You know, it's hard for me to say this, but... well... I'm happy. And it's not because today is another wonderful Thanksgiving holiday, a day to give thanks to God for all the many blessings poured out upon me and my nation. No, I'm happy because, for one weekend at least, the universe forgot to hate me. I freely confess I haven't a clue what to do with myself in light of it, but I can't avoid the fact that I had a weekend like none I've ever had before when it comes to technology. Here's the short vesrion: stuff worked.
I wrote previously about Gorthaur, the new, gaming-centric computer I've been planning for a while to build. At that time I was stuck in a holding pattern, waiting for all the system components to arrive. Well, not only did they all arrive in a timely fashion—I ordered everything Monday and had it all by Friday night—the components came to my door intact, undamaged, and working.
As such, I was able to build Gorthaur the Cruel last weekend. The set of experiences involved was so rare, so unique, that I can scarcely believe it. I have never had an experience like this with technology before, and I don't think it's possible for me to overstate just how amazed I am at how well everything went. So I'm writing today both to record the unbelievable for future reference and because it seems appropriate to express my gratitude for just such a weekend on Thanksgiving day.
| NB: | In the sections that follow, I provide links to the components and utilities I discuss, the same links I used for purchasing. To be clear, I receive neither money nor any other remuneration or consideration from these vendors. I chose them either because they have previously provided good service (e.g., Newegg.com) or they had the best overall price (TigerDirect.com, Best Byte Computers). I provide such links only as a convenience to my readers, and, as always, your mileage may vary. |
Because of the nightmares I've had in the past, most of which are already recorded in various Terrible Tales of Technological Terror, I had a detailed plan for building Gorthaur. The "short" version of that plan—believe me when I say it was a very detailed plan—was as follows:
I fully expected all kinds of grief at step (1), largely because I have never previously been able to install a motherboard without at least some major headaches. I also expected huge amounts of grief at step (4) because, in my experience, audio cards are positively fraught with driver problems, interrupt conflicts, etc. But what scared me most, to be perfectly honest, was step (7). I have never had a system with two video cards before, largely because I've had so many other problems I've been too scared even to contemplate opening myself up to such woes.
At a bare minimum, I figured the second video card would render the whole system unbootable. Indeed, I was fully prepared to fight some funky motherboard incompatibilities and power/thermal problems just to get the system booting, then fight Windows XP and the video drivers to get them to recognize the hardware and use it properly, then probably fight with every single game on my system to get them all to work. In short, I expected it would take me a month or more to get everything working, and that only after returning/replacing at least a few components along the way.
Saturday morning was the big first step, assembling the basics. My wife and I had company scheduled for that afternoon, so I had but a few hours on Saturday morning to pursue step one. Still, I figured that installing the most basic components wouldn't be too hard. I've built quite a few systems by now, so I know what I'm doing.
Surprisingly, my interlude of joy began with the case, a Cooler Master Wave Master TAC-T01-E1C. I say 'surprisingly' because that was the one component I wasn't entirely happy with from the outset. I had four criteria for the case: it should (1) have a removable motherboard tray, (2) use 120 mm. fans, (3) come without a power supply, and (4) ship with a black finish while being made of aluminum (for good heat dissipation). No one case met all four of these criteria, but the Wave Master seemed like the best compromise. It fully meets criteria (1) and (3) and partially meets (4), but I was unhappy about its use of 80 mm. fans and wasn't wild about the silver finish.
But when I removed the motherboard tray so easily, by undoing four simple thumb screws, I was hooked. I was able to get the short list of minimal components installed in that tray within an hour's time. Even the CPU cooler, about which more in a moment, was shockingly simple to install with the tray removed from the case. I didn't cut myself on any sharp edges, I didn't have to fight to get my large hands into small spaces, and I didn't once drop a screw into some freakishly small crack/corner from which it couldn't be extracted short of blowing up the world. It was bliss. Insofar as it is possible, I will never buy another case without a removable motherboard tray.
