M4800 Owner's Thread

Ok guys, just noticed something SUPER WIERD.

When I first boot the machine, performance is as above, good (especially with the undervolt slightly raising the clock speeds) but not class-leading, but also stable and not throttling below base clock so all good; it's not bad performance and it seemingly NEVER drops below the promised base clock, whilst keeping itself relatively cool and quiet. I was doing a bit of testing and tinkering ie turning off BD PROCHOT which didn't seem to change anything.

However, then I thought back to another Dell Haswell machine I had played with (a Latitude E6540, with more quirks, faults and issues than this M4800, that's for sure; that one had to go back, but same generation of Dell Machines), and I decided to try something I discovered with that. I sleeped the machine. And then woke it up. That's it.

Suddenly, the M4800/4900MQ realises that 4core turbo is up to 3.6GHz, and now throwing an all-core load sees that clock speed bouncing between 3.05 and a full 3.6GHz, often floating around 3.2-3.5GHz; and so far never going below 3GHz, fans are ramping up harder too so you can tell the machine is kicking in more fully. Throttlestop, if I monitor a simple CPU-Z Bench run, now indicates a maximum package power of 48W compared to the 40W previously. (Up to 58W if I'm testing everything and using the integrated Intel GPU also)

CPU-Z Multithread Benchmark goes from ~1850 (which is already up by about 100 points from the undervolt, from around 1750), to ~2000. Single core remains roughly the same, in the 400-410 mark, perhaps unsurprising as the machine is only using around 30W during this single thread stress test.

This is repeatable with a shutdown/reboot. (When I next boot the machine it will go back into the 'optimized' mode, with the sleep and wake bringing the machine to full power).

What the...

If I had to hypothesise, it is almost as though when the machine first boots, it does not load 'Dell's Ultra Performance' profile; instead it loads the default 'Optimized profile'; still good performance, but with lower power consumption and noise, and a capped multicore speed due to the lower wattage allowances.
When you sleep and wake the machine, the 'Ultra Performance' profile kicks in, with higher power limits; and therefore higher heat and fan curves. Possibly related to the age of the machine and being designed for Windows 7 and not Windows 10. Dell Command Power Manager was also last updated around 5 years ago, so wouldn't surprise me if Windows 10 treats something differently in terms of ACPI/Power profiles compared to Windows 7/8, and is overruling or not loading correctly the right profile on first boot, and only loads it properly after the sleep/wake cycle refresh.

Now, if this proves to be stable, this is actually not necessarily a terrible thing, as the machine is for my father in law, so means he can get a fast, and relatively quiet machine as standard, with 'beast mode' engaged if he toggles the machine in and out of sleep.
But it would be nice to know if there is a way to 'cure' the need for a sleep and have the machine load the appropriate, full performance profile on first boot.

Has anyone seen this 'feature' before; or even better managed to cure it?
Most of the performance issues I've seen have been due to old, failing batteries, or too low wattage power adaptors; I have the full 180W PSU, and a relatively new, excellent health rated battery (its not official, but Dell's utilities think it is, and that health is excellent).

 

Well, following the above, and revelation regarding the power limits (Optimised and Performance), I've now been able to make good use of this, and the overclockable turbo bins in Throttlestop to make practical use of this 'feature'. Albeit it'd still be nice to getting power modes working properly with boot, as the saying goes when life gives you lemons, make lemonade

I've had to reduce the undervolt to around 15mv to ensure stability, but have managed to get the 4900MQ running nicely on +3 turbo bins OC'ed it seems. This now means the Single Threaded clocks are now hitting as high as 4-4.1GHz in some cases, and the in that initial load 'Optimized' state and power caps, CPU-Z Bench score has increased to ~425-435 Single Thread hitting around 3.95GHz, Multithread still around 1750-1800 due to the power limitations. Performance under extreme load still settles at 2.8GHz with more moderate loads going around the 3-3.2GHz region as before due to the power cap.

When I sleep and wake her to unleash the beast mode however, all hell breaks loose and she is unleashed
CPU-Z bench Multithreaded score now in the 2090-2150 region, Single Core around 440-445!

Extreme load in this performance/power state (ie for example Passmarks Burntest with all tests runnning simultaneously and at 100% cycle) tends to have the CPU running around 3.05-3.6GHz all core, with the average a bit higher than before, with more moderate all core load for example CPU-z multithread bench, hitting 3.6-3.8GHz, so there is definately a performance boost offered by those turbo bin increases, even if there is still potentially a heat/power cap on this (albeit REALLY extreme loads like linpack or ASUS Realbench in stresstest mode testing everything push her as low as 2.8 base clock due to power and heat generation but that is ABNORMAL level load, not even extreme, and she never goes below base clock, ie proper/unacceptable throttling.
Machine is still passing stress tests and not crashing, albeit I intend to run more longer tests to make sure she is still stable with so much extra performance and higher average clocks on tap

Not at all unhappy with this result though if its stable. There are heat and power limits as its an older chip, but at its best, this is RIGHT on top of considerably newer high end Intel laptop chips and even some desktop high performance chips.
More single core performance for the balanced mode, more raw performance for the higher power mode when pure power and heat limits don't become the bottleneck. Excellent!

