Unsolvable performance problem...

[Ichinisan]

My friend is moving out of state soon, and he left his old computer with me for the last week so I could give it one final tune-up. This is an opportunity to give him some surprise upgrades as a going-away present...but I'm having a problem. Half-Life 2 is choppy. With all settings turned off / low, it's just as choppy at 640x480 resolution as it is when running 1920x1200! HL2 is an older game, and should run like a dream on this 7800GS graphics card. I doubt it's limited by the Celeron CPU. Why would HL2 be CPU-limited on a 2GHz 400FSB system? I just can't figure this out! I've tried everything short of reinstalling Windows...which I *really* don't want to have to do!

I really want my friend to be blown away by this upgrade. Do you guys have any
suggestions?

Intel 2.0Ghz Pentium 4 Celeron Northwood socket 478
ECS motherboard
-supports both PC100 SDRAM and PC3200 DDR400 SDRAM
-AGP 4x version 3.0
-Chipset "VIA P4X266 (VT8753)" according to CPU-Z
128MB PC100 SDRAM -> 2GB PC3200 DDR400 SDRAM (OCZ)
GeForce3 Ti500 -> GeForce 7800GS 256MB GDDR3
Maxtor 500GB IDE hard drive (WinXP Pro installed)
Antec 350 Watt power supply

Update: Here is a CPU-Z report while only one module was installed:
http://webpages.charter.net/ichinisan/cpuz.htm

The other hardware is irrelevant, because I have tried removing *all* non-essential hardware from the system and disabling all integrated devices. It makes no difference in HL2 (still the only game I have tried so far).

Since I got the system from him, I've switched it from 256MB of PC100 SDRAM to 2GB PC3200/DDR400, upgraded the video card from a Geforce3 Ti500 to a Geforce 7800GS, and installed some games to take advantage of it (from Steam gift tokens).

Upgrading the 256MB (2x128MB) of PC100 SDRAM to 2GB (2x1GB) PC3200 DDR SDRAM was more trouble than I expected. It required me to switch a HUGE bank of jumpers to enable the DDR slots and another jumper for changing the voltage for the memory type. I had already tried the DDR memory in the system before realizing that any jumpers had to be changed, and the system didn't boot. At first, I thought I might have fried the DDR memory with too much voltage, but everything seems fine. The huge bank of jumpers was probably keeping most of the traces to the DDR slots disconnected. There have been no stability issues at all after successfully upgrading the memory.

I still occasionally get a beep from the system when I'm using the keyboard. For instance, when I'm holding an arrow key to move through some text, it will occasionally beep from the system. It seems that it's triggered by holding down a key until it repeats for several seconds. This shouldn't happen unless there is a stack overflow, and shouldn't happen in Windows with working sound drivers installed.

I went as far as I could to free up resources: 1) Physically disconnected all peripherals and drives except the primary hard drive, keyboard, and USB mouse, 2) disabled all non-essential integrated devices (parallel port, secondary IDE, Game/MIDI device, modem device, floppy controller, etc) leaving only the USB controller enabled for the mouse.

I have made sure that it's running the latest BIOS from the board manufacturer (ECS). I've tried loading "optimal" defaults in the BIOS setup, and "performance" defaults. I've used Google to examine every BIOS option for relevance and CPU-Z to make sure that the memory and FSB running at the correct speed. I used ThrottleWatch to see if the Intel CPU might be throttling itself, and it shows a constant 2,000Mhz and a perfectly flat line on the frequency graph, so the CPU cooling is working as it should.

I've installed the latest VIA drivers from the board manufacturer, but I don't know if they are the latest from VIA for the chipset.

Windows XP is completely up-to-date, but I haven't installed the service pack 3 release candidate. There is no bloatware and I've made sure to shut-down all apps that load on startup. I've even stopped all unrelated background services and couldn't make a difference.

If I can't get HL2 to run correctly on this rig, there is an underlying problem that is going to affect other things too. It would be a waste of good memory and a good video card...so I'll need to figure this out QUICKLY before his flight leaves later today!

