Jump to content

sluggo

Members
  • Content Count

    80
  • Joined

  • Last visited

About sluggo

  • Rank
    Member
  1. I've spoken to a couple of other folks on this since I first posted. One claims that he's tested supplies that do have independent current limiting for each rail but could not recall the models. In any case, what I should have perhaps said was "all the "multiple-rail" designs that I've seen are just single rail supplies". Anything is possible, but independent regulation would raise the cost of the supply significantly, and as far as I know only the Enermax Galaxy line ($300+) offers independent regulation on even two of it's outputs, much less the three or four "rails" shown on some supplies. I'll continue to test new supplies as I get them and will try to remember to post here when I see something new. Cheers.
  2. For consumer ATX supplies, the "4x +12 rails" simply reflects a marketing trend that's caught on. Your supply's +12 output is 48A with at least 20A available from individual cable bundles (and probably more), but in no case can you draw more than 48A from the supply on +12, so 80 Amps (as technodanvan said) is simply not going to happen. The Mushkin supply is still good for 576 Watts on +12, which is a lot. Personally, though, if I was spending that sort of money I'd at least expect it to have active power factor correction, which it does not. The Corsair 620HX that I use has 600W capability on +12 (listed as "3x18"), has active PFC, is rated for full output up to 50 degrees C, is dead quiet and comes with a 5 year warranty. It's the same price as the Mushkin. PCP&C makes a quality product, no question. A 1kW supply would be way more than I need, and I suspect is more than most non-quad, non-SLI, non-RAID users would need. How much you need is up to you, but I'd look closely at your power budget before pulling the trigger on that purchase. You can choose from several very, very good supplies and spend no more than $150. I used to design mainframe power systems for one of the large computer companies. Power is still a good field, but the vast bulk of the basic consumer electronics product design work is now done in Taiwan and China with controllers made by companies like Maxim, TI, and PI. Right now I'd be looking more closely at automotive and solar applications. Those look like they'll be fun for the next 15 years or so.
  3. Good question. The short answer is that the supply is capable of sourcing 48 Amps of +12, and the "4x20" specification is simply marketing fluff. Much as it pains me (as a former power supply designer) to see "marketing" applied to something so basic to the operation of a PC, I suppose it was inevitable. The long answer is that the "split-rail" language that you see being used had it's birth in a well-intentioned but widely misinterpreted section of the ATX 2.2 spec. In that spec, a recommended guideline was proposed that would limit the energization (total energy available) of any particular line to 240VA in order to comply with a long-standing UL spec for user-accessible areas of listed equipment. The 2.2 spec also recommended, but did not require, an independently regulated +12 supply for CPU power. To this date, no PC-grade ATX supply has fully implemented either of these recommended specs. What they have done is simply to bundle cables in such a way as to create the "plumbing" for independent 12V rails, but the electronics are unchanged - it's simply a single 12V supply capable of prodigious output. The latest word from Intel is that in the next released version of the ATX spec (we're up to 2.9 on the preliminaries already), the recommendation for 240VA limiting is gone altogether. There will not even be the appearance of a requirement for multiple independent rails. The thinking behind the 240VA limit from UL is a little fuzzy. It's a spec that's been in place for a long, long time, and whether it's to reduce the likelihood of a self-sustaining fire or to protect the user from electrocution has been a matter of much discussion and one for which UL seems to offer strangely little guidance. My interpretation had always been that it was there to protect users of 240V AC lines from getting blasted into next Tuesday if they did something stupid. In no way does a 12V line offer any threat of electrocution, so applying a 240VA limit to user-accessible areas of a PC makes not a lot of sense to me. I think Intel and the ATX spec-builders may have finally gotten UL to come around to this way of thinking. We'll see.
