Eternal Battle Day 1: Ultimate Gaming Desktop June 26, 2005

Introduction

Page:: ( 1 / 14 )

Step 1: Build the ultimate gaming system with a no-budget, but don’t waste money approach
Step 2: Build a mission-critical workstation with a focus on stability
Step 3: Put both systems head-to-head in the first biennial Eternal Battle.
Step 4: In classic videogame fashion, reveal the true form of the final boss and continue the battle

Our results might surprise you…

Part 1: Building an Ultimate Gaming PC

Life is about struggles. As a kid, the struggles were simple. It was all about getting that kite to fly or whether you wanted vanilla or chocolate ice cream. As you grew up, it got a bit more complicated as you started to have responsibilities not only to yourself but to others, and one of the battles we all face daily is choosing between work and play. Admit it, if any of us won the lottery, we’d certainly go for an early retirement or at least cut back on our hours at work. This debate between work and play is reflected every time you build a computer. For any given budget, we’re always balancing our need for performance, reliability, and the realities of our budget. With an unlimited budget, we certainly can go all out and get the very best of products, but few of us truly have an unlimited budget…

For our Summer 2005 season of System Building articles, we’re actually going to publish five different articles, one each day! After we go over how these systems were built, we’ll pit them all against each other in one giant “John Woo” style shoot-out.

I’m going to start by making good on my age-old promise to walk through the creation of an ultimate gaming system. Our philosophy remains largely the same – build a killer system with a balanced approach to component selection. Our goal is not to have the fastest benchmarks alone using things like liquid nitrogen cooling. We’re actually running everything at spec, but our goal is to produce an enjoyable system to own, use, and of course, show-off. It’s not going to be cheap, and as always, few of you will actually buy every component we select, but the point isn’t to build an identical system to what we’re doing, but instead to check out our approach and see where you might want to splurge for a flagship component and where you might not.


We’re going to do something extra special this time too – I already said that we’d be building multiple systems, but it’ll be an unfair battle between unequal designs. We’ll have our flagship gaming system with performance, but we’ll also be building a high-performance workstation with maximum stability. Along the way, you’ll learn what makes workstations different from desktops and the differences in style and approach to system building. You already know that the gaming system will outperform the workstation for games, and that the workstation will outperform the gaming system for work – but just how big is the difference? What about low latency versus high-bandwidth DDR? PSUs? This isn’t just a short list of components, it’s all whole week of tutorials and the thought process we put into building our systems.

Of course, the above only describes three days worth of articles – the two system building articles and one benchmarking article. I’ve already said that there are going to be five articles in total; you guys will have to figure out what the surprise will be. Think like a videogame screenwriter, and you’ll figure it out.

This week will aggregate everything I’ve discussed in one form or another over the decade I’ve been writing. This will help make these articles a single-source reference – know this and you’ll be at the top of your system building game.

So, now without further ado, I present to you with our first article: Building the Ultimate Gaming System.



SIDEBAR: The reason we don’t adopt exotic cooling technologies is that these ultimate system guides still need to be a little bit practical.

Table for Two?

Page:: ( 2 / 14 )

Dual versus Single CPU

The debate between single and dual CPUs will be resolved by next year. Everyone will need two cores in the future, and the question will change to who needs four CPU cores (through a pair of Dual Core CPUs).

We’ll soon see desktop dual-core CPUs from AMD and by the end of the year, games will begin to take advantage of those multiple cores. When this happens, not only will games run faster on regular gaming desktop motherboards, but you’ll be able to enjoy the nearly 100% performance boost from parallel applications such as digital photography programs, and enjoy the faster day-to-day response from having a free CPU always ready to respond to user input and clicks. After all, no matter how intellectual and academic you want to be about Windows XP bloat and how silly it is to need 2 CPUs for maximum responsiveness for surfing the web, it doesn’t change the fact that faster hardware still makes Windows XP run faster. You can lobby Microsoft to design more efficient operating systems or simply be more pragmatic and solve the problem with faster hardware.

Dual-core for games today

Today’s games aren’t multithreaded. So, when designing a gaming system only one CPU core is needed. Therefore, the fastest individual core is going to be what’s important for having the fastest frame rates and the fastest benchmarks. In real-life, when you’re playing a game, your CPU still needs to spend time managing memory, the swap file, all while keeping your real-time anti-virus file scanner and firewall active. Everyone claims to run a clean system, but how many of us have been dropped out of a LAN game because we received an instant message? How many of you have a torrent downloading in the background while you game? Dual core CPUs would help there.


