Possibly Overclocking an Unclockable 3820TG.

Overclocking the Unclockable CPUs?

ORIGINAL SECTION MODIFIED - PLEASE READ.

I have modified this section from its original form, since there were a number of questions raised about the conclusion reached by Darth NVader thread on determining whether or not a PLL could be OC'd or if it needed a pin mod. http://forum.notebookreview.com/har.../429717-rtm875t-606-info-gathering-ocing.html. Accroding to this thread, a pin mod is only needed if a PLL shows a "1" in Byte 09h, bit 6. As you can see from the discussion below, there was some confusion about the location of Byte 09h, bit 6. After some debate, the consensus is that the location as shown in the Darth NVader thread is correct. The 09h, bit 6 is shown here.



If you have a "0" in the second slot from the left, your unit is not TME bound and likely OC'able. Once you know your unit can be OC'd the trick is to find the right clockgen. For my unit, the first clockgen that looked promissing was SLI505Y clockgen. The first one I selected turned out to be too unstable and limited.


Through trial and error, I found another SLI505Y clockgen that worked much better.




Follow Up:

This thread was started as a place to have an OC'ing discussion. When I looked back at the OC posts to make this thread, it looks like the Asus G73H clockgen has the Ram/PCIE/PCI Freq settings at 330.7/100.0/33.3mhz. The stock settings for the SLG505YC264BT are 266/100.0/33.3 mhz, so it is almost the same except the ram timing is low and probably not in a proper ratio with the base clocks. It should be 133.3/330.7. Instead it is 133.3/226.7. That is the only flaw with this clockgen that I can see. Nevertheless, it can still get a stable OC, albiet, not at the 3.2ghz speeds that Hendrickson is seeing from his 450M. But, this should be easy to adjust for the program creator, if he is so inclined. With this set-up, in order for the Ram to reach a proper speed, the CPU is running faster than it should and probably fails earlier. Thus, I still think that this clockgen is inherently unstable (maybe unstable is a bad word - it is more "performance limited" due to the improper ratio between CPU/Ram on upperslide), but it can yield decent results and holds-out hope for the future for these PLLs.

Here's a question ... can the clockgen definitions be easily changed by using the SetFSB editor?

Edit: Yuniu tells us that it is easy but still trial-and-error. http://forum.notebookreview.com/ace...info-benchmarks-tweaks-mods-upgrades-464.html. I have not had the time to test this out yet. IF he's right, the original SLG505YC264BT that I was using can be edited to bring the ram timing up into a better ratio with the CPU speed. This should allow higher OC'ing. It should also let me eidt the SLG505YC56DT clockgen to give me more room on the upside. Maybe I can equal Hendrickson's 3.4-3.5ghz if I can effectively edit the clockgens.

Bronsky

 

amazing................................................

 

Bronsky i set up the same clock generator u used on the pictures but i wants to know how do you get the clocks to start from 133 to 160? instead of starting at 133 like usual?

 

When you use "get FSB" you can move the slides. Use the top slide arrows. There are 2 settings boxes, one says "Current" and one says "Select". You will see the settings move on the bottom as you move the slide. Once it is at 160.3, you can hit SetFSB. It then changes the "current" setting. That is what the pictures depict. If this works for you ... please let us know.

Bronsky

 

Bronsky when i open setfsb version 2.3.149.115 and i put the two clock generators mentionnned on your post the [set fsb] button is always greyed out i can press [get fsb] without it giving me a pll error but i cannot change the fsb with the button set fsb i tried moving the top arrow as well but still nothing...

anything else you want me to try and have u gotten a reply for the program creator?

https://mail.google.com/mail/?ui=2&...ttid=0.1&disp=inline&realattid=f_gdfy0k170&zw link to the pic

 

When you run PLL diagnosis as in the top picture do you have a 0 in the correct spot or a 1?

You can "Get FSB" in a number of clockgens without the error? When you use SLG505YC256BT, you get a greyed slider that you can't move? Your picture link didn't work for me. I need to see what is happening there. Could you put it on a host site and link it.

If the SLG does not work for you, one of the others might. Did you try the SLG clockgen definition above the 505YC256BT? It is really a trial and error process at that point.

Bronsky

@louissasha - look at your bit 6 again. I had the photo wrong. Could you take a picture of PLL Diagnosis and your 09h registry entry? If it's 0, I would think about trying different clockgens to see if any of them might work.

