghz is just a timing issue being exposed. You need very rigorous testing at multiple CPU frequencies to confirm CR1 is stable.

Right now I'm trying 1x Kingston HyperX 1GB and 1x OCZ Platinum 1GB DDR400 together. CR1 may work with 2x Corsair or 2x OCZ Platinum, but I don't have those.

The IMC is very timings dependent, even stock timings might not be a good choice for a particular core revision and CR1 is even more difficult, sometimes a little ram overvolt might help a lot .

Mixing different brands with the same memory ICs may work, while same brand with different memoy ICs may not work stable at all.

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

Kruton 9000 wrote on 2026-04-29, 16:17:

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

This is not about standard 754 CPUs, these are mobiles, latest 754 core (basically single channel 939 with low leakage transistors and memory controller new stepping) IMC is revised and better than the 130 cpus one; the signal integrity is also influenced by motherboards traces/shielding and planes layout, given the same specs (chipset, VRMs, etc etc) there are better boards with less noise and there are some that are just terrible (when you push them out of the stock frequency comfort zone); but as you said less memory modules are better for signal integrity, it was also true that to maintain stability with something like winbond bh5 memory at very high clocks it was common to modify the motherboard ram vrm to provide up to 3.0/3.2V to memory slots, which was a massive overvolt from 2.6V, but bh5 loved it.
Addressing these caveats was the reason why DFI LanParty boards (6 layers, better traceroute, better shielding, huge ram overvolting capabilities) were so praised by the OC community.

Madowax wrote on 2026-04-30, 04:04:

Kruton 9000 wrote on 2026-04-29, 16:17:

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

This is not about standard 754 CPUs, these are mobiles, latest 754 core (basically single channel 939 with low leakage transistors and memory controller new stepping) IMC is revised and better than the 130 cpus one; the signal integrity is also influenced by motherboards traces/shielding and planes layout, given the same specs (chipset, VRMs, etc etc) there are better boards with less noise and there are some that are just terrible (when you push them out of the stock frequency comfort zone); but as you said less memory modules are better for signal integrity, it was also true that to maintain stability with something like winbond bh5 memory at very high clocks it was common to modify the motherboard ram vrm to provide up to 3.0/3.2V to memory slots, which was a massive overvolt from 2.6V, but bh5 loved it.
Addressing these caveats was the reason why DFI LanParty boards (6 layers, better traceroute, better shielding, huge ram overvolting capabilities) were so praised by the OC community.

Re: Battle of the platforms: socket 754!
To be honest, there isn’t much difference between the Venice core and the Lancaster.
It boils down to a difference of 512 KB versus 1 MB of L2 cache.

Venice wasn't even tested thoroughly in this thread, neither was Lancaster with L2 512kB. One LinX benchmark is insufficient. You are welcome to run test suite of nd22. Skip tests that are not essential. For CPU we have PCMark 2002, 2004, 2005, SuperPI and Cinebench 2003.

We have Clawhammer vs Newcastle (different L2 size) and Clawhammer vs Lancaster (same L2 size).

shevalier wrote on 2026-04-30, 07:27:

Re: Battle of the platforms: socket 754! To be honest, there isn’t much difference between the Venice core and the Lancaster. It […]
Show full quote

Madowax wrote on 2026-04-30, 04:04:

Kruton 9000 wrote on 2026-04-29, 16:17:

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

This is not about standard 754 CPUs, these are mobiles, latest 754 core (basically single channel 939 with low leakage transistors and memory controller new stepping) IMC is revised and better than the 130 cpus one; the signal integrity is also influenced by motherboards traces/shielding and planes layout, given the same specs (chipset, VRMs, etc etc) there are better boards with less noise and there are some that are just terrible (when you push them out of the stock frequency comfort zone); but as you said less memory modules are better for signal integrity, it was also true that to maintain stability with something like winbond bh5 memory at very high clocks it was common to modify the motherboard ram vrm to provide up to 3.0/3.2V to memory slots, which was a massive overvolt from 2.6V, but bh5 loved it.
Addressing these caveats was the reason why DFI LanParty boards (6 layers, better traceroute, better shielding, huge ram overvolting capabilities) were so praised by the OC community.

Re: Battle of the platforms: socket 754!
To be honest, there isn’t much difference between the Venice core and the Lancaster.
It boils down to a difference of 512 KB versus 1 MB of L2 cache.

