theamtrakvirus wrote on 2023-03-17, 02:16:

I have a pair of P3X 833mhz cpus in my Dell Precision 820 system
SIV shows that it is a Cascades P3X

There is no such CPU as a P3X 833, see https://wiki2.org/en/List_of_Intel_Xeon_microprocessors, and I suspect it's not an 820, see https://en.wikipedia.org/wiki/Dell_Precision# … wer_form_factor

What is the multiplier? Post the SIV initial screen, the Menu->Hardware->CPUID->CPU-0 and Menu->Hardware->CPU Detail->CPU Timers screens so I can check what is going on with all the clocks.

Whoops, that was two typos I totally missed, my bad. Yes, they're 866 mhz cpus and its a Precision 620. Here are those pics.

theamtrakvirus wrote on 2023-03-18, 04:33:

866 mhz cpus and its a Precision 620

Thank you for the screen shots, but I never expected them to be taken when Windows 98SE was running as it only supports one of the two CPUs! Just compare what was reported to what would be when running XP.

As they are from 98SE almost all the information I needed to check is absent, can you boot Windows NT 4.00, 2000, XP or 2003 and post new ones ideally using SIV32X.exe?

Maybe Menu->Machine->D + W Benchmark will give a clue to the actual speed of the CPUs by comparing the numbers to what I see on my 866 P3s.

Yeah I should have swapped os' for that, I have an sd-ide adapter mounted in the front I use to switch between 2000 and 98se. I'll switch it over to 2000 tomorrow and test it that way, though I don't think it'll show much different. I don't think its reporting incorrectly, it really doesn't feel much faster than my Katmai P3 500 that's currently sitting next to it here at my desk. Maybe even a bit slower at times.

theamtrakvirus wrote on 2023-03-18, 09:20:

I'll switch it over to 2000 tomorrow, though I don't think it'll show much different

I would not be that surprised if CPU-0 is @ 533 MHz with CPU-1 @ 866 MHz

OK and as it's 2000 rather than XP/2003 you will have to use SIV32L so we need to compare the SIV32L numbers which as you can see are different.

I just checked and I don't have SIV save files from a Dell 620 so from Windows 2000 please will you do Menu->File->Save Local to generate a set and post the two files?

I am wondering what the Intel utility will report so have attached fidenu08.exe 27-Apr-2000 + fidenu17.exe 20-Oct-2001, I am hoping at least one will run OK, does one?

Here is I believe everything you asked for. My hunch was right, neither CPU is running at its correct speed and its not a reporting error as it benchmarks significantly slower than yours. For a laugh I'm gonna do the same later to my Katmai system and see if my theory about it being faster really is right. The only thing that I couldn't provide was the Intel utility results, neither version seems to detect the cpus. I wonder if its because they're not running at the right speed they're not being detected properly?

Wait a minute, aren't Xeons totally unlocked multiplier wise? So they're just set to x4 either by DIP switches, jumpers or in BIOS?

theamtrakvirus wrote on 2023-03-19, 02:38:

neither version seems to detect the cpus.

Without seeing screen shots I can't sensibly comment. I would need to see both the initial screen and the CPUID data.

Looking at the screen shots I noted:

1. For MSR 0000002A the Cur 0C should be 0F, 0C means x2 which is why the BIOS reports 266 MHz which is 2 x 133 MHz
2. For MSR 0000002A the Max 0F is x6.5 which is as it should be
3. For MSR 00000017 the Multiplier 0F is the speed the CPU is specified to run at it's x6.5 which is as it should be
4. The 1.73GHz comes from 532 MHz * 6.5 / 2
5. The TSC is running at 532 MHz when it should be 266 MHz, this is why the FSB is reported as 266 MHz and the memory 800 MHz
6. I am sure the TSC is @ 532 MHz as there is no dilation on [CPU Timers]
7. Given 525 / 878 * 866 = 518 clearly the CPU is running @ 533 MHz and I suspect the multiplier is x4 even though it's reported as x2 and the BIOS thinks it's x2

The bottom line is that the CPU seems to be reporting x2, is x4 when it should be at x6.5. Given it's reporting 0C rather than 0F then I guess two pins are in the incorrect state, but I don't know enough about how the processor and motherboard communicate the multiplier ratio to use to know which pins these are.

In the BIOS is there a screen that allows the CPU speed to be set?