For sake of reference, the motherboard in question is a DFI Lan Party nF4 SLI-DR, sporting an AMD Athlon 64 x2 4400+ CPU with a Zalman CNPS9500 heatsink/fan and two 1 GB sticks of OCZ Gold Edition PC-4000 RAM for plenty of system memory. For the initial test, I plugged in only one of the two BFG GeForce 7800 GTX OC video cards I purchased. As my plan indicates, I wanted to start simple and make sure everything worked before I tried to get fancy. My main goal with the basic setup was to validate the memory; I've had memory problems in the past, and I wanted to make sure this memory was beyond any doubt before proceeding further.
With all the heat dissipation issues I've had over the last year or so, I was extremely careful to select a good heatsink/fan and get it installed properly. Thankfully, the Zalman cooler was amazingly simple to install. I was expecting some ponderously byzantine process, but it turned out to be nothing more than removing the screws from the plastic fitting around the CPU socket and threading them through the bracket that holds the cooler in place. I gather it's a wee bit more difficult with socket types other than the 939, but it was literally a two minute job with my particular setup.
Once I had the motherboard tray loaded, I hooked it all up to the PC Power & Cooling Turbo-Cool 510 SLI power supply unit (PSU) and promptly discovered two things worth raving about. First, the PSU supplied all the connectors I needed right out of the box. I'm so accustomed to needing adapters for the video card, front panel gadgets, fans, and so forth, that it never occurred to me it might have custom connectors for NVIDIA cards and SATA drives. And second, the DFI motherboard has a wonderful feature for enthusiasts: power and reset switches built onto the motherboard for ease in testing outside the case. I'm so used to fighting things that those two, minor innovations threw me for a loop.
Once I was ready, I triple-checked everything for safety's sake and hit the power switch on the motherboard. The chipset fan was clearly working—the nForce4 chipset must run pretty hot because the fan spins at something like 7,000 RPM virtually all the time!—as was the fan on the video card. The machine even made it through its power-on self test (POST) cycle without any issues. It didn't boot, of course, because I didn't have a floppy drive hooked up yet, but the motherboard checked itself out nicely the very first time I applied power.
The thing that scared the living daylights out of me was that the CPU fan wasn't turning at all! I knew I had hooked everything up correctly—I find triple-checking eliminates virtually all possibility of error on my part; I often screw up once, sometimes twice, but I virtually never make the same, stupid mistake three times in a row—but somehow the fan on that Zalman cooler wasn't turning. Still, the motherboard wasn't giving me any kind of error beeps, certainly not the awful wailing of thermal alarms, so I figured I would check the temperatures on its BIOS screen before I freaked completely.
What I found stunned me utterly: the CPU fan wasn't turning because it wasn't needed! The DFI motherboard has, in addition to all its nifty overclocking tweaks, a surprisingly robust set of fan controls. By default, the CPU fan isn't activated at all if the CPU temperature falls below a user-configurable threshold. I could hardly believe it, but my CPU fan wasn't turning because the CPU wasn't hot enough to need it. Seriously, the reading for the chipset and surrounding area was hovering around 35° C while the CPU temperature was hovering comfortably around 37° C.
I literally couldn't believe what I was seeing because of my experiences over the last year. My old computer has an Athlon XP 3200 CPU with a monstrous, noisy ThermalTake CPU heatsink/fan that runs around 4,000+ RPM. Further, the case has another powerful and noisy fan in the front, two fans at the rear, and one fan at the top with the PSU. The case temperature is typically around 38° C, but the CPU usually idles around 55° C. It gets up to 60° C (or higher) when I'm gaming, and that means a guaranteed, never-ending stream of crashes and reboots with Battlefield 2 and F.E.A.R. unless I turn down the front side bus (FSB) from 200 MHz. to 185 MHz.
When I next hooked up the floppy drive and booted the Microsoft Memory Diagnostic, the temperatures happily remained so constantly low. No matter what I did, the CPU temperature didn't change much at all. As I watched, the CPU fan would activate occasionally, turn for a few seconds, and then stop completely. After checking and re-checking for nearly half an hour, I figured that there must not be any danger of burnout after all. I could scarcely believe what I was seeing, but Gorthaur clearly wasn't having any thermal issues. Heck, the CPU temperature was actually reading lower than the surrounding air; that didn't make any sense to me, but the data wasn't changing.