There is a huge difference between Clawhammer/Newcastle and Venice/Lancaster/Newark though, especially the IMC, it is on par of the SanDiego core updates and the mobiles retain the full 1MB cache, they are basically 754 Single Channel SanDiego cores 😉

Kruton 9000 wrote on 2026-04-29, 16:17:

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

Socket 754 was launched in september 2003 as a high end platform. The older socket A became the budget platform. When socket 939 launched in june of 2004, 754 became the budget platform. Amd got this policy of launching a new architecture/platform as a high end option and keeping the old one/ones as a budget option all through the 1998-2008 period.

nd22 wrote on 2026-04-30, 17:36:

Kruton 9000 wrote on 2026-04-29, 16:17:

In the mid-2000s, it was common knowledge that socket 754 processors needed to be overclocked with a single memory module to reduce the load on the memory controller.
The quality of the memory itself isn't the issue; the weak point in this case is the processors themselves. AMD positioned this platform as a budget platform, planning for socket 939, which was released much later, making socket 754 the default platform for the time, while remaining budget-oriented by design.

Socket 754 was launched in september 2003 as a high end platform. The older socket A became the budget platform. When socket 939 launched in june of 2004, 754 became the budget platform. Amd got this policy of launching a new architecture/platform as a high end option and keeping the old one/ones as a budget option all through the 1998-2008 period.

True, that's why mobiles with 754 socket are an exception, some of them were actually top tier CPUs for desktop replacement notebooks (huge 17 inches beasts, within mobile workstation market segment) so until they moved to dual channel memory notebooks they used mobile 754 CPUs binned and specialized for low power consumption and high performances. With them you have a well refined 754 CPUs in your desktop platform than those AMD produced for the desktop 754 market segment that was moved to the budget section. I know sometimes they are more difficult to use on a desktop motherboard, but it is worth the fact that they received all the core revision fixes and enhancements that AMD skipped on the 754 desktop CPUs for budget market segmentation.

About the full 1MB cache against the half 512kB cache, it is a fact that AMD considered it to scale to ~200 MHz, but that is at stock CPU frequency and FSB, when you OC the CPU, that full cache scales up as well, so at 266 fsb that scales to ~300 MHz.

A major disadvantage is mobile s754 CPUs require OCing, many people don't like doing that. At stock frequency 1.8Ghz they aren't very useful. Without rigorous testing there can be instability in certain situations and people don't want to be dealing with that. It is just easier for them to go with AM2 for Windows XP or Athlon XP/Clawhammer for Windows 98.

With mixed RAM sticks and CR1, it seems my ML-37 is stable at 2.0ghz (stock frequency) and 2.4Ghz, but not 2.2Ghz (memtest86 fails). 2nd OCZ RAM stick may be needed to get it stable at 2.2Ghz. For ML-34 it may be the same, I just never tested it as rigorously. Running CR1 on s939 with 2 memory sticks will be a lot easier than s754.

I unfortunately do not have any Radeon GPUs to offer comparable results in games with nd22 and no longer have Clawhammer 3400+ as I ended up selling it.

AlexZ wrote on 2026-05-01, 08:20:

A major disadvantage is mobile s754 CPUs require OCing, many people don't like doing that. At stock frequency 1.8Ghz they aren't very useful. Without rigorous testing there can be instability in certain situations and people don't want to be dealing with that. It is just easier for them to go with AM2 for Windows XP or Athlon XP/Clawhammer for Windows 98.

With mixed RAM sticks and CR1, it seems my ML-37 is stable at 2.0ghz (stock frequency) and 2.4Ghz, but not 2.2Ghz (memtest86 fails). 2nd OCZ RAM stick may be needed to get it stable at 2.2Ghz. For ML-34 it may be the same, I just never tested it as rigorously. Running CR1 on s939 with 2 memory sticks will be a lot easier than s754.

I unfortunately do not have any Radeon GPUs to offer comparable results in games with nd22 and no longer have Clawhammer 3400+ as I ended up selling it.

Yep yep that is true, but they are still better than a budget Palermo sempron 3000+/3100+ even at stock.

AlexZ wrote on 2026-05-01, 08:20:

A major disadvantage is mobile s754 CPUs require OCing, many people don't like doing that.

