Gigger Hertz

Pentium Pro-blems Redux

138 posts in this topic

I took some of the suggestions from my last thread to heart and modified my system's RAM, video card, and operating system. After a grand total of $700 and the last week of my time she's fighting a little harder than last round. The current specs are as follows:

  • 2x Pentium Pro 266 MHz/1 MB L2
  • 66 MHz bus
  • 1,024 MB 5v 60ns non-ECC EDO RAM
  • 100 GB 7200 RPM EIDE hard drive
  • ATi Radeon 9200 Pro PCI
  • FreeBSD 5.1

The FreeBSD 5.1 kernel was compiled with support for both of my CPUs and optimizations for the Pro specifically and its cache. I am running KDE 3.1 and can't complain about the speed at all — it's much greater than Windows XP ever was. I have been in touch with a FreeBSD hacker who ensured my decision to move to this operating system instead of Linux was the way to go. The SMP and virtual memory subsystems in the FreeBSD kernel are much more efficient than either Windows or Linux. FreeBSD cost me nothing to install and only took a little time to compile the proper kernel.

The four 256 MB chips cost $410 after tax and shipping/handling. Considering the RAM alone, performance is not what I would call a 2x increase over the 512 MB, but overall it did augment my system. The output below from a FreeBSD utility called top illustrates this. One thing I want to try is ECC RAM. It's 72-bit vs. non-ECC RAM which is 64-bit, and I think that switching my gigabyte of non-ECC RAM for ECC chips will give me more memory bandwidth and ergo the performance I was expecting in installing another 512 MB.

Processes: 48 total, 4 running, 44 sleeping... 126 threads
Load Avg: 2.26, 1.69, 1.30 CPU usage: 43.2% user, 56.8% sys, 0.0% idle
SharedLibs: num = 117, resident = 52.8M code, 2.45M data, 7.87M LinkEdit
MemRegions: num = 5359, resident = 109M + 9.91M private, 84.0M shared
PhysMem: 260M wired, 390M active, 287M inactive, 937M used, 87M free
VM: 2.38G + 82.5M 71607(0) pageins, 5190(0) pageouts

Redraws are faster and I now have a vast array of bit-depths and monitor resolutions with my new Radeon 9200. I also went out and bought a 21" CRT monitor so I could enjoy a larger desktop without going blind. Though the 9200 has done little to affect overall system performance, I appreciate what it does for my graphics. In the very least any games for FreeBSD will fly 4x faster than before: The 9200 has 128 MB RAM where my Rage only had 32 MB. I'm happy with this addition to my system especially since I got a $20 rebate off of the $200 I paid for it.

Aside from these tweaks my system remains the same. I can not boost the bus clock above 66 MHz. My 100 GB hard drive, due to BIOS limitations, is partitioned into several 16 GB slices. (I wonder if anyone knows of any BIOS hacks to overcome this limitation.) I believe I may finally be pushing the limits of my system's hardware and am looking at ways to optimize my operating system. Supposedly FreeBSD 5.2 is in beta now, and I've heard good things about something called QNX. Here's to any more suggestions to eek all of the power out of my dual Pentium Pro system.

Share this post


Link to post
Share on other sites

I will not comment on anything you have said except for the HDD. I do not believe there is any such BIOS limitation that requires you to partition your hard drive into 16GB chunks.

Linux/BSD does not use the BIOS to read the HDD and should work fine with the standard /boot, /, /home and swap partitions... more if you want em.

Share this post


Link to post
Share on other sites
The four 256 MB chips cost $410 after tax and shipping/handling.

One thing I want to try is ECC RAM. It's 72-bit vs. non-ECC RAM

In the very least any games for FreeBSD will fly 4x faster than before: The 9200 has 128 MB RAM where my Rage only had 32 MB. I'm happy with this addition to my system especially since I got a $20 rebate off of the $200 I paid for it.

And to think, for less money you could have picked up a P4 2.6C, a gig of ram, and a new motherboard, and case. Or an Athlon 2500+ or something for that matter, you'd even get a decent video card instead of that Radeon 9200.