I made a note to double-check this later, figuring that things would probably change once I got the whole system assembled and packed into the case. All of my data to date had been taken with the components sitting out in the open air in the motherboard's removable tray, after all, and those temperatures seemed far too good to be true. By this time our guest had arrived, however, so I fired up the memory diagnostic in its extended testing mode and let it run while I enjoyed the rest of the day doing other things.
I'm happy to say that later, when I checked the system before going to bed, Gorthaur had run half a dozen or so passes of all eleven memory tests, some of which are really quite intensive, more than half a dozen times without a single error. That's the best result I've ever seen, and, coming on the heels of such a simple build process, I was completely shocked by it. Not even one bit wrong in so many reads and writes?! How could that be?!
You have to understand, folks, that things simply do not work like that for me. No, I'm always the "lucky" one who gets the motherboard with the bad capacitors, the PSU with the out-of-spec voltage ripple, the memory with a few bad bits (which usually require months upon months of life-mangling, spirit-destroying suffering to troubleshoot), the CPU that has a low tolerance for error, etc. But in this case, not only did all the components arrive on time, not only did they go together with the greatest of ease, they actually worked together and ran at lower temperatures than any configuration I have ever seen! That's just not what I expect when I build a computer.
Still, I figured the real test would come the following day. I mean, I had assembled the most minimal system possible for testing, and, sure, the memory had checked out, but that wasn't a real stress test. No, the real pain would surely come when I got Windows XP installed. Something would surely fail. I'd start getting reboots at high CPU utilization, the hard drive would have bad sectors, the optical drive would misread; something would have to go terribly wrong because that's what always happens. That's what has happened to me, practically without exception, for... well... as long as I've been using computers, and that dates back a few decades. As I've said so often, it couldn't be my life if it didn't suck.
Nevertheless, I shrugged off those thoughts and went to bed happy. Whatever terrible technology gremlins might yet plague me, I had just enjoyed the very best computer-building experience imaginable. What's not to love about that? I drifted off to sleep, figuring there would be plenty of time for the universe to resume hating me the next day.
After attending church the following morning, I settled down for a long day's work on Gorthaur. Even as awesome as the previous day's events were, I had completed but one of the nine steps in my plan and much remained to be done. So, I hooked up the Maxtor DiamondMax 10 300 GB SATA hard drive and the Plextor PX-740A DVD-RW drive, and I started installing Windows XP with service pack two. Formatting the hard drive took over an hour, but I stayed busy doing other things. Besides, I figured it was time well spent; doing a quick format saves time up front, but it can allow a host of problems to go undetected that will cost a lot more later.
The installation went as smoothly as any I've previously done. I'm a pro by now, you see, because I've had to install and reinstall Windows XP so many times. In just over half an hour, the operating system (OS) was installed and running. It didn't have any network access, audio, etc. because I hadn't installed the chipset or video drivers, but that was exactly what I wanted: a purely vanilla installation from which to make a drive image for use on other machines in the future. I installed Norton Ghost, used it to make an image of the drive, and proceeded to the next step.
Once that process completed, I powered the system down and installed the Creative Labs X-Fi Fatal1ty audio card. As usual, the drivers for that piece of hardware took forever to install, but I have to say this is the most painless audio card from Creative Labs that I've installed yet. Unlike most other audio cards, this one didn't screw up the system in any way or make me long for the "bad old days" of DOS, when I could simply edit a text file to get it working. No, the drivers sure took their own sweet time to install, but everything worked as it should right out of the box.
While I was at it, I installed the .NET framework, the latest chipset and video drivers, and went through the far-more-painful-than-it-should-be process of getting all the latest updates for the audio card and for Windows XP. And believe me, I was cringing every step of the way, expecting something to blow up in my face at any moment. I mean, come on, people, an audio card doesn't just install and work properly; the mere suggestion that it might is pure crazy talk. I'm positively flabbergasted to report that nothing went wrong. The dozens of updates installed and seemed to work just fine.
After picking up my jaw from the floor, I created another image of the hard drive, for future rebuilds of Gorthaur with its current hardware, and then started benchmarking the system. Before I give you the raw data, let me first explain the testing methodology. The basic steps I've been using for a while now remain unchanged: reboot the system, log on to Windows XP, then run the utilities one at a time. I know some people prefer to reboot the system between each benchmarking utility, and perhaps that is more "pure", but I figure benchmarking takes too long as it is. If you want advice from a purist, go read a site that accepts advertising revenue.
