|
Overclocking ReportDear friends, you may have noticed, I have been a little pre-occupied lately, and part of the reason has been my efforts to build a new computer. With this article I would like to share some of my experience. Read it, learn from it, suffer and enjoy. |
Originally I wasn't really planning to build a new computer. What I meant to do is to upgrade my old computer, PPro 200 with 66 MHz FSB on a Gigabyte 686FX, 128 MB EDO Ram, S3 Virge graphics card, IBM 8GB hdd (IDE) and a SCSI subsystem for DVD drive, internal ZIP and external scanner. However, I soon discovered that due to the intel socket 8 and the Baby-AT form factor, this would have meant to throw away almost the complete computer. I didn't want to do that, since it is still perfectly good, with some memories attached.
Thus I decided to put the old 16x CD-ROM back into it and transfer the SCSI subsystem and the network card to a new machine I had yet to build. Unfortunately the downgrade turned out to be quite time intensive because the ancient CD-ROM did not support busmaster DMA and in the process of adjusting the operating systems I made a mistake which meant I had to re-install everything. But that's a different story.
For the records (and due to it being mentioned below) I would like to add that I own another computer, a Pentium II 400 running at 124 MHz FSB on an AOpen AX6BC Pro, 128MB SDRAM, Matrox MGA-400, IBM 10GB hdd and Seagate 2GB hdd (which you might guess came with the PPro originally), DVD-ROM and CD-Recorder, all devices IDE.
After everything came together, this is how the new machine looks like:
| Mainboard | ASUS CUBX |
| Memory | 256MB Crucial 133 CL 2 |
| Cpu | Intel Pentium III 700E FCPGA, SL45Y |
| HSF | Golden Orb, lapped |
| Graphics | Hercules 3D Prophet GeForce 32MB DDR |
| Sound | Sound Blaster Live! Value |
| Network | Linksys Ethernet PCI 10MB/s |
| Hdd | IBM 27GB 7200rpm (U-ATA 66) |
| SCSI card | No-name with Symbios 53C810 chip |
| DVD | Pioneer 6x (SCSI) |
| ZIP | Iomega internal (SCSI) |
| Casing | H-24A |
| PSU | Enhance ATX 250W (temp. controlled fan) |
Some of you might ask, why did I choose the BX chipset, which is more than two years old? Well, BX boards are extremely mature and stable, are recognised by virtually all operating systems, have a superior memory performance and, despite opposite statements, overclock well, at least if you do moderate overclocking. And I had no illusions about the overclockability of the 700E.
I went one step further and connected the hdd to the U-ATA 33 controller rather than the U-ATA 66 which is on the board as well. But stability and avoidance of driver issues are more important to me. OK, it is also a fact that there is no application benchmark where U-ATA 66 makes the slightest difference.
Same goes for AGP 4x, nice feature on paper, but zero impact on application performance.
This proved to be less easy than anticipated. The memory I ordered some time ago over the internet from www.crucial.com, without checking the Australian import regulations first. I was very surprised when DHS rang me up and asked for a sales tax number. I didn't have any. Then you pay 22% sales tax, they announced. As it turned out, there is a bit more to the actual calculation plus a fee, which in the end adds up to a 34% surcharge. Well, it is a historical fact that Australia started out as a colony of convicts.
Mainboard, HSF, graphics, sound and hdd I bought in Sim Lim Square, Singapore. That's a good place for the enthusiast, but although Singapore didn't start as a convict colony, it started as a trading post and that can be as bad... Only joking, what I mean to say is, you really need to know your stuff, otherwise the sales people in Sim Lim Square sell you all the leftovers they happen to have in stock.
I then had to use up the tax free allowance of the whole family [cough] to bring the stuff into Australia, stashed away in a suitcase. Apart from the hdd, which came as hand luggage.
I still had to buy a case with PSU to get started. That was difficult. I went into numerous (smallish) computer shops in Melbourne, but they sold by the price, ie. the cases and in particular the PSUs were crap. I lost one mainboard in the past due to the PSU, never again. Luckily AOpen replaced it under warranty at the time. Perhaps that was a reason as well to order an AOpen HX45A case from somewhere near Sydney. It took another week to get the case delivered, and a long weekend came up, so I decided to build the new computer using the case of the PII-400. The PII will go in the new case, so all my AOpen components will be united. I am obviously hoping for some synergy, or good Feng Shui or so... ;-)
This went basically OK. When I say basically, I mean, I didn't spend weeks and months to solve most problems, like I did with my network starter kit from Linksys last year: I literally spent night after night on a freak problem (looked like a TCP protocol problem), until I replaced the Linksys drivers with Winbond drivers.