You will not get extra bandwidth for using ECC ram.

The amount of ram on the video card will not make a difference the way you think it will. I pray that was Canadian dollars or something for that Radeon 9200, as I just bought a ASUS 9600XT/TVD for around 260 CDN(about 200 USD). Now if only it'd work with ATI's multimedia center.

Share this post


Link to post
Share on other sites
After a grand total of $700 and the last week of my time she's fighting a little harder than last round.

*tears welling up from laughter*

The SMP and virtual memory subsystems in the FreeBSD kernel are much more efficient than either Windows or Linux.

in smp scaling windows is better than linux, and linux is better than freebsd. for vms windows is better than freebsd, and freebsd is better than linux. if smp is what blows your hair back, you should be using linux.

I think that switching my gigabyte of non-ECC RAM for ECC chips will give me more memory bandwidth and ergo the performance I was expecting in installing another 512 MB

1) it won't give you more memory bandwidth

2) having more memory is only useful if you have things IN MEMORY

In the very least any games for FreeBSD will fly 4x faster than before: The 9200 has 128 MB RAM where my Rage only had 32 MB.

the memory isn't what's making it faster, so that's yet another misconception to which you've subscribed. see this review for some digs on the 9200se/9200.

My 100 GB hard drive, due to BIOS limitations, is partitioned into several 16 GB slices.

there is no such bios limitation.

Share this post


Link to post
Share on other sites
*tears welling up from laughter*

OK, come on. Is there really something so funny about a guy with a hobby that you have to laugh at me? You did this in my last thread too. Cut it out.

in smp scaling windows is better than linux, and linux is better than freebsd.  for vms windows is better than freebsd, and freebsd is better than linux.  if smp is what blows your hair back, you should be using linux.

The engineer I am in contact with agrees with your claims if comparing Linux to FreeBSD 4.x, but I am using FreeBSD 5.1 which has superior SMP and VM to Linux 2.4.

1) it won't give you more memory bandwidth

2) having more memory is only useful if you have things IN MEMORY

Why won't it give me more memory bandwidth? 72 > 64, so the datapath is wider, allowing more data throughput. That works the same way that a six-lane Interstate does when compared to a four-lane Interstate. What am I missing here? Is it possible that you're mistaken?

the memory isn't what's making it faster, so that's yet another misconception to which you've subscribed.  see this review for some digs on the 9200se/9200.

It can hold 4x the data, and assuming that the GPU in the 9200 is 4x faster than the GPU in my Rage, why wouldn't it be? The chip and data bus even went from 32-bit to 128-bit in my upgrade: Everything about the 9200 is 4x greater than the Rage. The only thing that hasn't changed here is the PCI bus which is far and away from being saturated by the data from any video card.

My 100 GB hard drive, due to BIOS limitations, is partitioned into several 16 GB slices.

there is no such bios limitation.

Well, it's a limitation of using the BIOS PNP, which Windows NT et al required. Now that I am using FreeBSD, which doesn't use BIOS to access hardware, I can be rid of my partitions. Let's see how much speed gain I get from having a proper /swap under BSD after I repartition everything.

Share this post


Link to post
Share on other sites
Why won't it give me more memory bandwidth? 72 > 64, so the datapath is wider, allowing more data throughput. That works the same way that a six-lane Interstate does when compared to a four-lane Interstate. What am I missing here? Is it possible that you're mistaken?

Not at all. ECC is error correction, it'll actually slow the system down, as you have to check whether the data is corrupt or not.

It can hold 4x the data, and assuming that the GPU in the 9200 is 4x faster than the GPU in my Rage, why wouldn't it be?

Well, considering how low end the 9200 is, I have my doubts whether the GPU is 4x faster. It's like the GeforceFX 5200, high GPU and memory clock speed, but still a useless card, beat handily in most tasks by even the old Geforce4 Ti cards.

