Hello, Just had a thought. I wonder if socket 370 boards with LinuxBIOS (i810) would support Tualatin (FC-PGA2) processors as long as the FSB was compatable. You would probibly need a FC-PGA to FC-PGA2 converter because of the voltage differences, but this would allow much faster processors, up to 1.4Ghz :-)
Thanks - Joe
most boards are incompatible because intel chose a different pinout for fcpga-ii cpus on purpose. a hardware modification (or an expensive adapter) is needed to run a tualatin cpu on such boards. the cpu voltage is one of the minor problems. anyway, we should support all processors and their microcode updates. the processor support in general needs some rework as most cpus are completely unsupported. the generic functions are mostly useless and only control general parameters (if anything at all). Holger
Hello, Just had a thought. I wonder if socket 370 boards with LinuxBIOS (i810) would support Tualatin (FC-PGA2) processors as long as the FSB was compatable. You would probibly need a FC-PGA to FC-PGA2 converter because of the voltage differences, but this would allow much faster processors, up to 1.4Ghz :-)
Thanks - Joe
Quoting popkonserve popkonserve@gmx.de:
most boards are incompatible because intel chose a different pinout for fcpga-ii cpus on purpose. a hardware modification (or an expensive adapter) is needed to run a tualatin cpu on such boards. the cpu voltage is one of the minor problems. anyway, we should support all processors and their microcode updates. the processor support in general needs some rework as most cpus are completely unsupported. the generic functions are mostly useless and only control general parameters (if anything at all). Holger
I beg to differ Holger, I have bought these on ebay cheap, and they work just fine on boards that support them.
http://cgi.ebay.com/New-370-CPU-Celeron-3-Converter-Socket-for-Tualatin_W0QQ...
Here is one for 15$
Thanks - Joe
joe@smittys.pointclark.net wrote:
I beg to differ Holger, I have bought these on ebay cheap, and they work just fine on boards that support them.
http://cgi.ebay.com/New-370-CPU-Celeron-3-Converter-Socket-for-Tualatin_W0QQ...
Here is one for 15$
Thanks - Joe
Joe - What chipsets have they been working with? They should work just fine with LinuxBIOS if we have the chipsets ported.
Back around 99-01 we used to use PGA to Slot-1 adapters. They would violate the written spec by Intel for the FSB but they worked fine at the lower 66MHz vs 100MHz speed.
-Bari
Quoting bari bari@onelabs.com:
joe@smittys.pointclark.net wrote:
I beg to differ Holger, I have bought these on ebay cheap, and they work just fine on boards that support them.
http://cgi.ebay.com/New-370-CPU-Celeron-3-Converter-Socket-for-Tualatin_W0QQ...
Here is one for 15$
Thanks - Joe
Joe - What chipsets have they been working with? They should work just fine with LinuxBIOS if we have the chipsets ported.
Back around 99-01 we used to use PGA to Slot-1 adapters. They would violate the written spec by Intel for the FSB but they worked fine at the lower 66MHz vs 100MHz speed.
-Bari
I have used them on i810e, i815, and VIA VT82C694X boards. The bios's did not detect the correct speed but the OS and benchmark software did. I could not get them to work with i810 or SIS 630, that's why it would be cool to see if we could get LB to do the job, even better. It could bring new life to these older boards.
I know what you are talking about, the slockett adapters. I still have a few of these kicking around somewhere. When they first came out they would only do a FSB of 66 or 100. Later they did come out with 133MHz ones.
Thanks - Joe
I beg to differ Holger, I have bought these on ebay cheap, and they work just fine on boards that support them.
http://cgi.ebay.com/New-370-CPU-Celeron-3-Converter-Socket-for-Tualatin_W0QQ...
Here is one for 15$
THOSE adapters only reroute some signals. they leave the vcore generation completely untouched and you can easily burn your board's regulators. that's why they are so cheap. additionally those adapters don't change voltage levels of some important signals esp. PWRGOOD. this signal needs a level of 2.0V according to intel's specs. on some boards it is just connected to vcore. with vcore way below 2.0V the cpu will never start, not even with this adapter. everyone with medicore soldering skills could rework a S370 to achieve the same within 5 minutes.
but apart from all this: the bios (not even linuxbios) can distinguish between a fcpga/fcpga-ii board and a ppga only board. only if all boards were handtested by someone. this leads me back to my first thought: all slot1 and s370 boards get support for all slot1 and s370 cpus. the bios picks the correct handling routine during runtime like current bioses do,too. Holger
Quoting popkonserve popkonserve@gmx.de:
I beg to differ Holger, I have bought these on ebay cheap, and they work just fine on boards that support them.
http://cgi.ebay.com/New-370-CPU-Celeron-3-Converter-Socket-for-Tualatin_W0QQ...
Here is one for 15$
THOSE adapters only reroute some signals. they leave the vcore generation completely untouched and you can easily burn your board's regulators.
How? I thought FC-PGA2's have a lower voltage requirment than FC-PGA's.
that's why they are so cheap. additionally those adapters don't change voltage levels of some important signals esp. PWRGOOD. this signal needs a level of 2.0V according to intel's specs. on some boards it is just connected to vcore. with vcore way below 2.0V the cpu will never start, not even with this adapter. everyone with medicore soldering skills could rework a S370 to achieve the same within 5 minutes.
but apart from all this: the bios (not even linuxbios) can distinguish between a fcpga/fcpga-ii board and a ppga only board.