Me, I do this stuff in my very limited spare time, so I'll stick with consistency: running the same tools in the same order every time. It's not like Windows XP stands out in the first place as a remarkably deterministic OS; i.e., one that boots identically every time. I've seen far too many freaky things to count, things "fixed" by yet another reboot, so I figure the practical differences between the purist's approach and mine have got to be pretty negligible. At any rate, I currently run the following tools to benchmark my systems in the following order:
One through three are "synthetic" benchmarks; i.e., they are designed deliberately to exercise certain features and functions of a system, but they bear no necessary relation to real-world gaming performance. In contrast, four and five are actual games that give a pretty good idea of how the system will perform with the two most-used graphics APIs in the two most-licensed engines in the gaming universe (OpenGL in Q3A and DirectX in UT2k4 respectively). Finally, F.E.A.R. just happens to be the most system-killing game engine I've ever seen, so I've included it for that purpose: to beat the living daylights out of my new hardware and see how it responds.
In all three of the games, I run the benchmarks at 1600 x 1200 x 32 bpp with all of the options cranked as high as they'll go using the in-game controls. In the case of F.E.A.R., that includes 4x anti-aliasing (AA) and 8x anisotropic filtering (AF) as well. Hey, it's the system killer, so why not? Even if Q3A and UT2k4 run well, F.E.A.R. can still bring a system to its knees. Finally, I use the DEMO001 demo for Q3A, the Flyby-Inferno benchmark for UT2k4, and the internal "Test Settings" option for F.E.A.R., which give not only the minimum/average/maximum frame rates but also report percentages framerates at three key inflection points. Without any further adieu, the raw data for the single video card are as follows:
| Benchmark | Single Video Card | ||||||
| 3DMark®03 | 17,372 | ||||||
| 3DMark®05 | NA | ||||||
| Aquamark 3 |
|
||||||
| Q3A | 345 FPS | ||||||
| UT2k4 | 199.2 | ||||||
| F.E.A.R. |
38% below 25 FPS |
For the record, the 3DMark®05 score is listed as being "Not Available" because I couldn't get the benchmark to complete. It would run the first couple of scenes just fine, but then it crashed back to the desktop. I don't know why, and I didn't want to take the time to figure it out. I was interested in getting benchmarks for a single video card largely as a way to verify that SLI was working and gauge its performance improvement (assuming I could get it working).
Having made it through two thirds of my plan, and largely without so much as a momentary glitch, I figured it was time to pay the piper. I was surprised that nothing had gone wrong with the most basic system, and I was all the more stunned that the audio card didn't screw things up somehow. But now I stood at the precipice of doom: adding a second video card and enabling SLI would surely take me screaming into the abyss.
Or at least that's how I approached it. I powered down the system, pulled the existing video card, and switched all the jumper blocks. Unfortunately the DFI motherboard, wonderful as it is in other respects, requires the user to switch blocks of jumpers to enable/disable SLI support, and they're not at all simple to switch. Yes, I know a tool is provided with the motherboard, but I couldn't get it to grip the darned things. After a tricky five minutes of labor (using a pair of forceps instead), I was able to slot both video cards and attach the SLI bridge (a small, black plastic thingy included with the motherboard) to the two feature connectors at their tops.
After triple-checking that everything was ready, I powered the system on and waited for the worst. Gorthaur made it POSTed just fine, after which it sat at the Windows XP logo for quite a while. The progress kept moving, so I hoped I wasn't completely out of luck. I'm not sure what it was doing, but I wasn't about to interrupt it, not while I had any hope of it working. It took a solid two to three minutes before anything else happened.
And then the screen went black/blank. I didn't see anything obviously wrong, and I didn't get any error beeps or other indicators. It certainly didn't seem like the system had failed or restarted; no, I could clearly hear the hard drive spinning merrily away, seemingly continuing to load Windows XP. But I wasn't seeing anything more than a black/blank screen. I confess I was starting to think my fears might be confirmed, that something awful had finally gone wrong; a hard reboot of the system seemed like my only option, but I was afraid that doing so might screw something up in mid-configuration.