 

It might be that setFSB lacks the correct clockgen to work properly with those 3820TG models that can't use g73JH clockgen. So when selecting an clockgen that is close, you can adjust it, but it's not working properly.

No idea really, even I am at loss here

 

hmm..... so if anyone have a 1 in that spot instead of a 0 then a Pin mod is a must ??? or that is only for your notebook ?

 

According to the post I linked to above by DarthVader, if Byte 09h, bit 6 = 1 you need a hardware mod to OC with SetFSB. No idea if that is correct or not, but I have a 0 there and my unit turned out to be OC'able, so it worked for me.

Bronsky

 

hmm..... is that "6" a typo because in ur post and the linked one it is bit "7" that is highlighted ?

 

Could you be more explicit?

Bronsky

 

hmm.... actually both ur 6th & 7th bits are 0s so brother you are more than safe .... the thing is both you and the original post highlighted the 7th which in my case is 1 but on the other hand my 6th is 0 "by counting from right to left" thats why I wanted to make sure before I go on donating to a program and then find out that I can't use it to overclock

 

Thanks.

Bronsky

 

lol ..... so it is 6 after all

 

Yes ... 6 it is. Thanks.

 

here is the image i think it should work now



Uploaded with ImageShack.us



Uploaded with ImageShack.us

http://img843.imageshack.us/img843/3676/mmhz.png

looks like i have the 0 at the right place but none of the pll in the seem to help me with the [set fsb] button its always greyed out....

let me know what you think i should do thanks for the help

 

Your data looks identical to mine so far. You have a 0 in the 6th bit of 09h. Your unit is not TME bound. If you can find a clockgen that works, you are not prevented from OC'ing by your PLL.

Okay ... let's see what happens when you select the Clockgen. Select the SLG505YC256BT Clockgen, hit Get FSB and take another pictur eso we can see what we have. Post it here.

 

Hey Bronsky,

I just take a look at this thread

Just to your first post. The tool setfsb is old, very old from the technical aspect.
All the frequencies right of your 133,3 MHz are only a part of the history. A long time ago you were able to generate for each a clock cpu/ram/(agp)/pci.

But with the new bclk all of this is gone. You are only generating the blck and the gmch and ich generate the subclocks for all the other parts.

The technical point was, to get the development of the motherboard a little bit cheaper and to remove the problem with the clock skew.

 

So I don't have to worry about all of those other settings or I am using an old definition that won't make the automatic adjustments? The CPU/Ram ratio seems fixed. If I don't raise the CPU speed to a level where the ram timing is correct before I hit Set FSB, it freezes? It's perplexing.

Anyway, since my new board is comming next week, I only have so much time to play with this .

Bronsky

 

I think I'm starting to figure out this Overclocking thing on my machine.



When it became apparent that I would not be able to get a reasonably priced motherboard replacement, I decided to go back and see if I can solve the OC'ing problems I was having with my 3820TG. After a lot of fiddling with different clockgens and settings, so far I have been able to successfully set FSB to 2951.0mhz which resulted in an actual speed of 3.28ghz as recorded on CPU-Z. Temps were rock solid at 91C running WPrime 1024 for over an hour of stepped OC'ing. The best part is that I still had more room. The OC was incredibly stable. The temperature stayed constant at 91C without the slightest hint of throttling. What an amazing design.

BTW, I was using a different clockgen than before. I had switched to SLG505YC56DT. I'm also starting to get the hang of using the two slides. You have to alternate them and move them incrementally. I should thank Shark Technologies for sending me the wrong motherboard.

Bronsky

EDIT: At maximum settings for this clockgen, my 450M clocks at just under 3.3ghz and is very stable. I'll run some longer tests when I have the time. At least I know that the I5-540M I'm putting in here should OC at least as well. The 450M gained .7 GHZ or 27%. If the 540M clocks as well as the 450M with this clockgen at max settings, the 540M should attain speeds of 3.8-3.9ghz. I guess we'll get to see what an OEM CPU can do.

 

I'm here!

I've sent a PM to myOnlySt4r, thinking that he may have been the fellow who OC'd a staples unit...

Without OCing I am hitting 3.06GHz with cpu temperatures at 78 celcius in pc wizard

 

Okay. The first thing to do is load all three SLG505Y clockgens and post their settings so I can see them. Print Screen > Paint > save as JPG. That way I can compare them to mine and see what we can to do to adjust them. You should also post a baseline CPUZ printout of the I5-540M.