HAAALP! ;P

Any help is appreciated!

 

[Roguestar]

Well you see the problem is that the Celeron is the white trash chip.

You could buy a 3GHz P4 for a few dollars off ebay. http://cgi.ebay.com/Pentium-4-...VWQQrdZ1QQcmdZViewItem


Oh, you need it fixed today.

Well, you could buy it off ebay and mail it to him with instructions on how to change the CPU?

 

[Ichinisan]

Could it really be that? I wouldn't expect an older game like HL2 to be CPU-limited at all. And certainly not with all of the settings turned down.

 

[Ichinisan]

It's going to be a pain, but I do have a 3.0Ghz Northwood proc to try. It just seems like such a mis-match to put an FSB800 CPU in this board!

I really don't have enough time to do this, but I'm going to try anyway...

 

[Roguestar]

That Celeron isn't going to even be as good as a 2GHz P4, you see. HL2 shouldn't be nearly limited by the graphics card, the 7800GS would easily do but less than a 2GHz P4 would make me think that it's bottlenecked there. Try some other games and see how they do?

 

[Ichinisan]

I'll try some other games too. I believe I have run HL2 well on lesser systems before (AMD 1800 systems, etc), but I'm not sure. I know I've run it on laptops with slower clock speeds.

I still think it's strange that the CPU would be a limitation in some parts. For instance: When the game starts up, before a 3D scene is loaded, it's even slow to redraw the Valve image that is supposed to jump / flicker a little bit. Even more strange is that the opening scene of the game (where the G-Man is talking) seems very smooth, until the very point when the game opens up to player input. Strange.

 

[CZroe]

Might be time to offer a complete board upgrade. A 2GHz Celey shouldn't be a bottleneck for HL2 and, even if it is, you'll need it for the 800FSB CPU.

 

[Roguestar]

Originally posted by: Ichinisan
I'll try some other games too. I believe I have run HL2 well on lesser systems before (AMD 1800 systems, etc), but I'm not sure. I know I've run it on laptops with slower clock speeds.

I still think it's strange that the CPU would be a limitation in some parts. For instance: When the game starts up, before a 3D scene is loaded, it's even slow to redraw the Valve image that is supposed to jump / flicker a little bit. Even more strange is that the opening scene of the game (where the G-Man is talking) seems very smooth, until the very point when the game opens up to player input. Strange.

The part you're talking about it all pre-rendered movie, though, IIRC. Once you can start running about the train carriage it's up to the CPU to start crunching numbers, which is where it'll start getting bogged down.

 

[CZroe]

No, Valve takes pride in using NO pre-rendered movies (even their FMV teaser trailers are made with in-game engine footage). That part is most certainly real time, shifting "negative" black & white effects and all. The slow 2D intro splash screens, if FMV, only indicate that 2D FMV is slow while the real-time 3D intro is not, but real-time 3D gameplay is slow again. Unlike my Core Duo laptop that choked on the Battlefield 2 intro movie but sang with in-game performance (X1400), his inexplicable performance is all over the map.

FMV=Full Motion Video ("pre-rendered movie")

 

[Ichinisan]

Originally posted by: Roguestar
Well you see the problem is that the Celeron is the white trash chip.

You could buy a 3GHz P4 for a few dollars off ebay. http://cgi.ebay.com/Pentium-4-...VWQQrdZ1QQcmdZViewItem


Oh, you need it fixed today.

Well, you could buy it off ebay and mail it to him with instructions on how to change the CPU?