Thankfully, my caution paid off in the form of an idea that saved me a world of hurt: I connected the monitor to the second video card. As soon as I did this, the regular welcome/logon screen appeared as usual. When I subsequently logged on, the NVIDIA icon in the system tray popped up a message asking if I wanted to enable SLI, and I said yes! Unfortunately, it gave me some kind of error I didn't understand at the time—something about how my desktop configuration was incompatible with SLI—and it didn't bother telling me that I needed to reboot the system for things to work properly. I also discovered, after the reboot, that I needed to switch the monitor cable back to the primary video card. But beyond those hurdles all was good.
I may now understand some of what happened. When I added the second video card, I'm betting that the OS, or perhaps the NVIDIA drivers, assumed that I wanted to do some kind of dual-display thing, not SLI. As such, I'm guessing that what used to be the primary display went black/blank because the new card was now the "main" desktop, or something like that. I'm also betting that I had to shift the monitor cable back to the primary display after getting SLI enabled successfully because the OS, or again the NVIDIA drivers, then "knew" how I intended to use things.
That explanation is lacking a lot of details, but I'd be willing to wager that my problems were due, in the final analysis, to the differences between dual-head and SLI modes of operation. If anyone out there has substantive knowledge about such matters, please drop me an email and let me know. I'd be curious to hear from anyone with a clue on this subject. It's worth learning.
At any rate, once I had verified that SLI was indeed enabled and working, I took the final step of packing it all into the case. This was, surprisingly, the most difficult of all the steps in my plan. Despite having the skinny SATA cable for the hard drive, and the nifty, circular-ized IDE and floppy drive cabling provided by DFI, it still wasn't easy getting everything hooked up and all jammed into the case. I'm not very happy with the wiring layout as it is, so I'm planning to rearrange it all for better air flow in the future. But it was already nearing 23:00 hrs. by the time I had made it that far, and I desperately wanted to try the new system and still make it to bed by midnight.
With my eyes growing increasingly bleary after the long day's work, I got the system closed up and hooked into my sound system. I then ran through the same benchmark suite as before and achieved some odd but interesting results. The following summarizes the data and includes the data taken previously to make it easy to compare.
| Benchmark | Single Video Card | SLI Enabled | ||||||||||||
| 3DMark®03 | 17,372 | 28,127 | ||||||||||||
| 3DMark®05 | NA | 11,460 | ||||||||||||
| Aquamark 3 |
|
|
||||||||||||
| Q3A | 345 FPS | 345 FPS | ||||||||||||
| UT2k4 | 199.2 | 19.0 | ||||||||||||
| F.E.A.R. |
38% below 25 FPS |
00% below 25 FPS |
I don't understand at all why 3DMark®05 refused to run with one card but works great with two. And I really don't understand why Q3A shows no difference while UT2k4 is clearly mis-reporting the results; its demo absolutely tears up my eyes with its raw speed, despite the impossibly low score. But you know what? I don't really care much either, as long as actual gameplay is improved.
And good gravy is it ever! The first time I ran F.E.A.R. I couldn't believe what I was seeing. Even though the benchmark data doesn't really reflect it—the "Test Settings" feature is pretty optimistic compared to actual game play—the game was downright unplayable with the one video card. The game remained a slide show at times, despite having such cutting-edge hardware. With two video cards and SLI enabled, however, the game positively smokes, the clearest example I've seen to date of a video-limited game.
The bottom line is simple: I can now play F.E.A.R. at 1600 x 1200 with every setting cranked to the max, including 4x AA and 8x AF, without even the barest hint of slowdown! I bypassed the opening movies and cut-scenes and played a portion of the first "interval" with SLI enabled, and it was gorgeous. It was easily the best looking video game I have ever seen, without the slightest evidence of jaggies and smoothly blended textures as far as the eye can see. Let me tell you, folks, a gamer could get really comfortable with this level of quality.
If the two benchmarks that changed can be trusted, SLI in a 62% improvement in synthetic benchmarks and a whopping 66% - 224% in actual games (depending on which of the F.E.A.R. numbers one uses). From what I've seen since in other games, it looks to me like I'm averaging 80% better performance with a second video card. A gamer could get really comfortable with this level of performance too.