  4. Many folks wonder about "single-rail", "dual-rail", "triple-rail" supplies and which is better or more appropriate for what they plan to do with their machine. I just wanted to clear up what seems to be a common misconception, which is that these rails are independent of one another. In fact, with the exception of one family of supplies (that I know of), there is only one +12 rail in every ATX supply on the market. All the "rails" labeled on the power supply case are actually the same current source and labeled as separate rails only to comply with a UL requirement for consumer-class supply cabling. Every ATX supply (with the one exception) has a single winding on the secondary of the main transformer that generates the +12 voltage. The output of this transformer is filtered and fed to all the +12 circuits in the PC. There are no other isolation or protection devices that would constitute a separate "rail", and all of the +12 lines in the PC go back to a single point in the PSU. The one exception that I know of to this is the new Enermax Galaxy line, which starts at 850W for around $300. It has two separate secondary transformers for +12 and has separate output specs for each. To sum up, the only number that really counts w/r/t +12V is the total +12 current available from the supply. All the "multiple-rail" designs you see are just a single +12 supply with one point of regulation and protection. I hate to see people get all wrapped around the axle on this one issue, which really shouldn't be an issue at all.
  5. I won't comment on the memory speed - that's probably been covered. My 165 is running at 2.81GHz on 1.344V. Bus is at 312, memory at 255 (5/6). BIOS is 04/06. Upgrading from a single core (Venice) to the Opty I hit a snag. Instead of the Uniprocessor HAL (which, contrary to it's name, supports both single and dual cores), I had the single processor HAL. When I dropped in the Opty, the BIOS recognized it, the Windows Device Manager recognized it, but I was only running on one core. I had to delete the hal.dll and manually replace it with halaacpi.dll before both cores would run. I'm told that when you first install XP, the Uniprocessor HAL is loaded by default. In my case it was not. Look at /system32/hal.dll - right click it and see what it's internal name is. Check it against the names at MS support http://support.microsoft.com/kb/309283 If you're using halaacpi.dll, you should be good to go. If you're using halacpi.dll, you can delete it, copy halaacpi into /system32, and rename it hal.dll (this is what I had to do). If you don't feel like doing this then you're probably looking at reloading XP, if the guide above doesn't have any better tips.
  6. Got it solved. Whatever hal you load is copied into /system32 and is renamed hal.dll, which in my case was the single-processor dll. I deleted hal.dll, copied halaacpi.dll into /system32 and renamed it hal.dll. Halaacpi is called the "Uniprocessor HAL" and supports both single and multi-processor machines. One reboot later and all is well with my Opteron and the benchies went up by close to 2x. As far as I can tell, brute-forcing it like this is the only way to change hal's on the fly.
  7. No, I didn't. I think Windows correctly identified the new processor and loaded drivers. Device Manager, under Processors, shows two entries for Dual Core AMD Opteron Processor 165, driver version 5.1.2600.0, with one having a Device_Instance_ID of 0 while the other has a 1. Would I still need to re-install? EDIT: Boogers. I took a look in the Device Manager/Computer and I'm still using a standard HAL. ACPI single processor version. MS says I have to reload to change HAL's - is there no way around this?
  8. I'm using an NF4 Ultra-D with an Opteron 165. The problem is on all the benchmarks I've run, even those that support multiple cores, it responds like a single-core FX-57. Sandra, for instance, doesn't recognize it as a multi-core processor. I'm using the 0406 BIOS. Is there something I'm missing?
  9. Slow-closing applications sounds like page file problems. Check your page file (virtual memory) settings and make sure everything is as it should be. Maybe delete and recreate your page file if it's fragmented.
  10. It took a while (a looooong while), but I eventually found what looked like a solution at Adaptec's site. Although the conditions were different and it was for a different product, the suggestions worked. Rather than try to paraphrase and get it wrong, I'll just quote the section that got me fixed. "This problem can be caused when the motherboard BIOS does not correctly assign the proper resources for the 2200s RAID adapter. The motherboard BIOS is responsible assigning resources (such as IRQ, BIOS memory address, etc.) for 2200s RAID adapter. By adding 2 GB of RAM to the motherboard, the system would require the change in the BIOS load reserved memory area that normally would take place the next time the system was booted. This does not seem to have take place correctly and must be manually set to force the motherboard to re-allocate the resource for the 2200s RAID adapter and the process for this is as follows. 