Pegatron Ipmel-AE has me stumped

fincarre · 6 Jul 2020 at 20:32

Hello - this is my first post about my first attempt to upgrade and as I read quite a lot on this site and have now become interested in the technicalities of overclocking, I wondered whether someone could advise as to why my experimental el cheapo upgrade doesn't work please?

The motherboard is this one:
https://support.hp.com/gb-en/document/c01925513 and is circa 2009-ish.

It came with a Celeron E3200 2.4GHz 800MHz chip and I wanted to upgrade to a 1333MHz system with 8GB RAM so step one was to order 8GB of new 1333MHz returnable RAM as without that working, the exercise was a non-starter. Despite what HP says, the 8GB RAM was recognised, albeit downclocked to 800MHz and so I moved to the second stage which was to purchase a Q9650 quadcore 3GHz 1333MHz per HP's upgrade possibilities. This wouldn't work with the 8GB RAM (no POST, long double beep) but would work with the 3GB 1333MHz RAM that came with the computer originally - I don't recall whether that RAM ran at 1333MHz or less but I think it was 1333MHz.
I then wondered whether the 1066MHz DIMM limit was a problem so purchased 8GB of new 1066MHz returnable RAM - but same problem, no POST, double beep.
A local experienced computer tech suggested the quads sometimes didn't work in certain combinations so I then went next best (for me) and bought a returnable E8600 3.33GHz 1333MHz Core 2 Duo - same problem again - no POST, double beep with 8GB but worked fine with 3GB.
Then wondered whether the alleged max limit on the DIMM's - 1066MHz -was a problem in combination with the CPU so then bought a returnable E7600 3.06GHz 1066MHz so that the CPU matched the DIMM limit and again, with 8GB no POST, double beep but worked fine with 3GB RAM.
I just had enough thermal paste left to put the Celeron E3200 back in and am currently running it with 8GB 1333MHz RAM in it downclocked to 800MHz - as happened with the 1066MHz RAM too (I checked).

I was on the verge of calling it a day and then read about high and low density RAM chips and there doesn't seem to be a way of identifying either - and this idea started to suck me in until I remembered that whether high or low, the 8GB was working with the E3200 Celeron - thus suggesting the DIMM slots are reading the chips, irrespective of what they are.

So here lies the mystery - if the motherboard recognises the 8GB RAM of both speeds with the Celeron E3200 then why not with every other chip that is on the HP list of potential upgrades? I started to wonder about base clock multipliers and my eyes glazed over and so decided to ask you nice people

I have a few days left before I have to return everything but really would like to get the Q9650 running with 8GB's RAM - what am I missing please?

Thank you

Journey · 6 Jul 2020 at 21:16

You need BIOS version 5.07 or newer to get the Q9650 working, and that board only support 4GB RAM maximum.

fincarre · 6 Jul 2020 at 23:38

Thank you - my current BIOS is 5.07 dated 26 Jan 2010 and the Q9650 was released Q3'08, the E8600 Q3'08 and the E7600 Q2'09 so I assumed the BIOS would be OK for at least the Q9650 to begin and the others subsequently. I did check for an update at the time but HP didn't have an obvious one.

When the Q9650 was installed, I ran CPU-Z and on the Memory tab, it stated that DRAM Frequency was 531.9MHz and FSB:RAM was 5:8 - whatever that is.

Whilst the spec's state that only 4GB RAM will run, I am running 8GB right now with the E3200 at 800MHz - I am using it to type this.

Tetras · 7 Jul 2020 at 14:08

I don't know if this has anything to do with it, or not, but in several generations now, the Celerons are available in servers and they can have superior memory support to other desktop CPUs. It's also newer than the quad cores, so might have been updated for high density memory? I'd be confused if that was the reason though, because I thought that memory support was dependent on the chipset for 775 CPUs.

I guess it could be that it can only support this memory at 800. According to a post I saw by a Corsair rep, you need 128MB memory chips and possibly 256MB, but 512MB will not work.

https://forum.corsair.com/forums/showthread.php?p=552430

fincarre · 8 Jul 2020 at 03:57

Thank you and interesting as I now note that the humble E3200 Celeron is the most recent CPU, albeit by a month as compared to the E7600 but a year younger than the Q9650 and E8600. For those interested in double data rate and matching pairs of RAM modules, I just read this interesting article whilst trying to find out about memory controllers: https://en.wikipedia.org/wiki/Memory_controller

Your linked article was very interesting and I too wondered about 128MB chips. However, I think I have found the answer here:

https://www.intel.com/content/dam/www/public/us/en/documents/product-briefs/g41-chipset-brief.pdf

'Up to 17 GB/s (8.5 GB/s per channel with DDR3 1066 MHz) bandwidth and 4 GB supported memory; up to 12.8 GB/s (6.4 GB/s per channel with DDR2 800 MHz) bandwidth and 8 GB supported memory.'

