Home » overclocking » Page 3

overclocking

01/26/2001

Hey hey! It works! The server was down all day yesterday, which was a shame. I wanted to try a new experiment. So I’ll try it today.

I saw criticism over at Storage Review on Wednesday morning for their critiques of other hardware sites’ reviews. I disagree with this criticism; many of the reviews out there are atrocities, with poor methodology, hearsay, reviewer ignorance, and other shortcomings. Sometimes these reviews are more misleading than the information in the products’ advertising or packaging! I believe Storage Review is well within professional bounds to point out these shortcomings when they find them.

The mainstream media does this all the time. Columnists and editors will criticize the reporting done in other publications. Most newspapers also employ one person, known as the ombudsman, whose job it is to criticize and/or defend, as appropriate, the publication’s own work.

Seeing as the hardware sites out there often do very sloppy work, even compared to the mainstream media, some policing of it is a very good thing.

Then, over lunch, the idea hit me. Why not do some critiquing myself? I’m trained in editorial writing and editing. I have some experience as a reviewer. And I’ve published a fair bit of my own work in the arena of technology journalism–newspaper columns, a book, individual magazine articles, a series… So I’m qualified to do it, even though I’m not the biggest name out there. And that kind of content is certainly more useful than the “this is how my day went” stuff I’ve been posting way too often.

I’m not so arrogant as to assume that the webmasters of these large sites are in my readership and would take my advice. I don’t expect to change them directly. What I do expect to do is to raise people’s expectations a little. By pointing out what’s good and what’s not so good, hopefully I can raise the public consciousness a little, and indirectly influence some of these sites. If not, then at least my readers are better informed than they otherwise would be, and that’s definitely a good thing.

KT-133A roundup (Tom’s Hardware Guide)

This is a roundup of six VIA KT133a boards. Good review overall. It doesn’t get bogged down in three pages of history that tend to look like a cut-and-paste job from the last similar review, unlike some sites. But it does give just enough history to give proper perspective, though it would have been nice to have mentioned it took EDO and SDRAM some time to show their advantages as well–DDR is no more a failure than the technologies that came before. Unusual for Tom’s, this review isn’t obsessed with overclocking either. Lots of useful information, such as the memory modules tested successfully with each board. Inclusion of the DFI AK74-AC, which will never be released, is questionable. I can see including a reference design, but a cancelled commercial board doesn’t seem to make much sense. You can get an idea from its scores why it got the axe; it was consistently one of the bottom two boards in the roundup.

Emphasis was on performance, not stability, but Pabst and Schmid noted they had no compatibility or stability problems with these boards. Stability in benchmarks doesn’t guarantee stability in the real world, but it’s usually a good indication. As tight as the race is between these boards, stability is more important than speed anyway, and since the majority of people don’t overclock, the attempt to at at least mention compatibility and stability is refreshing.

Socket 7 Upgrade Advice (AnandTech)

This is a collection of upgrade advice for Socket 7 owners. This review, too, doesn’t get too bogged down in history, but the mention of fake cache is noteworthy. This was a PC Chips dirty trick, dating back to 1995 or so, before the K6 series. It wasn’t a very common practice and didn’t last very long–certainly not as long as the article suggests.

Lots of good upgrade advice, including a short compatibility list and pitfalls you can expect. Also included are some benchmarks, but it would have been nice if they’d included more vintage chips. The oldest chip included was the K6-2/450, and AMD sold plenty of slower chips. You can’t extrapolate the performance of a K6-2/300 under the same conditions based on the 450’s score.

Also, the rest of the hardware used is hardly vintage–you’re not likely to find an IBM 75GXP drive and a GeForce 2 video card in an old Socket 7 system. Using vintage hardware would have given more useful results, plus it would have given the opportunity to show what difference upgrading the video card and/or CPU makes, which no doubt some Socket 7 owners are wondering about. Testing these chips with a GeForce does demonstrate that a more modern architecture will give better peformance–it exposes the weaknesses of the CPU–but indication of how much a new CPU would improve a three-year-old PC would be more useful to most people. Few people have the delusion that a K6-3+ is going to challenge an Athlon or P3. They just want to know the best way to spend their money.