Of course, nothing on this Earth is perfect. I have encountered a few problems with the system, having tested it briefly with almost two dozen games in addition to the benchmark applications. Thankfully, the problems can be summarized by the following short list:
As to (1), I find that enabling audio hardware mixing works just fine, so perhaps there's some kind of funky problem in F.E.A.R. when using software mixing with EAX features. I don't know; I just know that enabling audio hardware mixing works fine. As to (2), a bit of Internet-forum crawling lead me to a workaround: disabling the native command queueing (NCQ) option on the SATA controller gets rid of both the boot delay and the errors. It makes the drive run noticeably slower too, but that's not a big deal. I'm happy to report that (3) is easily addressed by downloading some tweaked textures (as described in this forum post). Apparently the game has a problem with the way it handles DirectX device capabilities, and some clever person came up with that nifty workaround.
Sadly, the only issue that can neither be fixed nor worked around is (4). No matter what I do, anti-aliasing causes the freaky connection issue described above. This would have been a deal breaker a couple of years ago, when RS was my favorite multi-player game, but it's not such a big problem these days. I still wish NVIDIA would fix this problem, but I can live with it as it is. I'm especially pleased to see that the glowing, green artifacts in Prince of Persia are gone. All things considered, NVIDIA has done a lot to clean up their drivers since last I wrote about their many failings.
It has been less than a week since Gorthaur first took physical form, and I'm still downright bubbly with exuberance at how magnificently everything has worked out. I have never in my life, and I mean never, previously had an experience this trouble free with computers and technology. The case, despite its 80 mm. fans and silver coloring, is well designed and great to use. I've even grown to enjoy its look. The power supply is magnificently robust and provides all the connectors I need without any adapters. The motherboard, CPU, and memory all came together perfectly and work beautifully and run cool. The video cards and audio are the very best I'm capable of imagining.
I can't believe I'm saying this: I didn't have one substantive problem in the entire process. That's simply unheard of in my experience. I ran into two or three extremely minor glitches, and that's it. True, my benchmarking tools don't all seem to give good data, but why should I care? My new system has chewed up and spit out every game I've installed thus far. It runs them all staggeringly well, barely breaking a sweat it seems in light of the consistently cool temperature readings. Hell, even under positively crazy stress testing, the system runs responsively, snappily even, without anything even vaguely resembling a complaint or a mistake.
Honestly, this is the very best money I have ever spent on a computer. I got almost exactly what I wanted in every respect, it all went together without a hitch, and the resulting system has outperformed my expectations in every respect. Even the most demanding games I own positively quail before mighty Gorthaur the Cruel, who delivers what I thought would be impossibly smooth framerates at high resolution with maximal image quality enhancements. Gaming just doesn't get any better than this, folks!
I keep having this nagging tug at the back of my mind, the weight of my personal computing history telling me that this can't be happening, telling me that the sky will indeed come crashing down soon enough. Perhaps the universe will start hating me again at some point. At the moment, though, I'm just too happy to worry about it. All hail the mighty Gorthaur the Cruel! Long may he bend software in this (Middle) Earth to his malevolent will!
Given this set of experiences and today's holiday, I have to close with a note about how grateful I am. I have had so many problems with technology that I approached this entire process with a mixture of excitement, dread, and outright loathing. I literally got down on my knees and prayed to God for wisdom and discernment in choosing components, patience in assembling them, and whatever good fortune he would grant. I'm not taing it as coincidental that I had a nearly miraculous experience.
To be clear, I'm pretty cautious about identifying God's involvement in the world, despite my Christian belief, because there is always the question of religious epistemology: how can one know with any certainty that God has been involved? The honest answer is that one cannot, short of God revealing himself directly. But I'm not going to fall prey to the trap C.S. Lewis identified either; i.e., I'm not going to blame God when he doesn't answer but deny him the credit when he does and I'm simply too limited/dense to see it. No, today I thank God, for building Gorthaur has been a nearly miraculous success, whether or not he intervened directly in that process.
May all have as blessed a Thanksgiving as I enjoyed.
11/24/2005