Although I installed 8GB of both 1333MHz and 1066MHz DDR3 memory, both were downclocked to 800MHz because the E3200 Celeron runs at 800MHz and so it looks like the chipset then acts as though I have got 8GB's of 800MHz memory (and seemingly ignoring the DDR2 vs DDR3 difference) - which the flyer states is supported. Then reading https://www.transcend-info.com/Support/FAQ-296 I started to understand how the differences between DDR2 and DDR3 are all about squashing more data down the tubes per clock cycle of the computer, of which the G41 chipset is a main part. Reading transcends flyer, I can see that this motherboard (more specifically, its chipset) is designed to transfer data at the upper range of DDR2 (viz. 400 x 2 = 800MHz) but the lower range of DDR3 (viz. 533 x 2 = 1066MHz) which explains why the HP stats state that only 800 or 1066Mhz memory can be used. The intel flyer quoted above shows that the G41 chipset is actually transferring less data per second with more DDR2 memory (8GB) as opposed to more data per second with less DDR3 memory (4GB) and that the G41 chipset is thus designed only to allow a maximum throughflow of data within the parameters of 12.8 to 17GB/s TOTAL bandwidth. I therefore assume that 8GB of DDR3 RAM of either 1066 or 1333MHz is simply going to try to squash more data down the tubes than the tubes have been designed to carry.

Assuming this mathematical limitation is absolute because it is linking computer and RAM clock cycles which cannot be changed, I assume that the question of high versus low density memory chips is moot.

It seems like my buying an experimental 8GB RAM and testing it with the Celeron E3200 first before buying the Q9650 caused me to assume that 8GB would work accross the board when clearly as above, I was lacking an understanding of how everything worked together which buying extra CPU's wasn't going to fix because the problem wasn't the CPU.

My current experience is of having the Celeron E3200 paired with 8GB 1066GHz RAM downclocked to 800MHz and results in using only 3.7GB RAM but almost maxing the CPU (mid 90% usage at 2.20GHz speed out of 2.4GHz max) when used with Win 10 Pro on an encrypted disk doing encrypted transfers to a cloud backup. I thought I had a memory problem but I have a CPU problem.

So as I only have 3GB RAM other then the two sets of 8GB (which is being returned), I am going to buy 4GB of 1066MHz DDR3 RAM and refit the Q9650 (with 12M L2 cache and 4 cores - I have a 95 TDP fan) and run the encryption on Win 7 instead (I have dual boot) and limit the number of Firefox tabs I have open and then watch how it performs. This is old hat to you maybe but as I have now established a benchmark with this machine in terms of understading, I am curious to know whether it can be made to perform better with the Q9650 and only 4GB RAM transferring at the higher data transfer rate than with the Celeron E3200 and 8GB RAM transferring at the lower data transfer rate - and yes, I will try both Win 7 and Win 10 for the hell of it but Win 10 will surely max out the memory, methinks.

Unless somebody is going to come up with anything else, this puzzle now seems to have a coherent explanation - but thanks to everyone who contributed

fincarre · 19 Jul 2020 at 17:11

Here is an update for anybody who is interested in this particular motherboard. I ordered 4GB of 1066MHz RAM and due to supplier mislabelling, one stick of 1066 and one stick of 1333MHz arrived so I combined their 1333MHz stick with my own single 1333MHz stick and can now confirm that the motherboard downclocked both to 1066MHz - so the DIMM slot limit is the limit = 1066MHz.

I refitted the Q9650 Quad and ran it with both Win 7 Ultimate and Win 10 Pro. Surprisingly, Win 10 manages the 4GB memory better than Win 7 and I have done various experiments with my eyes glued to task manager graphs. I configured the Pagefile to 16GB (the max. for 4GB RAM) and with all three of my disks encrypted (the OS is on an SSD) and using encrypted cloud backup and using memory-hungry Firefox, the computer runs without a struggle. I even had just over 60 Firefox tabs open once (admittedly with just text thereon) and the CPU has never hit 2.8GHz in speed, I have seen its load go up to just over 90% momentarily on occasion when opening programs but it is usually in the 7-40% load range and have seen the Pagefile using around 6GB. Overall, where I previously expected to have to be satisfied with staying with Win 7, I am now permanently migrated to Win 10 Pro with four screens and no problems. The original Intel fan (bought separately off ebay new and unused but stored for 11 years) ) for this 95TDP chip keeps the CPU around 45 deg C average accross 4 cores most of the time and when I open a few programs more and give it some load, I have seen it go up to around 66 deg C but no more (so the original 65TDP fan would have been no good!).