No deceiving graphics or lack of knowledge here; what’s in this article is good stuff and well written. It’s just too bad the testing didn’t more closely resemble the real world, which would have made it infinitely more useful.

Memory Tweaking Guide (Sharky Extreme)

This is a nice introduction to the art of memory tweaking, and it explains all those weird acronyms we hear about all the time but rarely see explained. Good advice on how to tweak, and good advice on how to spend your memory money wisely. They disclosed their testbed and included the disclaimer that your results will vary from theirs–their benchmarks are for examples only. The only real gripe I have is that the benchmark graphs, like all too many on the Web, don’t start at zero. From looking at the graph, it would seem that Quake 3 runs six times as fast at 640x480x16 than at 1600x1200x16, when in reality it runs about twice as fast. Graphing this way, as any statistics professor will tell you, is a no-no because it exaggerates the differences way too much.

Asus CUSL2C Review (Trainwrecker)

This is a review of the Asus CUSL2C, an i815-based board intended for the average user. This review has lots of good sources for further information, but unfortunately it also has a little too much hearsay and speculation. Some examples:

“Of course, Asus won’t support this [cable] mod and we’re pretty sure that doing it will void your warranty.” Of course modifying the cable on an Asus product, or any other manufacturer’s product, will void your warranty. So will overclocking, which they didn’t mention. Overclockers are either unaware or apathetic of this. In matters like this, assertiveness is your friend–it gives a review credibility. One who is assertive and wrong than is more believable than one who is wishy-washy and right.

“Arguably, Asus provides the best BIOS support in the business. We believe Asus develops their BIOS’s at their facility in Germany.” Indeed, Asus claims to have re-written over half the code in their BIOSes, which is one reason why Asus boards perform well historically. Most motherboard manufacturers make at least minor modifications to the Award, AMI, or Phoenix BIOS when they license it, but Asus generally makes more changes than most. This claim is fairly well known.

I was also disappointed to see a section heading labeled “Windows 2000,” which simply consisted of a statement that they didn’t have time to test under Windows 2000, followed by lots of hearsay, but at least they included workarounds for the alleged problems. Including hearsay is fine, and some would say even beneficial, as long as you test the claims yourself. This review would have been much more useful if they had delayed the review another day and tested some of the claims they’ve heard.

There’s some good information here, particularly the links to additional resources for this board, but this review is definitely not up to par with the typical reviews on the better-known sites.

DDR Analysis (RealWorldTech)

Good perspective here, in that DDR is an incremental upgrade, just like PC133, PC100, PC66 SDRAM, and EDO DRAM were before it. But I don’t like the assertion that faster clock speeds would make DDR stand out. Why not actually test it with higher-speed processors to show how each of the technologies scale? Testing each chipset at least at 1 GHz in addition to 800 MHz would have been nice; you can’t get a P3 faster than 1 GHz but testing the Athlon chipsets at 1.2 would add to the enlightenment. Why settle for assertions alone when you can have hard numbers?

Also, the assertion “And don’t forget, even though things like DDR, AGP, ATA/100 and other advancements don’t amount to a significant gain all on their own, using all of latest technology may add up to a significant gain,” is interesting, but it’s better if backed up with an example. It’s possible to build two otherwise similar systems, one utilizing AGP, ATA-100 and DDR and another utilizing a PCI version of the same video card, a UDMA-33 controller, and PC133 SDRAM, and see the difference. Unfortuantely you can’t totally isolate the chipsets, so minor differences in the two motherboards will keep this from being totally scientific, but they’ll suffice for demonstrating the trend. Ideally, you’d use two boards from the same manufacturer, using chipsets of like vintage from the same manufacturer. That pretty much limits us to the VIA Apollo Pro series and a Pentium III CPU.