When I looked in the save file at [SMB Bus] I noted that [ 0_63 ] + [ 0_64 ] have very high peak times so SIV should not try and read them. I have adjusted the attached SIV32L 5.70 Dell-00 test SIV to skip them so use this rather than SIV 5.69. It should also report the CPU and MB temperatures, does it?

SIV was unable to read the RIMM SPD, do any programs report the SPD and if so which? I expect there is an SMBus multiplexor as the system supports eight RIMMs and wonder if the SPD would be reported if you moved them from RIMM2_A to RIMM3_B or RIMM4_B. Were it my system I would install all 8 RIMMs, then if the SPD for 4 was reported start toggling GPIOs in the hope one controlled the mux.

And how come the fsb report as 266? AFAIK there's no such thing as a double pumped fsb on P6 architecture?

H3nrik V! wrote on 2023-03-19, 09:16:

So they're just set to x4 either by DIP switches, jumpers or in BIOS?

How did you come up with x4 ? AFAIK there are no jumpers to set the ratio

red-ray wrote on 2023-03-19, 09:18:

The TSC is running at 532 MHz when it should be 266 MHz, this is why the FSB is reported as 266 MHz and the memory 800 MHz
I am sure the TSC is @ 532 MHz as there is no dilation on [CPU Timers]

H3nrik V! wrote on 2023-03-19, 09:20:

And how come the fsb report as 266? AFAIK there's no such thing as a double pumped fsb on P6 architecture?

Before posting you need to read what has been specified.

red-ray wrote on 2023-03-19, 09:23:

H3nrik V! wrote on 2023-03-19, 09:16:

So they're just set to x4 either by DIP switches, jumpers or in BIOS?

How did you come up with x4 ? AFAIK there are no jumpers to set the ratio

I guessed from 533 and 133 fsb 😀

H3nrik V! wrote on 2023-03-19, 11:22:

red-ray wrote on 2023-03-19, 09:23:

H3nrik V! wrote on 2023-03-19, 09:16:

So they're just set to x4 either by DIP switches, jumpers or in BIOS?

How did you come up with x4 ? AFAIK there are no jumpers to set the ratio

I guessed from 533 and 133 fsb 😀

Guessing when the actual multiplier is reported via MSR 0000002A bits 22->25 is pointless and at best misleading

I'll have to look at the cpu cards and see if there's something wrong, the previous owner did any funny business with it (this computer was a recycling rescue, so no clue as to what its previous life was) or if the pins just need cleaning. I'm hoping its the latter. From what I can tell there's no jumpers for the cpu settings on the board anywhere.

To answer your questions, there is a 'compatible' setting for the cpu(s) in the bios, however from what I gather all that does is disable the L2 cache and nothing more. I'm going off what Memtest86 v4.3.7's info says because I tried booting Windows 2000 like that and it was making no progress after about five minutes. Unfortunately this system's ram configuration is a bit... odd. Putting rimms in slot 3 on each ram riser is as far as I can tell, from what I had seen about the 'acceptible' ram configurations an illegal configuration for the most part, so it probably wouldn't boot with that. I would load it up further but my selection of rimms is very limited compared to pretty much every other common ram type unfortuately. I think the only way I could get rimms in slot 3 exclusively to work is if I had either 512mb or 1gb rimms which I don't remember which it wanted like that. I've only got a couple of rdram systems in my collection and I know one does not have 512mb rimms I could steal out of it for this and the only other one I have that may actually is in storage for the time being. Plus I'm not sure if it will accept non ecc memory which that other one would be. The Intel utility just reported it was unable to recognize the processor and showed only some basic information of which we seem to already have all of it here, however I will still add a screenshot of it. Only the later version of the two worked. I do now get temperatures with the version of SIV you posted there, I don't know if you want me to go back and re take all the screenshots you previously wanted again with that info now available. Unfortunately I can't try just a single cpu card either, I believe it would require a terminator card which I do not have nor probably would be able to find.

Also, if anyone is curious I did run the benchmark on my 500mhz Katmai system... 515 for integer and 288 floating. Just about within margin of error for interger so no wonder it feels so slow..

theamtrakvirus wrote on 2023-03-20, 08:21:

I don't know if you want me to go back and re take all the screenshots you previously wanted again

Good luck with finding the issue and I don't really need screen shots. Using the SIV save files to run SIV in test/debug mode I can see what they look like.