I have returned all the unused memory and CPU's and have obtained refunds and have learned a hell of a lot about hardware combinations whereas previously I had only ever loaded Windows of various versions and fitted various internal cards when that was required. This computer is still 11 years out of date even with the upgrade but this exercise has established a benchmark in my mind for where I was, where I am and where i could be in the future if I were to get the itch and start investigating how to build a really modern computer. I prefer to learn and understand rather than throw money at something just because I can. That said, I have also learned that computers are not like tinkering with old cars - with computers, I now appreciate that there comes a time when rather than tinkering, the technology becomes so obsolete that serious consideration has to be given to how worthwhile it is to upgrade rather than replace. The Q9650 was a faster chip in its day but CPU benchmark sites show that modern CPU's leave it standing and then some - so the question is, what do you want to achieve? It seems to me that the faster computers get, the more advanced gaming becomes and thus the one feeds off the other - but if one is not gaming, then this constant advance seems unnecessary.

Bottom line, for ordinary computer use this motherboard with this chip will run Win 10 Pro on 4GB 1066MHz RAM with 8-16GB Pagefile and manage it nicely. Even when physical RAM use was up to 3.8GB out of a possible 4 GB, there was no visible slow down. It plays HD YouTubes fine and runs 4 screens of graphs via an nVidia NVS510 GPU (which is for PCIe 2.0 but this board is v 1.0) which is actually all I wanted it to do. When I tried to do this on the Celeron E3200 800MHz chip with 8GB RAM, the CPU was running in excess of 90% load a lot of the time and I could tell that the chip just wasn't 'man enough' for the job thus the Q9650 on 4GB RAM is a better choice that an 800MHz CPU with 8GB RAM.

Tetras · 19 Jul 2020 at 19:39

It sounds like it was ultimately a worthwhile experience, despite being a lot of faffing about!

There's a good bit of benefit to tinkering around with old stuff, for sure.

Quite surprised to hear that 4GB is adequate for Windows 10, though I suspect the pagefile being on an SSD has something to do with that.

Dual cores seem to be reaching the end of the line now, even for desktop use, or at least the slow ones are.

Journey · 19 Jul 2020 at 19:54

I use Windows 10 on a Core Duo laptop with 2.5GB of RAM, runs fine for managing VM's etc.

fincarre · 26 Jul 2020 at 19:33

It was a very worthwhile exercise from a learning point of view and I am still using the Quad as my everyday computer with Win 10 Pro on SSD. When I switch to Win 7 loaded onto a 1TB 7200 rpm HDD on the same computer, the speed difference is stark to the point where I start to question whether something is wrong but when I inspect task manager, I can see that several times there is no memory left if I have a few Firefox tabs open so I have to shut a few down. Win 7 never stops and works fine but is significantly slower than the Win 10 setup. Whether Win 7 would work better on an SSD I don't know, I wanted to keep it off the SSD away from Win 10 as a backup OS in case Win 10 or the OS failed one day as I have all my data on two HDD's.

I also discovered the reality of bottlenecks - the Celeron E3200 was the bottleneck when it came to computing power even with 8GB RAM and having cured that by changing to the QuadCore with Win 10 memory management, I now notice in task manager that my old 2TB Seagate HDD with 32MB cache becomes the bottleneck when uploading to an encrypted Cloud - it running at 100% continuously when the QuadCore is operating at less that 50% and the RAM having adequate spare capacity. At other times the upload limit of 18Mbps on my copper wire internet connection is stuck permanently at 18Mbps thus becoming the bottleneck whilst the CPU, RAM and Seagate HDD all have spare capacity so it was educational to gain a reality on how the whole system slows to the speed of whichever bottleneck becomes relevant to whichever operation I am doing at that particular time and that bottlenecks are not necessarily caused by the same item every time. This was quite a 'wow' moment of realisation for me - I have read it so many times in various articles and understood it conceptually but to have graphs showing it actually happening on a system which I have tweaked myself as much as I could, opened up a new facet to my understanding of how the concept actually worked practically in a system to stop me achieving a performance that other components might otherwise allow me to achieve if it were not for that particular bottleneck.