I wouldn’t have minded...
Which is how I ended up with a Turion-based rig.
I bought the Turion to replace my 1.6 GHz Sempron.
It’s certainly faster, but...
And the VIA K8T800 non_PRO doesn’t support asynchronous overclocking at all. 🙁
I had to look for a processor with a Venice core.
And for the Turion, I had to go for the nForce 4.

shevalier wrote on 2026-05-01, 08:30:

I wouldn’t have minded... Which is how I ended up with a Turion-based rig. I bought the Turion to replace my 1.6 GHz Sempron. It […]
Show full quote

AlexZ wrote on 2026-05-01, 08:20:

A major disadvantage is mobile s754 CPUs require OCing, many people don't like doing that.

I wouldn’t have minded...
Which is how I ended up with a Turion-based rig.
I bought the Turion to replace my 1.6 GHz Sempron.
It’s certainly faster, but...
And the VIA K8T800 non_PRO doesn’t support asynchronous overclocking at all. 🙁
I had to look for a processor with a Venice core.
And for the Turion, I had to go for the nForce 4.

I find nforce3 better for my use case (agp and WIN98) but NForce4 is a great platform for XP and cheaper pciE GPUs. AlexZ is right though, so please I beg you nice people: stop buying top tier boards, if you don't want to overclock, buy nice and well built budget motherboards and leave the beautiful ones to those sick with this OC disease. 😇😉😆

Madowax wrote on 2026-05-01, 08:41:

shevalier wrote on 2026-05-01, 08:30:

I wouldn’t have minded... Which is how I ended up with a Turion-based rig. I bought the Turion to replace my 1.6 GHz Sempron. It […]
Show full quote

AlexZ wrote on 2026-05-01, 08:20:

A major disadvantage is mobile s754 CPUs require OCing, many people don't like doing that.

I wouldn’t have minded...
Which is how I ended up with a Turion-based rig.
I bought the Turion to replace my 1.6 GHz Sempron.
It’s certainly faster, but...
And the VIA K8T800 non_PRO doesn’t support asynchronous overclocking at all. 🙁
I had to look for a processor with a Venice core.
And for the Turion, I had to go for the nForce 4.

I find nforce3 better for my use case (agp and WIN98) but NForce4 is a great platform for XP and cheaper pciE GPUs. AlexZ is right though, so please I beg you nice people: stop buying top tier boards, if you don't want to overclock, buy nice and well built budget motherboards and leave the beautiful ones to those sick with this OC disease. 😇😉😆

It would be hard to call the Jetway K8T8AS a pretentious overclocking motherboard 😀
What’s more, the Gigabyte GA-K8N51GMF is, in essence, a motherboard designed for office computers.
Although, following the upgrade to polymer capacitors, Turion ML-30 work excellently at an FSB frequency of 275 MHz.

Those are good choices... You are saving a lot of overclockers from their unsatisfied motherboard cravings. You are a good person. 🙂😄

Yep those Turions go up a lot when you add power stability to the mix, I still have rubycons mbz on my board.

Madowax wrote on 2026-05-01, 08:53:

That is a good choice... You are saving a lot of overclockers from their unsatisfied motherboard cravings. You are a good person. 🙂😄

In my previous post, I went into more detail about exactly how I use my Gigabyte motherboard.
Unfortunately, I’m prone to overclocking. But I don’t overclock it any higher than the top-of-the-range model in this series.
It’s not really overclocking, but rather restoring of the natural balance.

Continuation of testing at 2407Mhz CPU, 267Mhz base, 9x multiplier, HT 802 (3x), 400 memory. This time, Turion 64 MT-40 is used, but since clock speed and memory timings are the same, it is equivalent to ML-34 at the same clock speed.

See Re: Battle of the platforms: socket 754! for the beginning

I got Command Rate 1T stable by using OCZ Platinum 1GB stick in 2nd slot. This worked for both TL-37 and MT-40 at 2.4Ghz, but not 2.2Ghz. At 2.2Ghz we can only use CR2 on ML-34, ML-37 and MT-40. This shows a different CPU will not help you.

CPU runs at lower voltage 1.26V instead of 1.42V. This doesn't translate to lower temperature under load as the temperature was already very low before. MT CPUs have lower TDP and are more efficient at the same clock speed.

PCMark 2004 benchmark results

Previous benchmarks with Turion 64 at 2200Mhz:
Re: Battle of the platforms: socket 754!

We got a 9% boost over 2.2Ghz in CPU score

PCMark 2005 benchmark results

Previous benchmarks with Turion 64 at 2200Mhz:
Re: Battle of the platforms: socket 754!

We got an 8% boost over 2.2Ghz in CPU score

Far cry benchmark

Previous benchmarks with Turion 64 at 2200Mhz:
Re: Battle of the platforms: socket 754!