My first major activity was to disassemble the Golden Orb, which was quite hard and involved filing off some material from the metal pins sticking out underneath. The Orb looks actually quite cheaply made, "tinny" if you know what I mean. The bottom surface had a dent right in the centre of the contact area with the flipchip and the rest of the surface was not very flat either. Anyway, I fixed all that.
For a first test I put the case, board, memory and GeForce together, connected monitor and keyboard and switched on. I was quite happy to see the components were working! From here it could only get better. I played with the BIOS setup and checked the fans were running. Since the Orb is completely hidden under the PSU, I carefully lowered a finger to where I thought the centre was. It gave a little thump and I knew I hadn't quite hit the centre. Anyway, I didn't think anything of it. It was a minute later when I discovered I had blood all over my hand! I wasn't badly hurt, but it gave me a bit of a shock. My shock grew significantly when I took the board out of the case and I noticed a fin had broken off the fan inside the Orb! I didn't take any photos, so take my word for it, OK? I managed to glue the fin back on, but I am not sure how long it will hold. I might contact Thermaltake for a replacement fan, but I suppose at the end of the day it might be cheaper to get a new one in SLS...
Another little problem was the second power connector in 3.5" floppy style that I needed for the ZIP drive: I carefully pushed the pins out of an hdd power connector, soldered a floppy power connector to them and pushed them back in. No big deal, but takes time.
After putting everything together I set up 100 MHz FSB and started the software setup. First I created three primary partitons (2GB, 2GB, 22GB) which involved switching between DOS and Linux. Next I set up Windows 98 in the 22GB partition and installed drivers and benchmark programs. With this minimal setup I intended to establish the limits of overclocking, before completing the system setup. Since then I haven't progressed much further in terms of system setup... Btw, I installed the Nvidia detonator 5.22, not sure whether that was a good choice.
Fuelled by encouraging reports in www.hardwarezone.com, www.overclockers.com and other sources, my aim was to reach 133MHz FSB which means 933MHz CPU clock. To cut it short, after spending a lot of time I am now aiming for 124MHz which is a challenge already. But one after the other.
I went straight for the 133 setting, which resulted straight in a hung computer - the electrical power supply had to be switched off and on. Next I tried 124, and slowly realized what I had read before: the BIOS has the habit of setting Vcore last thing before it gives control to the boot mechanism, so you basically have to complete a boot before the new Vcore is effective. So I had to boot at 100MHz with Vcore 1.85V, reset and select 124 to get it running... At the highest Vcore (and 3.6V I/O voltage) I managed to boot Win 98 at 133 FSB, but it crashed immediately. With that I am quite certain, that my CPU cannot sustain 133 FSB stable. Perhaps with active cooling, but that is beyond me. I am definitely aiming for 124 MHz now.
In hindsight, the same die would have worked better for me with a 6.5 multiplier. Not all dies are the same, so I am not necessarily saying it is better to buy the 650E instead of the 700E, but for me it works out this way. What I am trying to say is, 868@133 is better than 868@124. Anyway, as you can see below, the system is still kicking ass at 124 FSB.
After one morning of no problems, 3D Mark 2000 started freezing which made me despair because I tried and tried and in the end I went to 100 FSB and it still happened. I discovered the GeForce had come out of the slot by a fraction of a millimeter, I pushed it back in and it worked again.
I am now test running 124 FSB at a Vcore setting of 1.75V (in reality 1.79V). This setting performs exactly like 100 FSB at 1.65V as far as stability is concerned. And this is the sad part: on both FSB 100 and 124, 3D Mark 2000 will perform an illegal operation and terminate after some time, say, 2 hours. I am not sure whether these problems are hardware or software (driver) related, but I recall some people have reported similar problems in the past... After one of these crashes, BurnInTest 2.0 also crashed, in particular it was the 3D Graphics test that hung. This makes me wonder, whether it is the GeForce giving me problems here. I can't really blame the overclocked AGP since it happened at 100 FSB as well.