Everything about the 9200 is 4x greater than the Rage.

You made the claim it was faster simply because it had 4x the memory.

The only thing that hasn't changed here is the PCI bus which is far and away from being saturated by the data from any video card.

In fact, the entire point of AGP is that PCI just isn't fast enough to move data for a video card. Even a low end video card can move gigabytes a second back and forth in its memory banks, PCI is only 133MB/s at best.

Share this post


Link to post
Share on other sites
The engineer I am in contact with agrees with your claims if comparing Linux to FreeBSD 4.x, but I am using FreeBSD 5.1 which has superior SMP and VM to Linux 2.4.

better vm, worse smp. linux is 2.6 now.

Well, it's a limitation of using the BIOS PNP, which Windows NT et al required.

wrong, and wrong. there is no bios limitation for 16gb anywhere. futhermore, 'pnp os' is not recommended for any windows operating systems. see here for proof.

Why won't it give me more memory bandwidth? 72 > 64, so the datapath is wider, allowing more data throughput. That works the same way that a six-lane Interstate does when compared to a four-lane Interstate. What am I missing here? Is it possible that you're mistaken?

it uses it for, duh, the ecc. see here for proof.

It can hold 4x the data, and assuming that the GPU in the 9200 is 4x faster than the GPU in my Rage, why wouldn't it be?

you seem to think that by adding more memory, everything will magically go into it. i don't doubt that the 9200 is MORE than 4x faster, but it's not because of the AMOUNT of memory it has. it has mostly to do with the cpu and memory SPEED.

go look at reviews contrasting 64mb vs 128mb cards of the exact same class. in most games there is little difference. why? because people use texture compression and don't load everything into vram because most of their market isn't going to have enough to handle it.

i'm glad you wasted your money. you were provided with all the information you needed to make the right decision by endless sources in your original thread.

Share this post


Link to post
Share on other sites
in smp scaling windows is better than linux, and linux is better than freebsd.  for [virtual memory] windows is better than freebsd, and freebsd is better than linux.  if smp is what blows your hair back, you should be using linux.

OMFG LOLF! Please tell me you're kidding, that's the most ridiculous assertion I've heard all week.

Share this post


Link to post
Share on other sites

no, i'm not kidding. only with 2.6 has linux gotten their heads above water in the 4+ cpu segment, and that was literally released like a week ago. if you want to look it up, just search for ibm's statements on it. they waited for 2.6 with bated breath so that they could 'get real' with 16-way systems.

if you want a testament to smp scale or robust vms, look to the hundreds of thousands of n-way citrix servers in use right now.

Share this post


Link to post
Share on other sites

What a monumental waste of seven hundred dollars.

Oh well. Hopefully it was at least an entertaining waste. For seven hundred georgies, I would have gone with an athlon 2500+ system, but....whatever.

Share this post


Link to post
Share on other sites
no, i'm not kidding.  only with 2.6 has linux gotten their heads above water in the 4+ cpu segment, and that was literally released like a week ago.  if you want to look it up, just search for ibm's statements on it.  they waited for 2.6 with bated breath so that they could 'get real' with 16-way systems.

if you want a testament to smp scale or robust vms, look to the hundreds of thousands of n-way citrix servers in use right now.

That's cute, but SGI's Altix has been using Linux on systems with 64 processors and up for more than six months now. Incidentally, SGI would never be dumb enough to put Windows on such high performance systems.

Share this post


Link to post
Share on other sites

ECC:

Each byte of data contains 9 bits, not 8, ergo 72bit memory.

The bandwidth to your CPU is the same...while technically the bandwidth to the memory controller is ~12% greater. Again, your CPU doesn't know you're using ECC. Bandwidth doesn't increase, but your reliability will increase.

I can vouch for IBM's findings regarding Windows Server versus Linux. Improving scalability (i.e. efficiency) is arguably the biggest feature of the 2.6 kernel.

I didn't think it was a secret that the enterprise market was demanding better performance from Linux companies.

