Laptop memory overclocking?

[superkdogg]

Hey gang,

I have a Toshiba L25-s1216. It has a mobile Celery M 1.6 and DDR2-400 stock. The Toshiba specs say that it's supposed to be DDR2-533, but even after a successful pin mod, the lappy insists on underclocking the ram which is GSkill rated for DDR2-533. So, now after a successful pin mod, I have the Celery rockin' at 2.13 and 133 FSB, but waiting on the slow DDR2.

I loaded up Clockgen, but that only appears to work on the FSB and proc. speed-not on the memory clock.

Is there a software tool out there to help me?

Failing that, what are the chances that I could successfully flash my bios with that of another Toshiba L-series that has the same chipset, but came with a Pentium-M so should have a 133 memory clock?

[STEvil]

if clockgen is working for your FSB then the ram is changing clock as well. Sure you're looking at the right tab? Can you post some screenshots?

[eToh]

Is there anyway to overclock from bios?

[superkdogg]

Is there anyway to overclock from bios?

That's a negative-it's a stock Toshiba laptop. if I had bios options, it would be easy.

if clockgen is working for your FSB then the ram is changing clock as well. Sure you're looking at the right tab? Can you post some screenshots?

When I move the sliders, the FSB moves, but ram stays the same. I'm benching some stuff to see if it changes, but CPU-Z and Sisoft Sandra say it doesn't.

*update-the ram latencies due change somewhat with the FSB moves, but it's really weird. My pin mod means that the CPU is native 133 fsb, but the memory appears to be running 100 at startup, but if I lower the FSB by a touch to say 128, the latencies go down by about 50 ms according to CPU-Z's utility.

Is it possible that the memory strap somehow responds positively (faster) to what it thinks is an underclocked FSB? I'm puzzled and there's more benchmarking being done to figure this out.

[superkdogg]

My results are pretty sketchy.

The CPU-Z latency seems to have an inverse relationship to FSB speed after 1MB (once the pattern is running in RAM rather than on the CPU cache). That's weird.

Everything else likes more FSB speed, but I can't isolate the memory speed because the benchmarks should improve with more FSB and more CPU speed.

The sliders in clockgen are working, and they definitely adjust the speed of the FSB/CPU speed, pci-e and pci buses. This is confirmed by software tools. The ram is still being read as 400 MHz by Sandra, Everest, and CPU-Z so I assume that the Ram is set at 400 in bios, not using a ratio. I kind of assumed that already, since the ram was 400 before the pin mod and 400 after the pinmod made the CPU run 533.

So, does anybody know of a software tool to adjust ram ratios? Anything that supports the XPress 200 chipset should work or is at least worth a try.

[STEvil]

ram should stay with the FSB, there is no separate PLL to drive it at speeds other than ratios of the northbridge.

What may be happening with the cpu-z latency is that chipset strap may be changing over a certain FSB. Everest read/write/latency tests are a more accurate way of testing your ram.

[superkdogg]

ram should stay with the FSB, there is no separate PLL to drive it at speeds other than ratios of the northbridge.

What may be happening with the cpu-z latency is that chipset strap may be changing over a certain FSB. Everest read/write/latency tests are a more accurate way of testing your ram.

Everest does better the higher I can get the FSB.

CPU-Z shows the ram speed as DDR2-400 asynch. regardless of what the fsb is running.


I suppose I should just be happy with a laptop Celeron M running 139x17=2363 MHz, which makes for a damned fast lappy and a hot left hand and leg!

I guess I just have two more questions:

1. Is it possible that CPU-Z, everest, and SiSoft Sandra are just getting it wrong when reading the ram speed?

2. Would it be worth it for me to undo my pin mod and overclock just using clockgen to increase the FSB in case FSB and memory are 1:1 in bios, provided FSB is 100, but when bios detects FSB 133 it changes the ratio to 4:3?


I guess #2 provides the most likely scenario for best success because I doubt the chipset is dynamically changing ram speed ratios on the fly and like you said-there's no seperate PLL for changing the ram speed independently. I'll have to run all the benchies and see if FSB 139 with the pin mod is actually slower than 139 via clockgen which will then tell me more about what the mobo and bios are doing.

Any other suggestions for me are more than welcome!