1. Turn power off on the system and remove the 2200s RAID adapter from the system. 2. Simply insert a bootable floppy disk into the floppy drive and power on the system to boot to the floppy disk multiple times (between 2 and 5 times). 3. Power off the system and put the 2200s RAID adapter into the system and do not connect any cables/drives to the 2200s RAID adapter. 4. Boot to the floppy disk again. This will check the BIOS of the 2200s RAID adapter to make sure it will now show the BIOS banner and search for attached devices and successfully boot to the floppy disk. When the 2200s BIOS is identified on the screen, the motherboard has successfully assigned the resources needed for the adapter to function properly. 5. With the Adaptec BIOS now showing up, power off the system and attach the cable(s) to the 2200s with drives and boot normally as this should now boot to the attached array drives that contains the operating system. The 2200s SCSI RAID adapter should now function normally as before and no harm to the array drives or data should have take place, but it is a good idea to re-boot the system a few times to test and make sure. Checking the operating system and any data files randomly to make sure the 2200s SCSI RAID adapter is working. " Several DOS boots off a floppy was all it took for the system to finally hook into the card's BIOS. I doubt I'd have ever thought of this. Hope this is useful info for someone sometime.
  11. I've tried the following: - added an HDD delay in the system BIOS - moved "Add-in Bootable card" to the top of the boot priority - reset ESCD - changed to a 3rd SCSI card - confirmed that the cards are being initialized at boot-up and have been assigned an IRQ - mounted a bootable HDD as the only device on the SCSI bus - changed PCI slots - loaded DFI BIOS 0406 After all this, the system is still not recognizing/acknowledging the SCSI cards' BIOS at boot time and so I cannot do low-level configuration of the cards. I'm really stumped. Has anyone seen this sort of thing before?
  12. I'm having a problem in my NF4 Ultra-D motherboard. I tried two different Adaptec SCSI host bus adapters - a 2940UW and a 2920C. The DFI board will not play nice with either board, but I can put either of them in an ancient ECS K7S5A (Barton 2500+) and they both work perfectly. The Adaptec cards allow you to modify the low-level operations of the board by pressing CTRL-A at some point during the boot. When loaded in my DFI, the 2940 displays the CTRL-A message, but when trying to load, the utility comes back with "There is no SCSI host bus adapter". With the 2920C installed the CTRL-A message never appears and the board will not respond to CTRL-A. Windows XP loads the correct driver and all appears normal in Device Manager, but trying to poll the SCSI bus on either card causes the DFI machine to immediately reboot. I'd be interested in knowing if others have working setups with these two older HBAs. EDIT: I recently installed a Buslogic BT-958, and it has the same problems - cannot access the card's BIOS configuration utility.
  13. The Adaptec cards might not be his problem, but it's interesting that I'm having the same problem in my NF4 Ultra-D motherboard. I tried two different Adaptec cards - a 2940UW and a 2920C. The DFI board will not play nice with either board, but I can put either of them in an ancient ECS K7S5A (Barton 2500+) and they both work perfectly. The Adaptec cards allow you to modify the low-level operations of the board by pressing CTRL-A at some point during the boot. When loaded in my DFI, the 2940 displays the CTRL-A message, but when trying to load, the utility comes back with "There is no SCSI host bus adapter". With the 2920C installed the CTRL-A message never appears and the board will not respond to CTRL-A. With either card in the DFI board, trying to poll the SCSI bus causes the machine to reboot. I'd be interested in knowing if others have working setups with these two older HBAs.
  14. The PWMIC is indeed one package. PWM in this case stands for Pulse Width Modulator, and pulse width modulation is what that part does. It sends control signals to the mosfets, turning them on and off in three phases. The mosfets and the PWMIC both get hot, but the mosfets are designed to be abused. The PWMIC needs the most cooling.
  15. Not to confuse the issue further, but you might want to take both a DC and an AC voltage measurement. All of the rails are primarily a DC level with a smaller AC component riding on top of it. Ripple, it's called, and ripple can often make two voltmeters report different values, even if they measure the same valus for the DC component. Different meters use different DC sampling and approximation methods, and ripple affects them differently. If your DMM and the chip on the board read the same value for a rail (+/- .01V) over a .25V range I'd be amazed.
×
×
  • Create New...