I do need to recheck later to make sure it is thermal or turbo limits, but it now feels like like I am hitting good old fashioned thermal or proper turbo limits with the likes of Linpack and Realbench stress, both of which are extreme and hit the system harder than realworld usage, rather than more annoying and arbitrary initial power limits set lower than the system and CPU can take.

Now just to get the faster RAM in, and maybe a CPU repaste to increase thermal transfer with some fresh MX4. Could also test +1/+2 bins and a higher undervolt to see if this increases longer term performance without too much short term detriment; higher voltage and higher clocks both increase power draw and heat afterall... +1 with roughly the same undervolt as stock multis could be a very nice balance. Tweaking is a slippery slope!

Either way with or without the overclock, performance is now about where I expected it to be with the silly power limits not the main issue. For what it's worth in CPU-Z bench this is now only about 10% slower than the desktop I gave him, which was an i7-2700k clocked close to 5GHz on all cores, still no slouch even by modern standards and not at all unimpressive that this 6 year old laptop is so close.

But yes if anyone has encountered that power cap on boot which is resolved by sleep and worked out how to solve it, let me know!

 

Well I think I've settled on +2 bins for 1 core, +3 bins for dual core (aka both set to 4GHZ), with 3 and 4 core set to the default 3.6GHz. Helps stop heat and power consumption getting out of hand, whilst still offering a nice bump to single and dual threaded tasks and snappiness, and retaining a 50mv undervolt to help with long term performance at the long term power allowances.

Running the likes of Linpack it does get quite toasty even with the downclocking due to power and heat, but crucially it's not crashing. I did consider opening her up and repasting the CPU, but having watched a few videos it seems a bit of a ball ache and as it's not needed right now, I'd rather avoid the risk of accidentally damaging one of the tiny connectors until she starts getting too hot in general usage.

CPUZ bench scores now 2015 approx multithread, 435 approx single thread. Pretty decent.

MSata Samsung SSD arrived and transferred the Win10 install to it, as although the 2.5" was slightly nippier in some loads, it turned out the Samsung kicked the snot out of it on longer loads etc once the cache ran out and was all round more consistent, especially with a larger load. Either way the Samsung was also slightly larger and that's never a bad thing for an OS drive. Dual SSD so pretty sure the father in law will not have speed complaints due to storage!

Also ended up ordering the RAM from Taiwan via eBay from a well rated seller as couldn't find any good value stuff in the UK, so got 16GB (2x8) of 1866 CL11 HyperX on the way. I would have preferred CL10 but it was quite literally twice the price. As more stuff can take advantage of higher mem speeds these days, 1866 CL11 seemed more prudent than 1600 CL9, especially for the same price, and 16GB of 1866 CL11 will be better than 8GB 1600 CL11 that's currently in here anyday.

Turning into quite a solid machine now, and as I got it for a swansong, should work out just over £300 total. Given M4800s often go 400+ on eBay without the dual SSDs, Wireless AC Intel 7260, faster ram or 16GB of the stuff, pretty happy and father in law is looking forward to getting hold of it.

She will now handle 4k60 YouTube vp9 video footage with only a small smattering of dropped frames (ie less than 1%, 75 in 12517 frames in a video I just watched, so closer to 0.5% and at 60fps one frame now and then is a much smaller deal than at 30 or less), so I am pretty happy with that. 4k30 is a breeze, and if she can handle that any other general purpose computing should be a breeze. Hell the faster, lower latency ram might even help with that video decoding as going by task manager the load is partially CPU partially Intel integrated GPU and we all know anything involving integrated graphics loves bandwidth and lower latency.

 

Ok it's passed hours of stress tests with no issues and various different core loads. Then I got a blue screen whea uncorrectable. Given I was pushing +3 bins on 2 cores AND an undervolt of -50, I reckon I'm close but just a little unlucky, so Ive dropped my UV slightly to -45 and dropped my dual core back to +2 bins Vs +3. Between the two changes hoping I've caught the cause of the WHEA UNCORRECTABLE.