That's a good thing right? This way you can do some things you normally wouldn't be able to do. Like, the wiki home page says "Various non-standard scenarios (e.g. FPGA in Opteron socket)"
only if all boards were handtested by someone. this leads me back to my first thought: all slot1 and s370 boards get support for all slot1 and s370 cpus. the bios picks the correct handling routine during runtime like current bioses do,too. Holger
Correct me if I am wrong, but isn't it ok if the LB processor code is a bit generic. All it needs to do is get Linux going and as long as the Linux kernel is able to detect the processor (vender/device id, etc) it can take over and unleash the processors full capabilities?
Thanks - Joe
THOSE adapters only reroute some signals. they leave the vcore generation completely untouched and you can easily burn your board's regulators.
How? I thought FC-PGA2's have a lower voltage requirment than FC-PGA's.
yes, lower voltage but HIGHER current. the problem is that the mosfets drivers are not capable of driving such high currents. in general the complete circuit (capacitors, inductors, diodes) is not designed for the current requirements. even IF the
but apart from all this: the bios (not even linuxbios) can distinguish between a fcpga/fcpga-ii board and a ppga only board.
That's a good thing right? This way you can do some things you normally wouldn't be able to do. Like, the wiki home page says "Various non-standard scenarios (e.g. FPGA in Opteron socket)"
well..this kinda makes things complicated. LB has to contain handling routines for all intel processors. handling can differ between models even if they are one family. a good thing is: handling routines can be divided into core handling routines ex. all coppermine processors have the same initialization and handling routine.
as i already stated the problems that Uwe had with his M6178 mainboard booting dead slow was the missing L2 cache initialization routine for the cpu he used.
LB should contain handling routines for all processors that COULD be run on a mainboard. on a VIA 694x slot 1 mainboard you could run everything from intel pentium ii 233 to intel pentium iii tualatin 1.4GHz
Correct me if I am wrong, but isn't it ok if the LB processor code is a bit generic. All it needs to do is get Linux going and as long as the Linux kernel is able to detect the processor (vender/device id, etc) it can take over and unleash the processors full capabilities?
another problem: the kernel relies on the bios setting up the processor correctly but only partly..it's a real pain. some things are done by the kernel and some things are not.
case 1: if you start the kernel with a intel pentium iii with disabled L2 cache the kernel will leave it unconfigured. you will run a processor without L2 cache then. case 2: if you start the kernel with a cyrix 6x86(MX) various cpu registers are altered that (can) affect performance (and power consumption). the same if you run an AMD K6(-2/-III).
so LB has to set up all hardware registers (in this case: the cpu registers) to correct values. we can not (and we should not) rely on the kernel to do the work. and in my understanding the kernel either shouldn't alter any of those registers or set up all registers correctly. Holger
On Fri, Nov 16, 2007 at 07:20:42PM +0100, popkonserve wrote:
but apart from all this: the bios (not even linuxbios) can distinguish between a fcpga/fcpga-ii board and a ppga only board.
That's a good thing right? This way you can do some things you normally wouldn't be able to do. Like, the wiki home page says "Various non-standard scenarios (e.g. FPGA in Opteron socket)"
well..this kinda makes things complicated. LB has to contain handling routines for all intel processors. handling can differ between models even if they are one family. a good thing is: handling routines can be divided into core handling routines ex. all coppermine processors have the same initialization and handling routine.
Yep.
as i already stated the problems that Uwe had with his M6178 mainboard booting dead slow was the missing L2 cache initialization routine for the cpu he used.
Fortunately that's only the smaller of two problems. The main problem was my fuckup in Config.lb which lead to the CPU init not being done at all, thus the slow-down. This is fixed now for all 440BX boards and the MS-6178.
But yes, enabling L2 should be done, too.
LB should contain handling routines for all processors that COULD be run on a mainboard. on a VIA 694x slot 1 mainboard you could run everything from intel pentium ii 233 to intel pentium iii tualatin 1.4GHz
Ack.
Correct me if I am wrong, but isn't it ok if the LB processor code is a bit generic. All it needs to do is get Linux going and as long as the Linux kernel is able to detect the processor (vender/device id, etc) it can take over and unleash the processors full capabilities?
another problem: the kernel relies on the bios setting up the processor correctly but only partly..it's a real pain. some things are done by the kernel and some things are not.
case 1: if you start the kernel with a intel pentium iii with disabled L2 cache the kernel will leave it unconfigured. you will run a processor without L2 cache then.
Yes. It's even worse, the kernel sometimes uses the fact whether the L2 cache is enabled or not to determine the CPU type. Turn L2 on/off and it'll report a different CPU in /proc/cpuinfo.
case 2: if you start the kernel with a cyrix 6x86(MX) various cpu registers are altered that (can) affect performance (and power consumption). the same if you run an AMD K6(-2/-III).
so LB has to set up all hardware registers (in this case: the cpu registers) to correct values. we can not (and we should not) rely on the kernel to do the work. and in my understanding the kernel either shouldn't alter any of those registers or set up all registers correctly.
Agreed.
Uwe.
-
bari -
joe＠smittys.pointclark.net -
popkonserve -
Uwe Hermann