And just because some company is running Linux on 64 CPUs doesn't make it the most efficient. I would think cost has something to do with their decision. ;)

I will be very interested to see what results from these megacluster super-computers once they run the 2.6 kernel. B)

DogEared

8^)

Share this post


Link to post
Share on other sites
That's cute, but SGI's Altix has been using Linux on systems with 64 processors and up for more than six months now. Incidentally, SGI would never be dumb enough to put Windows on such high performance systems.

here is a pdf written by sgi less than 6 months ago - the topic is the altix and scalability. a quote: The CPU scheduler in the 2.4 (and earlier) kernels is simple and efficient for uniprocessor and small MP platforms, but it is inefficient for large CPU counts and large thread counts.

now, will you please shut up?

Share this post


Link to post
Share on other sites

I really need to follow this thread. Seems another funny one.

Share this post


Link to post
Share on other sites
What a monumental waste of seven hundred dollars.

Oh well.  Hopefully it was at least an entertaining waste.  For seven hundred georgies, I would have gone with an athlon 2500+ system, but....whatever.

What you think is a "monumental waste" of $700 makes a lot of sense to some people, buddy. Maybe some of us would like to push the limits of efficiency rather than getting speed simply by throwing money at some new system with questionable technology. This is a pet project, perhaps you should go and look up the definition.

Share this post


Link to post
Share on other sites
no, i'm not kidding.  only with 2.6 has linux gotten their heads above water in the 4+ cpu segment, and that was literally released like a week ago.  if you want to look it up, just search for ibm's statements on it.  they waited for 2.6 with bated breath so that they could 'get real' with 16-way systems.

if you want a testament to smp scale or robust vms, look to the hundreds of thousands of n-way citrix servers in use right now.

Thanks, but I think I'll wait around to see how Linux 2.6 pans out. It's too early in its lifecycle to trust running my system to. FreeBSD 5.1 has been around for a little while now and is rock solid. Linux, I would imagine, still has a couple serice packs and hot fixes before it's ready for primetime.

Share this post


Link to post
Share on other sites

ok, you have got to be a troll.

freebsd 5.1 is not considered to be stable. linux 2.6 is, and has been in testing for a long time (by large contributing companies such as the already-mentioned ibm and sgi). furthermore, the idea of calling updates 'service packs' or 'hot fixes' for linux is ridiculous.

the 5.x freebsd tree is not expected to be declared stable until 5.2-RELEASE at a MINIMUM.

Share this post


Link to post
Share on other sites

Does your pet have a name? ^_^

You gotta love the "If it was my money..." posts. It shows people care.

Here's mine:

"If it was my money, I would buy another Dell 2001FP monitor." :lol:

They probably aren't that cheap anymore.

If you know anyone looking to get rid of Sun equipment in Northern Illinois...let me know. :)

DogEared

8^D

Share this post


Link to post
Share on other sites
ok, you have got to be a troll.

This kind of thing got said in my last thread. I appears to me that this board has a problem with someone with a hobby that flies in the face of modern-day computing trends. If that makes me a troll, then so be it. I happen to think that some of us have a problem with groupthink and consumerism. Knock off the ad hominem attacks or leave or quit posting in this thread.

freebsd 5.1 is not considered to be stable.  linux 2.6 is, and has been in testing for a long time (by large contributing companies such as the already-mentioned ibm and sgi).  furthermore, the idea of calling updates 'service packs' or 'hot fixes' for linux is ridiculous.

What's the difference in what updates to an operating system are called? If you want to be all hoity-toity be my guest, but it's kind of offtopic whether I am familar with Winows Update and use the terminology for another product. Everyone, including yourself, knew what I was talking about.

the 5.x freebsd tree is not expected to be declared stable until 5.2-RELEASE at a MINIMUM.

5.1 released June 2003, Linux 2.6 released December 2003.

Share this post


Link to post
Share on other sites

5.1 is NOT considered a stable release. i provided urls which plainly state that fact. linux 2.6 is, and has been heavily tested.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now