And if you’re ambitious, you can test each possible combination of parts. It’s a nice theory that the whole may be greater than the sum of the parts, and chances are a lot of people will buy it at face value. Why not test it?

This reminds me of a quote from Don Tapscott, in a Communication World interview from Dec. 1999, where he spelled out a sort of communication pecking order. He said, “If you provide structure to data, you get information. And if you provide context to information, you get knowledge. And if you provide human judgment and trans-historical insights, perhaps we can get wisdom.”

This analysis has good human judgment and trans-historical insights. It has context. It has structure. The problem is it doesn’t have enough data, and that’s what keeps this from being a landmark piece. Built on a stronger foundation, this had the potential to be quoted for years to come.

Windows NT on hardware it has no business on

A partial retraction. OK, Southwestern Bell isn’t responsible for all my missing mail. I had a second POP3 client running that I forgot about, which was grabbing some of my mail. But my computer couldn’t find a DHCP server all day, so even though one problem wasn’t their fault, another one was. So I’m still gonna write Casey Kassum with a request and dedication: Todd Rundgren’s “I Hate My Frickin’ ISP,” dedicated to my beloved Southwestern Bell.

Running, uh, no, executing Windows NT 4.0 on a Pentium-75 with 16 MB RAM. Disclaimer: Before you start thinking things that include my name and words like “crack” or “LSD,” let me state emphatically that this was not my idea. I was only following orders. (I’m not on drugs. I’m not nuts–I’m certifiably sane. I’m not even depressed.) All that clear? Good.

That said, the stated minimum hardware requirements for NT 4 are a 486 CPU with 12 MB RAM. And I did once build a print server out of an old IBM PS/2 that had a 486SLC2/50 CPU and 16 megs of RAM. Hey, I was young and I needed the money, OK? Besides, it was a very experimental time and I didn’t think anybody would get hurt…

OK, I’m done turning druggy double entendres.

Needless to say, NT on this machine is anything but pretty. (And I’ll put a marginal machine into service as a server where no one ever interacts with it directly long before I stick one on an end-user’s desk.) The video card in my flagship PC has more memory and processing power. But we’re out of PCs, and this poor girl needs a computer on her desk (though she’s never done anything to deserve this fate), so here’s what I did to try to make life on this machine more tolerable. These tricks work much better on fast machines.

  • Pull out all network protocols except TCP/IP. I also double-checked all TCP/IP settings and made sure the closest DNS server was first on the list.
  • Use a static IP address. The DHCP service uses memory and CPU cycles, and on machines like this, every byte and cycle counts.
  • Remove Office Startup, Find Fast, and LoadWC from Startup. The first two are in the All Users start menu. The last is in the registry. All eat memory and provide no useful functionality.
  • Move the swap file to a second physical hard disk. This machine happened to have a second drive, so I put the swap file there for better performance.
  • Turn off unnecessary services. The Scheduler service and Computer Browser service normally aren’t needed. If the network never sent out notifications (ours does), I’d also turn off the Messenger service.
  • Remove unnecessary fonts. I won’t do this without her present, since I might inadvertently nuke her favorite font. But if she doesn’t use it, it’s gone.
  • Keep free space above 100 megs. Windows slows to a crawl when forced to live on a drive that’s as crowded as a mosh pit.
  • Defragment! Making matters worse, this drive didn’t seem to have a single file on it that wasn’t fragmented. I ran Diskeeper and there was more red on the screen than at a Cardinals game when Mark McGwire’s chasing home run records.
  • When you have two drives, put the OS on the faster of the two. Unfortunately, the OS is on an ancient Seagate 420-meg drive, with a 2.1-gig drive in as the secondary drive. The roles really should be reversed. When in doubt, the bigger drive is usually faster. The newer drive almost always is. I may just Ghost the OS over to the 2.1-gig drive, then switch them.
  • Switch to Program Manager. She’s probably not comfortable with the old Windows 3.1 interface (I’ve only ever met one person who liked it) so I probably won’t do this, but that’ll save you a couple megs.

