Zombie Mods - The I9 Steambox

[ALXAndy]

Hit it with a hammer, you are in the UK so you understand jeremy clarkson logic, trust me im an engineer, in tf2.

 

We call it the Irish Screwdriver

[alpenwasser]

Hmmm. Well I've been doing some research this morning and I think I know why it doesn't work.

 

Basically a Xeon has two QPI links. In the 980x the second QPI link is disabled. In the Xeon the second QPI link is active. This is the *only* difference between the two chips.

I think ECC memory support is the second difference, but yeah the chips are

basically the same silicon. This is also why you can't use regular CPUs on

the SR-2. If you put two regular ones onto the board, they won't be able to

communicate with everything they need due to their lack of a second QPI link.

Interestingly enough, if you put only one LGA1377 i7 onto the board, its QPI

link will be pointing to the second CPU socket instead of the board, which is

why you can't use an i7 even in single-socket mode.

So, how do Xeons run on regular X58 boards? well the answer to that is simple - those boards have a bios that disables the second QPI link on the CPU 'soft mod'. There was, at one point, a hard mod for a specific EVGA board. EVGA soon released a bios but people were hard modding their boards to disable the second QPI link.

 

So in theory the same mod (though obviously not the same logically) could be performed on any board, because there is a circuit for the QPI link and it's usually on show on the board itself. You just short a connection.

 

Now I'm not saying that there's any hope tbh. If the Alienware board had a removable BIOS chip then I could possibly hack the bios itself (I've done this before many years ago). However that's not going to happen really.

 

What I am wondering is if the CPU itself could be shorted in any way or form to disable QPI link 2.

You could try insulating either the pins on the socket or the pads on the

CPU? Put a bit of lacquer or something on it?

[ALXAndy]

You could try insulating either the pins on the socket or the pads on the

CPU? Put a bit of lacquer or something on it?

 

Indeed. If I can find out what ones to short I could easily do that tbh. I'm pretty sure that this is why once I exit bios the system does not work.

[ALXAndy]

/extremely technical mode. Warning - geeks only !!!111oneoneeleven.

 

OK. So I have been doing some research. It seems that Intel have technical data available for their sockets. You can find it here.

 

http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/core-i7-900-ee-and-desktop-processor-series-datasheet-vol-1.pdf

 

OK. Now from what I can gather Intel call the pins inside a LGA 'lands'. I don't know if they mean lands as in like, geographical lands or whether they're calling a spade an island (IE - it only makes sense to them). Now I've been studying the document above and from what I can see 'lands' are pin outs on LGA. So for example if you look at this screen grab here -

 

 

You can see that it lists the 'lands' and what they do. It also tells you their state. So digressing away from QPI for a moment we can see how the socket approaches 'DDR2'. Now to explain a little, DDR2 is not DDR2 as in DDR2 memory type. It is DDR2 as in the second lane of memory out of six.

 

Now. If we take a look at this picture here.

 

 

We can clearly see the 'lands' for the QPI. Now if I can find out which of these lands need to be shorted to disable the second QPI link then I could, in theory, disable it.

 

I'm not sure how many of you guys have done any research into bending pins in the socket (it happens a whole lot) but basically if you bend a pin and it shorts in the socket of an Intel board strange things happen. As an explanation; a couple of years back I bought a Foxconn Flamingblade GTI and built my system. Now I put in 3gb of memory over three DIMMS. This should have meant triple channel. However, when I booted the machine one of the DIMM slots was showing as empty. The thing is, I had the ram in. When I got to Windows it said (in the system information/rating etc) that I had 3gb of memory but only 2gb was usable.

 

Now this had my mind bent in all sorts of directions and left me completely and utterly perplexed.

 

So why did that happen? well basically it all worked perfectly, sans the missing 1gb of ram and triple channel. I used the rig for about a year like that and aside from the odd reset at boot (it would post, then reset and post again then work) it was fine. Then I bought an Alienware Aurora chassis with a board in it and when I removed the CPU I noticed that the socket had three bent pins that were touching one another. I got out my magnifying glass and a needle and moved them back into position.

 

A few months later I found a cheap I7 920 and decided to build up a spare rig for sale. Guess what? all 3gb showed up and I had no more boot issues.

 

So the moral of that story? if you short pins inside a LGA socket you will lose functionality. What does that have to do with the issue I am having here? everything, basically. It basically means if I find the correct 'lands' for the second QPI link I can, in theory (whether or not I would actually risk it remains to be seen !) cover those pins preventing them from being detected and thus, force the CPU onto a single QPI output.