BTW, you will need CPUZ, CPUZ hardware monitor, WPrime, as well as the SetFSB program. If you're going to be running WPrime to test your OC'ing steps, make sure you monitor temps.

Bronsky

 

Funny thing is this: I get a pll error for all three slg505y clockgens.

The ICL9LPRS919BKL looks like it might work...



Then again, maybe not... the following settings resulted in a PLL byte error when clicking set fsb



*edit* I just found another ICL clockgen that looked like it may be a contender but Unfortunately I get pll byte errors when I try to setfsb on these ICL clockgens...

Could the 3820TG-3022 varient be that different??

Heading out to the art gallery soon. Will experiment more this evening.

 

Every 3820TG can have a radically different clock gen from another. Even two units in the same model series can have different clockgens. It's random.

Actually, your bottom screen does not look that different than mine. I have a larger value in the FSB slot but I think this might work. You probably need to feed in a lot more lower slide to get the MHZ values in the lower right hand box closer to 3ghz. I think this is workable.

First, switch to control. Then, try starting with the Set FSB values at around 160 and see if the unit freezes with the lower slide set where it is. Then feed in 50 units of set FSB to one or two clicks on the lower slide. See if feeding the lower slide values increases the MHZ speed settings on the right. Let me know how it goes. My OC has a PCIE value of 180 I think. Looking at the ratios among the settings, I would suspect you need a PCI-E setting of around 150. Your PCIE frequency is correct at 33.3. One of the tricks is to keep the stepped increases from changing that frequency.

It looks to me that you can probably max the upper slide and control speed on the lower one ... but we're not there yet.

Bronsky

 

I get a pll byte error when I try to set the fsb with this clockgen. I'm heading back to working my way through the list.

 

LOL... more waiting...

I decided to do a clean install last might. Turns out that the Machine ID (license) is tied to the machine's SID.

Have to wait for another key from Abo.

 

OK, Editing the Hex values of the control registers has been a revalation! I have been using ICS9LPRS355BGLF as the clockgen, and so far I have set 0x00 to 29, 0x0f is left at D8 and I can get 0x10 to 34 and stable, getting the i5-540M up to 2679.1 MHz.

Core temps running wprime 1024 are 82/84 celsius
Exhaust temps at the vent is 61 deg C.

Screen shots to come soon.

The Intel Core i5 turbo monitor desktop gadget is showing 2.85 GHz with the meter just shy of 1/2 way, and the CPUZ is showing 2961.3MHz for the core speed while WPrime runs. Score is 645.311 seconds for WPrime 1024.

 

I'm starting to play with the registery as well. I want to see how high I can get this chip before I reaplace it.

 

Managed to hit 3.2GHz in turbo without touching 0x0f

 

So far playing with the 0x00, 0x0f, 0x10 bytes the best I can do for base CPU clock (non-turbo) is 2742MHz or, about 8.3% over the factory clock of 2530MHz

That's set to 29, D7, and 33 respectively

 

That is the clockgen that gives me a pll error no matter what I do.

Bronsky

 

Clearly there are different chips being used in various models.

We should try to keep a record of which models like which clockgen until one or more of us get inside a system and identfy through observation what chip is being used.

In my 3820TG-3022 (Staples.ca) I am using the ICS9LPRS355BGLF clockgen in SetFSB

 

OK, The best I have been able to do so far is 2689.9 MHz on the base clock. This is a 6.4% increase in base clock speed, which equates to a turbo speed of about 3.22GHZ

Here's my most recent run for score.





Is expecting more from this CPU unreasonable? I still have room left on the thermal side of things...

I am hoping that when I am able to supply a correct clockgen ID to abo, he will be able to make a SetFSB configuration that allows greater granularity in the settings.

Also, I have noticed that WPrime doesn't get the processor running in full-bore Turbo mode (note the intel gadget)

 

I don't know yet. I'll let you know as soon as I install my I5-540M. Is yours an engineering sample or OEM. Hendrickson had an ES that did not OC very well at all. I bought an OEM to see if I can get a better OC. Your speeds are about the same as my I5-450M OC'd but we're not sure if that is due to the incredible OC'ability of the 450M or not. Hendrickson had a similar result with his ES I5-540M.