With the Intel utility it's as I was expecting apart from Packaging N/A which on my system is S.E.C.C./S.E.C.C.2. I have attached two later versions, maybe they will do better if they run on 2000. I wonder if the Intel FIDDRV driver fails to run on Windows 2000, looking at Menu->Windows->Services->Driver Control does FIDDRV show up for a second or two on start-up or when you select the other CPU?

OK about the RIMM possibilities, I had forgotten CRIMMs might be needed, do you have any? I suspect CRIMM + 256 + 64/128/256/512 + CRIMM would work.

After looking at Menu->Hardware->H/W Detail->I/O GPIO there are 6 GP Outputs that might control an SMBus mux. I suspect GPIO23 + GPIO25 are the more likely so if you are feeling adventurous/brave you could try toggling them and see if Menu->System->SMB Bus reports any RIMM SPD @ [ 0_50 ] to [ 0_57 ] or if the Xeon PIR Data goes away. To toggle the GPIO state click on the blob. I would run two instances of SIV, one showing the [GPIO of Port 0880] panel and the other the [SMB Bus] panel.

What SIV32L 5.69 reported is a classic case of Garbage In -> Garbage Out and after pondering I am wondering if SIV should tidy up the garbage, what do you think please?

The basic issue seems to be than any multiplier < x4 results in x4 and as you can see below SIV32L 5.70 Tidy-00 compensates for this.

What do you see when you run SIV32L 5.70 Tidy-00 on the real hardware please? Rather than screen shots post new Menu->File->Save Local files.

I wonder what CPU-Z would report, I suspect much the same as SIV 5.69.

Here's the file from SIV tidy and a text file of what CPU-Z found. The Intel utility's driver was not loading. Unfortunately I can't find any of my other crimms at the moment. The Dell manual even stated if you only have two modules installed you put crimms in the first set of slots and the rimms themselves in the second set of slots and leave the 4-8 slots open. I was surprised but it obviously works since this thing posts no issues ever now. However, I'm putting this thing aside for the time being as I'm genuinely at a loss as to why it refuses to run at its full speed. I even tried making sure the intrusion switch (even though I have it off in the BIOS) was fully pressed down on startup and even tried a slower agp video card to make sure it wasn't some strange issue with agp 8x. The only thing I can think of is maybe its the latest bios causing it. I don't think it was an issue before I updated the bios but I can't remember. The power supply I completely recapped so I doubt its some sort of power issue and the motherboard mostly uses smd tantalums with the exception of two OS-CON caps which I highly doubt are the issue. Parts are non existent for this thing and it needs a new fan anyways since one of the JMC/Datech fans are rattly and I'd need to make an adapter to use a new fan with it. Just too much work for a system I'm unsure if I can get running 100% right ever again since the issue is still up in the air, I'll probably work on it again sometime in the future but for what I want here set up properly it's way too close (even beaten in some cases like 3DMark99) to my Katmai 500mhz. No doubt the 600mhz system I have waiting to be repaired would actually beat it heartily.

theamtrakvirus wrote on 2023-03-21, 02:38:

The Intel utility's driver was not loading.
The only thing I can think of is maybe its the latest bios causing it. I don't think it was an issue before I updated the bios but I can't remember.

Thank you for checking out 5.70 Tidy-00 which worked as I hoped it would. I plan to leave the garbage correction code there and release 5.70 Beta-00 later this week. As I expected CPUZ reported much the same as SIV 5.69.

OK, maybe it needs XP or later. I will check what happens on one of my Windows 2000 systems. I tend to use 2003 Enterprise Server on my systems, but I must have a P-3 with 2000.

It could easily be the BIOS messing up so feel it would be worth trying an older version. Maybe they added support for the 100 MHz FSB 1MB/2MB cache CPUs and messed up.

Yeah, Dell still holds a few different versions up on their site and I might try a couple this weekend to see. Hopefully it might help, though I find it odd that the last couple of revisions they did were so far apart. I think A11 was in 2001 and then A12 wasn't until like 2005 and the only documented change was something to do with changing the NIC hotkey. They did change the microcode of some cpus in A10 so I'll probably try A09 and see if that works or not. From the documented changes in 10 and 11 I don't think I'd be losing too much anyways.