 

I've seen pins bent in other areas and actually caused a short that caused the socket to burn. I would imagine these pins carried the voltage to the CPU..

 

/technical headfuck mode off.

 

I will continue, unabated, to do my research

[ALXAndy]

Hmm. I've also noticed that there are two QPIs in the land listings. One is called DRX, the other DTX. Now if I can work out which is which I should, again, theoretically, be able to short one and disable it. I would imagine this is done internally on a 980x so it can't be hacked back into life, turning it into a Xeon.

[ALXAndy]

 

EUREKA !!

 

I was absolutely correct. There are a piss load of 'lands' in the socket for power (see above, bent pins, shorted and burnt)

 

All I need now is a diagram with the code land numbers that point to what pin is what number.

[ERSCreeper]

This is getting interesting

[LukaP]

 

 

EUREKA !!

 

I was absolutely correct. There are a piss load of 'lands' in the socket for power (see above, bent pins, shorted and burnt)

 

All I need now is a diagram with the code land numbers that point to what pin is what number.

did you figure wether to short the DTX or the DGX?

[ALXAndy]

This is getting interesting

 

LOL check this out...

 

 

I can now work out what every single pin in a 1366 socket does. I can also work out now what the pin that I believed to have snapped off did, and if it would affect the PC from booting. It could work out that my board did, once upon a time (before that single pin was damaged) boot perfectly with the Xeon in. If not, and it does not affect it then I can now work out where the QPI pins reside.

 

And this could effectively lead to being able to mod the CPU to force it onto one QPI.

[LukaP]

 

LOL check this out...

 

 

I can now work out what every single pin in a 1366 socket does. I can also work out now what the pin that I believed to have snapped off did, and if it would affect the PC from booting. It could work out that my board did, once upon a time (before that single pin was damaged) boot perfectly with the Xeon in. If not, and it does not affect it then I can now work out where the QPI pins reside.

 

And this could effectively lead to being able to mod the CPU to force it onto one QPI.

 

oh yes this is still going places

[ALXAndy]

All it would need is a very small piece of insulation tape.....

[LukaP]

All it would need is a very small piece of insulation tape.....

what about (idk how its called in english, but its used on thin copper wires when they are used in coils. they are coated in it.

 

[brownninja97]

what about (idk how its called in english, but its used on thin copper wires when they are used in coils. they are coated in it.

 

the brits call it a coil, nothing fancy.

[LukaP]

the brits call it a coil, nothing fancy.

yeah, but the copper wires are insulated with a material, its really thin, similar to what women use for their nails, and its so damn annoying as every single wire i buy here is coated in it, and i have to grind it down

[brownninja97]

yeah, but the copper wires are insulated with a material, its really thin, similar to what women use for their nails, and its so damn annoying as every single wire i buy here is coated in it, and i have to grind it down

But then again, this wouldnt be as fun if it was easy right.

[ALXAndy]

the brits call it a coil, nothing fancy.

 

It's called Kapton mate.However, insulation tape would work perfectly and wouldn't damage anything or be permanent. It really is as simple as just blocking the second QPI on the CPU from connecting to the motherboard.

 

IE - for a Xeon to work in a X58 single socket board with one QPI you have a couple of choices.

 

1. Find a board that has been coded (in the bios) to support the Xeon. All this bios does is disable the lands on the socket in some sort of 'soft' mode.

 

2. Find a single socket X58 board that has a bios that supports the 980x, then physically disable the second QPI either by modding the motherboard (IE the EVGA mod) or modding the CPU or socket itself to disable QPI2.

 

Hilariously EVGA saw this mod (I have a pic if you're interested) and then just updated the bios code to support the Xeon, making the mod pointless and obsolete.

[brownninja97]

It's called Kapton mate.However, insulation tape would work perfectly and wouldn't damage anything or be permanent. It really is as simple as just blocking the second QPI on the CPU from connecting to the motherboard.

 

IE - for a Xeon to work in a X58 single socket board with one QPI you have a couple of choices.

 

1. Find a board that has been coded (in the bios) to support the Xeon. All this bios does is disable the lands on the socket in some sort of 'soft' mode.

 

2. Find a single socket X58 board that has a bios that supports the 980x, then physically disable the second QPI either by modding the motherboard (IE the EVGA mod) or modding the CPU or socket itself to disable QPI2.

 

Hilariously EVGA saw this mod (I have a pic if you're interested) and then just updated the bios code to support the Xeon, making the mod pointless and obsolete.