Usually load variation is what tends to bring out the worst after all

 

Well managed to get 16GB of HyperX CL11 1866MHz.
Disappointingly it's running at 1600 CL11 rather than CL9 (or full speed 1866 CL11)
HX318LS9IB (X2)


Anyone got any good tips for tweaking the memory timings on this machine, by rewriting spds or similar or another tweak? Still trying to resolve with the seller incase this particular RAM is off as notice people have had a lot of success with HyperX.

Would really like to get this stuff running with lower latency or full speed, albeit the main thing is the capacity is working and it's stable. Need to investigate the SPD profiles.
Father in law is very happy though with the machine overall.

 

Ugh. I suspect the HyperX is either faulty, mislabelled or offspec.
Whilst the labelling is 1866/PC14200 (yes I realise this should be 14900 really but seems 14200 is a known 'thing' when it comes to 1866) as read by various bits of software, when I dig into the SPD profiles with HWinfo, CPU-Z and Thaiphoon, there are no proper RAM profiles above 1600 CL11, CPU-Z shows the highest RAM profile as 838MHz/1672.

I've checked with Thaiphoon, HWinfo and CPU-Z, all read the NAME/LABEL of the RAM as 1866, BUT the SPD tables go up to ~1600MHZ CL11.
There are no 1600 CL9 OR 1866 CL11 SPD profiles (or XMP profiles) as you'd expect from performance RAM.

The RAM labels as HX318LS9IB also, and all Kingston's spec lists seem to show HX318LS11IB as being the correct part (with HX316LS9IB being the 1600MHz, not the strange mix of the first part and second...); so again I am tempted to suspect these are maybe seconds/offspec/fakes that have had the header profiles changed.

Every genuine HyperX kit I've worked with has had profiles for all rated speeds, so I am a bit suspicious these could be fakes or failed bins, that have just had the labels reprogrammed and are in essence 1600 CL11 kits. Manufacturing date is listed as 2018 also, so too recent for it to be before XMP/higher SPD profiles became normal and again expected that 2018 SODIMMs with higher spec speed profiles would have those correctly programmed into the RAM.

The seller SO FAR has at least responding and trying to see if we can get this working as he also seems to believe it is 1866MHz, so I have sent him pictures of the proof of the missing higher SPD profiles and waiting for his response. He is based in Taiwan so likely tomorrow. As much of the RAM is produced out there, I've got no reason to suspect it is a deliberate con, but more reason to suspect they might be seconds.


PS. Sorry if it seems like I'm monologuing but good to get this off my chest and no way I would put this all in one post. It'd be too bloody long

 

m4900mq only supports up to 1600mhz
https://ark.intel.com/content/www/u...4900mq-processor-8m-cache-up-to-3-80-ghz.html

Even looking at the 4940mx i have it still says 1600, i dunno how other people are getting their systems to run with 1800 as this was a very limited option .

according to this document it's limited to 4x4gb 1866 to enable it to run in this speed, which must mean bios/firmware has strict control over memory channels, like we've seen before, dell firmware is a really advanced piece of engineering

https://www.dell.com/support/articl...-mobile-m4800-m6800-portable-computer?lang=en

 

Yes I am aware the 4900MQ only officially supports 1600MHz. To be fair, I only expected it to run at 1600MHz and not at 1866MHz (that would have been a bonus). What I DIDN'T expect, is for the RAM to be missing all SPD profiles above 1600CL11, which makes it virtually useless even for machines that will support higher unless they have manual memory speed and timings control. Even when you cannot run the RAM at the higher speed, from prior experience, the RAM would normally have SPD/XMP profiles present for 1866 CL11, and usually 1600 CL9, which you'd be able to view if you inspected the tables at the very least, even if you couldn't manually select or use them.

It's the fact that this RAM's idents do not tie up with those published by Kingston, and the SPD tables are missing any 1866CL11 or 1600CL9 profiles, or XMP profiles, that has me suspicious. I have a strong suspicion its 1600 CL11 memory (as the Samsung memory modules on these SODIMMs appear very similar and almost identical timings to my other 1600 Samsung memory), with the EPROM reported names edited and fake stickers/idents slapped on. Or its factory seconds that never passed higher qualification, and never had the higher profiles added.

 

Well i had a HP Probook 6470b, i tried some friend's 2133mhz and it was working fine (i was running a 3540m at that time, so 1600mhz max according to intel). It would be weird that a hm76 chipset and a 3rd gen i7 could run 2133mhz, but a qm87 and a 4th gen i7 couldn't.

(both the CPU and Ram aren't on the supported list for the 6470b)

 

Yes pretty much confirmed the RAM is fake and returning. Seems there is a growing market for fake Kingston with very similar modus operandi.
Have some stuff that looks more genuine coming from locally.