I had a couple of weird experiences with the GeForce: Once the video BIOS reported 16MB instead of 32 during booting, and all benchmarks were substantially slower than normal. After the next restart the problem had gone. Another time, after a 3D Mark crash, the GeForce was in disarray and created garbage when doing bit block transfer operations on the screen.
Benchmark Results
OK, some of you are only interested in the numbers, I guess. Well, I don't have the time to do a proper line of application benchmarks, so I can only give you some of the popular synthetic benchmarks. But that is all you need to put the performance into prospective, I suppose. Here we go. First WCPUID:
Next the SiSoft Sandra CPU benchmark.
This is the SiSoft Sandra memory benchmark.
The result of the 3D Mark 2000 default benchmark (1024x768x16). Note that the GeForce has not been overclocked (other than the AGP).
Unlike my other two computers, this one will get warm after a while. These are the readings at 124 FSB:
As you can see from the screen shot, the CPU idles at 20 to 30 degrees, but reaches 40 to 50 under load. I also get fairly high readings from the motherboard probe, up to 30 degrees. I am sure you have to take these readings with a pinch of salt. For instance, on the screen shot the CPU reading is below the motherboard reading, which doesn't really make sense. Nonetheless, it does worry me a bit, since it is quite cold in Victoria at this time of year, and my study is not heated. It is also quite possible that the stability problems are heat related.
I am not in the same position as a professional hardware reviewer in the sense that I do not have access to many components, but I will share my experience nonetheless.
First thing is the GeForce. I am still in the process of getting used to it, but compared to my Matrox MGA-400 you notice two things straight away: It is significantly faster, and the images don't look as good. There's no bump mapping, the mip mapping sometimes creates obvious stripes of different textures, and the overall impression is just not as good. But it kicks ass, the 3D Mark helicopter ride in high detail is a mere slide show on the Matrox, whereas the GeForce shows a fluid ride. As mentioned above, the GeForce hardware or software might be responsible for some crashes I experienced. To me it looks like there was only one goal in designing the GeForce: frame rates. Everything else has been given lower priority.
The board is of outstanding quality and design, but the BIOS has driven me up the wall. The habit of setting Vcore too late I have described above. If you reset or switch off before the boot process, you end up in the recovery screen next time. Sometimes you get a popup to select the boot device, but I haven't figured out when and why. Also noteworthy, the Vcore measurement is always a little higher than the actual setting, resulting in "errors reported by the hardware monitor" if you exceed a certain limit.
The Golden Orb is definitely a low cost product that requires some mechanical attention. It also needs to be lapped before use. How the cooling performance compares to other HSFs I cannot tell, the initial reports were very good.
It appears, my Pentium III 700E does not overclock too well. But I haven't ruled out the GeForce as a cause yet.
I have described my bad experience with the drivers of the Linksys PCI Ethernet Card above. The Linksys support is not much better, I have never received a reply to my emails.
I can't say much about the Crucial memory other that I have never gone to its limits.
The Sound Blaster Live! Value is just that: a great soundcard at good value.
The SCSI card and devices are just perfect, I've never had any problems.
The Enhance PSU seems to be of high quality and quite powerful. I like the temperature controlled fan, although the new machine produces so much heat that it doesn't really help cutting down the noise level.
It was an interesting experience building this machine, since it was the first time I did it completely from scratch and completely on my own. I would quite happily settle for 124 FSB. The only thing I expect from a computer is 100% stability, and I am not confident I have 100% right now - even at 100 FSB. It is too early to say what the problem is, and more experimenting is needed, but my feeling is the GeForce. If any of you has the same problem, I am very interested in sharing our experience, and perhaps finding a solution for it. Just drop me an email.
That's it for now, but I might follow this up with another article.
Martin
When the new case (AOpen HX45A) finally arrived, I decided to move the new machine into it, since the position of the PSU and the air intake is best suited for the CUBX/P3/Golden Orb combination. The other case is ideal for the AX6BC/P2/Global Win combination.
I also swapped graphics cards and did some stability testing. Interesting result: whichever computer has the Hercules 3D Prophet DDR shows the instability problem, with whatever version of the Nvidia driver. This indicates to me that the GeForce is the culprit.
Back in the CUBX, I started overclocking the GeForce a bit. Could get to 135/340 which gives me 5500+ in 3D Mark 2000. It didn't solve the instability problem, though. ;-)
| < Home | Part II > |