I think I'm going to use my Sempron 64 2800+ as a donor of the heat spreader.

I always chose the processors with the larger cache back in the day if the prices was close. I felt like I was getting better performance per watt.

I usually got one or two speed grades down from the fastest because the price and heat was often better.

After the hot age of humming fans when the choices where Palomino vs P4, I followed this upgrade path:

• Athlon XP "Barton" -- Mounting a heatsink was like balancing a bowling ball on a shot glass though
• Athlon 64 "San Diego" -- Did great in Half Life 2 & CounterStrike Source, but many of those early 939 boards had short life spans
• Opteron "Denmark" -- Was available at a good price in 2007 and held its own for quite a while. Didn't have the PREFETCHW instruction, so it is 32 bit only for >= Win 8.1
• Core Duo E8400 "Wolfdale" -- My dad still runs this system today

Those were all great chips

marxveix wrote on 2025-01-28, 12:51:

Huge price, i skip at the moment. 😀

Yeah, I was about to comment on that... $65 for an old single-core 64-bit s754 CPU? No thanks! 🤣 Not worth it, I think.

I noticed all CPUs for 754 are much more expensive than the 939 versions in general. I paid $50 for mine.

At that price, you might as well get a whole motherboard + CPU combo or some laptop that has the CPU you want in it.
But I guess the problem is the seller(s) usually don't list that kind of information, especially on scrap/old hardware.

Two years ago, I put a bid on eBay on a FIC K8M-something s754 motherboard. It was listed for parts/not working and that's all that was mentioned in the description. Through the pictures, though, I saw that it came with an Athlon 64 3700+ CPU included, so I knew it was a good deal. I managed to win the auction easily for like $18 total shipped to my door. Unfortunately, the seller had a goof-up and sent me a completely different motherboard by mistake (a Via-based s478 P4) and the motherboard I won was sent to someone else... which is a shame, because the FIC board I purchased was riddled with bad/burst caps all over it. It would have been a super easy fix/restore if I got it. And the CPU would have been a nice addition to my collection. Sadly, I didn't get lucky this time. Instead, I now have yet another socket 478 motherboard that's not really anything special. On the plus side, it does work... and didn't leave me bored in terms of having nothing wrong with it. One of the mounts for the NB heatsink had pulled out and there was also one bulged cap that needed replacing. Easy-peasy. I will eventually do a full recap on it. Turns out, the board does unofficially support P4 Prescott CPUs, but only with 533 MHz FSB IIRC... which is perfect, as I have a 2.8 GHz Prescott with 533 MHz FSB that's doing nothing.

Is a ML-44 the same as a 3700+ ClawHammer in terms of speed? Maybe a 3700 is an option.

I've compared specs of those two. Besides ML-44 supporting SSE3 and being much more efficient, the performance should be the same.
Both CPUs are hard to find and very expensive compared to other models.

zuldan wrote on 2025-01-28, 19:37:

Is a ML-44 the same as a 3700+ ClawHammer in terms of speed? Maybe a 3700 is an option.

Just go with much cheaper and available 3400 😀.
SSE3 came later with 90nm cores.

I still believe the 3200+ Clawhammer has the best current performance/price ratio.

Sometimes its possible to find these cpus with cheaper prices if you by broken laptop and another time asked price for the cpu is just better.
Turion64 is great and Mobile Semprons are ok and they dont have ML (35w) and MT (25w) versions, they are 25w or similar to desktop 62w.

Mobile Sempron 3400+ 2GHz 256kb 25w
https://www.cpu-world.com/CPUs/K8/AMD-Mobile% … 3400BQX3LF.html

I know I am very late to the discussion, but since this topic comes up pretty often during my online research on 754, I though I'd share some input.

First of all, 130nm CPUs have inferior memory controller compared to 90nm ones, but that applies actually to the performance. At the same settings, 130nm has up to 10ns longer memory latency and that can lead to few percentage differences in performance. Obviously, every usecase is different, but in my testing, Venice is faster than Clawhammer, even though it has less L2 cache.

Another misconception relates to improved IMC on AR revision Clawhammers. The only difference is support for 2T timing option on AR revision, which should be avoided anyway as it's detrimental on memory write performance. The IMC itself is poor anyway and I had better memory results even on NF2 boards.

Clawhammer IMC isn't that bad after tuning memory timings. I'm using Command Rate 2T as I'm using two modules.

Memory latency wise it's better than what I achieved on Phenom II or Vishera.