I have been busy at work, etc., but as soon as I get some time, hopefully in the next few days, I will swap in my new CPU. Since I already have a clokgen that I can max out, it should be a straightforward comparison between the two CPUs.

Did you increase the multiplier to get x21?

Bronsky

 

My 540M is an OEM unit.

I'm not sure if I changed the multiplier or not. The only changes made were to the byte codes at 00x00, 00x0f and 10x00 as shown in the screenshots earlier in this thread...

Much like you I would love to a manual that correlated the registers to the names of the variables being changed (multiplier, base clock, etc.)

 

Have done a little bit of research. You can download a manual for this Clockgen here: 9LPRS365 (ICS) - Search -----> ICS9LPRS365 (9LPRS365.pdf) then scroll down and klick on locate datasheet. When you have the document opened search for Byte 0 FS Readback and read downwards.

It looks that byte 01 19 20 (PLL3) are required for some fine tuning, but further investigation needed. I played a little bit with it without concrete results. Could also be that i'm wrong with this idea.

 

This may be a really stupid question but ...

if I want to set a lower value for a registry value of BB, is the lower value BA? Is the next lower value AF? And so on?

Bronsky

 

The BB A0 etc are hexidecimal bytes (8 bit words)

They go

A0
A1
A2
A3
A4
A5
A6
A7
A8
A9
AA
AB
AC
AD
AE
AF
B0
B1
B2
B3
B4
B5
B6
B7
B8
B9
BA
BB
BC
BD
BE
BF

etc...

 

Thanks ... that was what I was doing wrong.

 

This is my most stable clockgen so far. It's a 26% increase from the base clocks. I don't think you're expecting too much. Hendrickson had a hard time with hiis 540M too but he has en ES, not an OEM. Maybe there is something wrong with the first gen of Arrandales and the Madison chips OC far better. Mine (540M OEM) is going in this weekend. I wanted to find a good base line to compare the two CPUs.

How come you can't feed in more clock speed by moving the slide bar to the right? See how yours are all the way over to the left and mine are all the way over to the right? Does your system freeze or something?

Bronsky

 

so far my results have been acheived through direct register manipulation. The sliders and the current/select fsb/ddr/pcie/pci value have no resembalance to reality so I am ignoring them until I know exactly which clockgen chip is really in there.

 

My 540M is in and seems to OC quite well. So far, I've been able to run it with base clocks OC'd in the 2.9 ghz range and turbo to 3.5-3.6 ghz. I have not had the chance to do any stability testing. I've had base clocks OC'd to 3.1ghz but get a memory dump blue screen. I'm trying to figure out the cause of that. OC'ers forum suggests that I have some driver remnant issues. I think my ram is being out-performed.

I also put a switch in my backlit KB that I am not happy with. As soon as I fix that, I will get back to this OC issue. I'm mad at myself for trying to do too many things at one time and not getting the Clockgen ID'd. Oh well!

I am, however, really happy with the Temps from the IC Diamond paste.

Bronsky

 

Bronsky,

I know a lot of guys are protective of tyhis info, but would it be possible for you to tell me what tools and settings/tweaks you're using to get 2.9GHz? The best I can manage is 2.7 and it's unstable.

Also, what stepping is your 540M?

Thanks,

Tom

 

LOL I'm an open book. I am using the same clockgen that I used on the 450M. I'll try to post the settings I was using as soon as I have some time. I'm going to have to reinstall the backlighting switch I installed and then will get back to OC'ing.

 

Back to the clockgen question. I have the same clockgen as Kindan does.

ICS VS3197BL



I've emailed IDT asking for a datasheet.

 

Mine is an SLG8LV595V. While I had it apart to finish the switch, I checked it out.

Bronsky

 

That explains a lot.

Hopefully IDT will be forthcoming with the datasheet...

 

Yea - does explain a lot. That is probably why my OC works so well. I seemed to have found the right clockgens by trial and error.

Hey ... that is some real impressive work on the motherboard soldering.

 

Thanks,

It takes good light, patience, fine solder, a fine soldering iron, NO COFFEE and lots of flux.

My eyes aren't what they used to be.

IDT has come through with the datasheet. I have passed it along to Abo-san with an offer of another donation.

 

That must be why I'm no good at it ... I have no patience, my lighting is bad, my equipment is more suited to woodburning than soldering and I drink wayyyyy to much coffee.

Bronsky