[STEvil]

1: I think you are reading the wrong info in the programs
2: starting at 100mhz FSB and overclocking to 139mhz FSB should give you a higher final memory clockspeed if dividers are being used and your ram is running 200mhz (400mhz DDR) stock.

Which screen are you reading your info from? The CPU and Memory tabs will have the info you want, the SPD tab only displays what the module is programmed to do at stock settings.

[superkdogg]

I can see why you would think that, but I'm honestly reading the info from the memory tab. It stays at DDR2-400 no matter what. The results from benchmarks change though, so I don't know where CPU-Z gets its information from but it looks to me like it just doesn't work accurately on this particular motherboard.

I ran SiSoftware Sandra's memory and cache benchmarks, figuring that it was the only tool I know of that has some credibility and also would allow me to isolate out FSB and faster L2 cache versus speeding up the memory. The larger pieces that were too big for the cache were in fact faster with the higher FSB. I'm pretty sure at this point that the memory is running at the speed it should but the tools are not accurate in reporting it's speed. I know that's at least partially true because when I OC with CG, the memory definitely responds but no matter my FSB setting all the tools I have available will report it at DDR2-400 running asynch with FSB.

Any more ideas?

[STEvil]

If everything says its faster then its faster.

To guage how much faster you might have luck with 3DMark 2001 "Dragothic Low" (Game 2 test 1) which is nearly 100% bandwidth limited. Test each of your settings (with a reboot between each test) and see which one gives you the best performance.

[superkdogg]

If everything says its faster then its faster.

-That's what I thought. I wonder why CPU-Z doens't pick up the changes, but the proof's there.

To guage how much faster you might have luck with 3DMark 2001 "Dragothic Low" (Game 2 test 1) which is nearly 100% bandwidth limited. Test each of your settings (with a reboot between each test) and see which one gives you the best performance.

* bold added by me.

Thanks for the suggestion regarding Dragothic. I'll do that as well as the Sandra and Everest testing.

It looks like I'll be removing my pin mod tonight, because a clockgen OC with a 1:1 ratio to start out with could put me right where I want to be. Right now any bandwidth gains I get are presumably starting at 100 MHz and growing with a 4:3 ratio. I'm hoping that clockgen allows for the ram to run 139 at fsb 139 as opposed to running only 104.5 MHz at 133 fsb if it really starts out at DDR400 after the pin mod.

[superkdogg]

Just to wrap this puppy up, here's what I found:

I benched some vary basic quick and dirty numbers for 3DM01 and Everest with the pin mod @ 133, underclocked back to 100 and then without the pin mod at both speeds. My goal was to figure out if the memory was using a ratio or a flat asynch setting.

While the results are somewhat ambiguous, I'm pretty sure that all my messing around was for no reason. It seems as though no matter the FSB setting, the memory is running a flat DDR-400 asynchronous with the FSB. There are some slight differences in the benchmarks, but they aren't nearly enough to explain a 35-40% change in the speed of the memory bus so I presume they are due to the CPU overclock and/or cache speed changes.

Long story short, I'm just going to leave it overclocked with Clockgen because that did give slightly better results but I'm fairly sure that my L25-s1216 memory cannot be persuaded to run its advertised speed (even with DDR2-667 ram it wouldn't run 533). Too bad because the shared memory for graphics was the big reason I even cared and it pretty much cripples the performance for all but the most basic games.

Text-based sims, Warcraft III, Starcraft, and some others still work though, so I guess I won't whine too bad.

[chris89]

Can I mod a ddr3-1333 dimm to a ddr3U-1333 stick?

The a10-5750m supports only ddr3-1866, ddr3L-1600, and ddr3u-1333... Normal = 1.5v L = 1.35v U = 1.25v

[Charles Wirth]

Typhoon burner is for flashing the SPD on the ram

You would have to manually test the settings first and make sure the ram can do the UL settings first.

Next would be learning typhoon burner to find and modify the bits you need.

There are other SPD programs/tools out there, typhoon is not free.

[Don_Dan]

Can I mod a ddr3-1333 dimm to a ddr3U-1333 stick?

The a10-5750m supports only ddr3-1866, ddr3L-1600, and ddr3u-1333... Normal = 1.5v L = 1.35v U = 1.25v

Are you sure your memory can run at 1.25V?
As far as I know the SPD doesn't contain any Vdimm information, but I might be wrong.