I have that exact CPU and I tried it. The board doesn't recognize the FSB pins specified by the FSB800 CPU, so it defaults to 100MHz (100MHz "quad-pumped" = FSB400) and won't let me modify it to the highest supported by the board. This means that my 3.0GHz CPU runs half-speed @ 1,500MHz; which is SLOWER than his original 2.0GHz CPU....*BUT* ...the in-game performace with HL2 still improved dramatically! Again: Performance in HL2 is *MUCH* better with the P4 @ 1,500MHz than with the P4 Celeron @ 2,000MHz. I would have thought that *any* 2.0GHz CPU, Celeron or not, should have been able to handle a game like HL2 just fine. I was wrong. It's frustrating that I can't override the FSB to the board's maximum (133MHz, "FSB533"). It's an ECS-brand board with no FSB adjustment or FSB800 support, so it locks at 1,500MHz instead of 3,000Mhz. If I could at least over-ride to 133MHz/FSB533, I could get 1,995MHz.

The CPU is a mis-match. I'm going to need to get a 533FSB P4 Northwood that is not considered a "Celeron".

 

[Ichinisan]

Annoying story:

I tried *everything* else that was possible before changing the CPU for a very good reason: This system EATS THROUGH CPU MOUNTS LIKE NOTHING I HAVE EVER SEEN.

I met my friend when I was installing his Internet service 2-3 years ago. His computer was extremely slow and I was able to diagnose the problem while I was installing the Internet service. One corner of the HSF mount was broken (the bracket that stays on the board around the CPU socket), and the heatsink was not being pulled flat against the CPU. The CPU was getting hot and throttling its speed to compensate. A recent TomsHardware demonstration video was still fresh in my mind showing what happens when the HSF is lifted off of an Intel CPU versus an AMD CPU. The AMD CPU burned up immediately with smoke and fire, and the Intel CPU throttled itself internally and the game (Quake 3) froze. When the HSF was set back down on the Intel CPU, the game resumed and the system never even crashed.

For my friend, I was able to scavenge a mounting bracket from a board at work (very lucky).

The day before I started this thread, I carefully removed the fan and plastic clamp from the bracket, remembering what had happened years before. I made sure that there was absolutely no stress on the board's bracket while removing the clamp and fan. I removed a carpet of collected dust that was blocking pretty-much all airflow through the CPU heatsink.

The computer's chassis is designed with an air duct, which ensures that the air is not coming from anywhere near the floor/carpet. Still, the HSF managed to get completely choked up with accumulated dust / lint. So, it seems that any similar system would eventually *need* to have the clamp removed and the HSF cleaned.

I tried to remove the fan from the heatsink, but it wasn't coming. After fumbling with it, I realized that there was no longer a seal between the HSF and the CPU, so I would have to re-apply the Arctic Silver 5 thermal compound. I took it off cleaned everything, and re-applied AS5. I carefully placed the clamping cage assembly over the HSF and locked one side at a time...and heard a *SNAP*.

"$#@%!" A corner broke off from the bracket.

I was as careful as possible. There was nothing I could have done to prevent it. The HSF *needed* cleaning. The only stress on the assembly was the closing of the clamping levers. I had to scramble to find a computer shop that could scavenge one from another board, and time was running out to finish this system upgrade.

Replacing the bracket was a huge hassle because I had to remove the motherboard to take the old one off. Still, it was done. I upgraded the memory and graphics card and did EVERYTHING I COULD to troubleshoot the performance issue without changing CPUs (and dealing with the bracket-eating HSF from hell). I stayed up through the night, trying to get it resolved before my friend moved out of state. The next day, I started this thread. There was apparently nothing left to do in troubleshooting except change the CPU...so I did.

Well...changing the CPU fixed the problem, but it broke *ANOTHER* mounting bracket! That's the THIRD TIME! It always happens when closing the clamping levers, where I have *NO* control over how tightly it clamps. It's frustrating that I have no idea why each bracket didn't break when first installed. It only breaks when the clamp is carefully removed, put back on, and closed.

When I find a 533 FSB CPU for the system, I'm going to change out the entire clamping mechanism.

Now he's gone and I'm going to have to find a way to ship this HEAVY system to him. It's one of the heaviest systems I've seen (for that size). It's a full tower packed with 3 hard drives, 2 optical drives, a SATA card, USB2.0/1394 combo controller card, large-size graphics card, several fans, and a weighty Antec PSU... Shipping is going to cost a lot.