Yes, even with these adjustments, it’s still awful. So I’m gonna see if I can dig up some memory from somewhere. That’ll help more than anything. But as tempting as overclocking may be, I won’t do it.

Overclocking Pentium-75s

I had an overclocking conversation at work today. A coworker wanting to overclock a laptop. I told him I didn’t think that was a good idea. Then this was waiting for me at home:

From: Curtis Horn
Subject: Pentium-75

Hello, I’m one of your readers and I check your view for the tips you sometimes put up. I’m working on a compaq pentium 75 also and maybe you can do what I did. Overclock the chip to 90Mhz. the way I did this is by changing the bus speed to 60Mhz, from 50. this has speed it up significantly I think because the memory is also speed up. You’ll also get a laugh out of this, it’s a Compaq 972 and it has 8MB of memory — ON the Mother board!! I could not believe it. but it’s there. Luckily this leaves 4 simm slots open, so i can add 4 8MB SIMMs. (16MB SIMMs are way to expensive and I have some 8MB laying around and can buy some more for 10$ each) I convinced the person to buy a 5Gig quantum drive, so they have something they can use when they upgrade. Well hope the P75 you’re working on OCs as easily as the one I have here.

Curtis

http://homes.arealcity.com/cis/

Compaq used to put a fair bit of memory on the motherboard itself. My Presario 660 (a 486/66) has 4 MB on the board. There are a couple of Compaqs from the 900 series still floating around at work that have 8 MB on the motherboard as well. But it’s not a common practice anymore, and I don’t recall any other manufacturer who did that regularly–I remember Compaq doing it because I used to sell Compaqs by the truckload and frequently I ended up adding upgrades to them.

Bus speed isn’t nearly as important in the Pentium Pro/II/III/Celeron and AMD Athlon arena, but in Socket 7 and earlier, you’re right, it makes a huge difference. Remember, the bus speed determines the speed at which the CPU can access the memory and the cache, and as the Mendocino Celeron illustrated, cache speed is more important than CPU speed or cache size. In the early days of Tom’s Hardware Guide, Tom Pabst revealed that a Pentium-150 running at 75 MHzx2 outran a Pentium-166, and a Pentium-166 running at 83MHzx2 outran a P200. So what was the point of buying a P200 if you weren’t going to overclock it, right? Ah, the good old days…. This, of course, was one reason Intel decided to start locking CPU multipliers.

The speed of the PCI bus was also tied to the bus speed. A good Pentium-100 could outrun a Pentium-120 because the Pentium-100 had a full 33 MHz PCI bus while the Pentium 120’s PCI bus ran at 30 MHz. The Pentium 75’s PCI bus ran at a pokey 25 MHz. Nobody wants to slow down their video and disk performance by 10 percent, let alone 25 percent.

Overclocking P75s is risky business though. Intel never intended to make a P75. The problem was, they had terrible yields initially on their P90s, but they found a good percentage of the bad chips would run reliably at 75, so they created the P75 and phased out the P60 and P66. (The P66 was actually a better performer because of the bus speed.) The P75 sold like crazy, and Intel wasn’t going to can a best-seller, so once they got over the yield problems, they still marketed P75s. I’ve heard of people going as high as 133 MHz with P75s. I experimented once with a P75 and took it as high as 120 MHz, but couldn’t get 133 (I suspect people getting to that level may have been increasing the voltage). It didn’t run reliably at 120 MHz for long, though I know of people who swear up and down they got 75’s running at that speed reliably with no special tricks.

Overclocking an old chip like that is fine, as long as you’re aware of the risks and willing to live with them. I’d definitely put a heavy-duty CPU fan on it (like a PC Power and Cooling fan for a high-end K6-2). In my case, I’m more interested in having a PC that’s as reliable as possible. Her life’s plenty complicated enough without having and overclocked P75 to deal with.

And we now have better ways to measure overclocking’s effects. Microsoft doesn’t have a dog in this fight but they see the weirdness.

But thanks for the idea, and for the stroll down memory lane, definitely.