My design technology teacher lied to me, 3 or 4 years ago, he was a tw** anyways, well then thank you for teaching me something i didnt know

[ALXAndy]

OK so I found this link..

 

http://www.qdpma.com/systemarchitecture/systemarchitecture_qpi.html

 

Look at this picture..

 

 

Now the only thing left to work out is which of the two I need to disable. IE - unscramble DRX and DTX.

[The_Strict_Nein]

This went from freaking awesome build to a thesis on intel processors. I love you you beautiful man

[ALXAndy]

OK guys so an update.

 

What I didn't realise was that I was reading from a 900 Extreme series (so 980x 990x) CPU schematic. Thus, I went off to find a Xeon one. I used the 5000 series Xeon schematic as that's pretty much what I have. A hexa cored, Westmere-EP.

 

And I started to look for the QPI pin outs. Basically I was wrong. DTX and DRX are indeed two different types of QPI labels. However, one is input and one is output. Blocking any of these off would likely result in the CPU not working or, being terrible unstable.

 

But.....

 

I then took screen grabs of the sections of the schematic where the QPI pin outs were listed. And guess what? The 980x uses exactly 75 pins in the socket for QPI. The Xeon uses 170 pins in the socket for QPI.

 

As such all I (in theory) need to do is print out two empty socket schematics and colour in the circles where the QPIs make contact. Then do the same for the Xeon and then overlay the two to see what's going on.

 

It's going to take many, many hours. I also need to look at the states of these pin outs (0 or 1, binary for on and off). Then I need to try and match the Xeon as closely as I can. What I do know is that there is a pissload of 'reserved' pins in the 980 socket......

[Paolo Verwey]

OK guys so an update.

What I didn't realise was that I was reading from a 900 Extreme series (so 980x 990x) CPU schematic. Thus, I went off to find a Xeon one. I used the 5000 series Xeon schematic as that's pretty much what I have. A hexa cored, Westmere-EP.

And I started to look for the QPI pin outs. Basically I was wrong. DTX and DRX are indeed two different types of QPI labels. However, one is input and one is output. Blocking any of these off would likely result in the CPU not working or, being terrible unstable.

But.....

I then took screen grabs of the sections of the schematic where the QPI pin outs were listed. And guess what? The 980x uses exactly 75 pins in the socket for QPI. The Xeon uses 170 pins in the socket for QPI.

As such all I (in theory) need to do is print out two empty socket schematics and colour in the circles where the QPIs make contact. Then do the same for the Xeon and then overlay the two to see what's going on.

It's going to take many, many hours. I also need to look at the states of these pin outs (0 or 1, binary for on and off). Then I need to try and match the Xeon as closely as I can. What I do know is that there is a pissload of 'reserved' pins in the 980 socket......

Well there goes your sunday

[The_Strict_Nein]

Also, this really shows the lengths you can go to and how much you can learn when you set yourself to a limited budget. A lesser man would have just brought a new CPU by now

[ALXAndy]

Also, this really shows the lengths you can go to and how much you can learn when you set yourself to a limited budget. A lesser man would have just brought a new CPU by now

 

I probably will end up doing just that tbh. Thing is, I'd like to try if I can to get the Xeon running. It's not a cheap or low end CPU here, we're talking 32nm six cores and 12 threads. It's a massively capable CPU. Otherwise I wouldn't bother..

 

I just finished mapping out the QPI use of a 980x in the socket. That's the easy one, next it's onto the Xeon lol.

[ALXAndy]

Holy crap..... OK so look at this.

 

 

Those are all of the QPI pin outs for the 980x. Now let's take a look at the Xeon...

 

 

OK now let me explain what I have done here. Basically I have taken the Xeon picture, laid it over the 980x one and then blended it. Now. Note how I've written 980x in DARK green. This is because that is the red showing through. Then note the light green.

 

 

OK so what's really weird here? well, look back at the 980x pin out diagram. Note that it does not use a single pin out from the bottom of the socket. The Xeon, however, does. BUT ! it does not use any single different pin from the top area than the 980x. In fact, when over laid? every single pin in the top area of the board is dark green. Not a single one changes to any other colour which indicates that both CPUs use EXACTLY the same pin outs from the top area of the socket.

 

The upshot? well if I were to start covering pins they would ALL come from the bottom. So the difference between the 980x and the Xeon are simply that the Xeon uses the pins from the bottom of the socket.

[Chris is awesome]

This is really bloody interesting. Please continue.