The other issue is the ability to multitask while gaming. With Dual Core CPUs, it’s possible to encode a DVD while playing a single-threaded game at competitive frame rates (you’d expect a small drop in performance from less memory bandwidth being available). The obvious question is how many gamers actually encode MPEG-2 while playing games on a day to day basis? It’s an easy answer: none. This brings up an interesting question though. Is this something that is not done because there’s no demand? Or is it something that we’ve convinced ourselves that we don’t want or need, because it’s impossible?

Here’s the analogy: How many of us in California will catch a flight to Japan to spend a day shopping and enjoying fresh sushi and then fly back home in the evening? None. It’s too expensive and you waste too much time on the plane anyway. Now, fast forward 20 years and imagine that all that scramjet technology really pans out, and so we’re looking at a short 2 hour flight, and then imagine that it’s doable at Southwest or JetBlue budget pricing where you can do it for $69 round-trip. It changes things doesn’t it?

Want another analogy? How much free HDD space do you have? Isn’t it odd that no matter how big or small your HDD is, it quickly fills up to say 90% of capacity? When you have more HDD space available, you will find ways to use it. Likewise, when you have more CPU cores and power, you will find ways to use it.

I’m a believer of dual core CPUs be it Intel or AMD, but the point when the number of multi-threaded games being released will likely exceed the number of single-threaded games being released won’t happen until Summer 2006. So, for the fastest benchmarks in games today, our ultimate gaming system needs the fastest single core CPU.


SIDEBAR: AMD used to be the company that had to compete on price against Intel. How the tables have turned…

Wonderchild

Page:: ( 3 / 14 )

Choosing The Platform

In the choice between AMD and Intel, there’s little debate. If we’re looking for the fastest gaming performance, it’s going to be with NVIDIA SLI. With Intel SLI products not yet widely available, we’re going with AMD. (This article is being written in May 2005). However, even if Intel SLI boards were widely available, the Athlon FX-57 CPU would still our choice for the fastest gaming system. The FX-55 is already faster in real-world gaming than the P4 EE 3.73GHz CPU (and does so at almost a $250 savings) and the Athlon FX-57 should be even faster. Our FX-57 was a pre-production model.

AMD Athlon FX-57
$950
http://www.amd.com

What makes Opteron 100’s different from Athlon64’s?

While the original Athlon FX-51 was a Socket 940 part and essentially identical to an Opteron 146, today’s Socket 939 Athlon64 FX’s are different. The Athlon 64 FX-57 differs from the traditional Athlon64 line-up in that it features 1MB of L2 cache rather than 512KB found in some AMD designs, and because it represents the fastest core clock speed. Both Opterons and Athlon64s have the same fundamental AMD64 core design.

The Opteron platform is designed to handle mission-critical tasks. For any given clockspeed, there are three versions of the Opteron: the 1xx, 2xx, and 8xx series. That is, the Opteron 152, Opteron 252, and Opteron 852 all run at the same 2.6GHz. In a single CPU system, they will all perform identically. The Opteron 152 only works in a single-socket system. The 252 can work in dual-socket systems and the 800 series will work in an 8-way system. Athlon64 CPUs only function in single-socket systems.


While both Athlons and Opterons have on-die dual-channel DDR controllers, the Socket 940 Opteron architecture uses Registered ECC DDR-RAM. Registered RAM is slower, but more reliable. The Athlon 64 FX supports ECC memory but not registered RAM. I’ll talk more about Registered ECC in tomorrow’s article.

Running Total: $950

SIDEBAR: If they had kept the Athlon name for the servers, we would have been able to talk about a “biathlon” when referring to a 2 CPU system or a “decathlon” for a 10 CPU system. Oh well.

CPU Cooling

Page:: ( 4 / 14 )

Heatsink/Fan

The days of not knowing if your cooler would be sufficient for your CPU is gone. Manufacturers now make it easy to identify which heatsink works for which CPU, and with standardized thermal requirements, today’s heatsinks consistently overshoot the minimum requirements. For non-overclocked systems, the differences in CPU coolers is more about ease of installation, operating noise, reliability, and looks than it is performance. Everything changes when you need to overclock.


In the nearly two years after our last system building article, our design choice for the CPU coolers remains largely unchanged: the hybrid aluminum-copper Zalman CNPS7000B-AlCu. This design from Zalman is one of the easiest heatsinks to work with and clean and the B-spec version offers an incremental improvement over the older A version. This is due to a minimal change in the physical layout and improvement in the smoothness of the CPU-side surface. Although Zalman offers a larger 120mm cooling fan (7700 series) based cooler, the smaller unit works almost just as well without needing nearly as much space. The difference in thermal resistance is “0.22 to 0.29” for the CNPS7000B-AlCu and “0.21 to 0.28” for the bigger CNPS7700-AlCu.

By using a low-RPM, large-diameter fan with a high-surface area heatsink, noise is kept to a minimum, the hallmark of Zalman cooling designs. Our rationale for the aluminum-copper hybrid over the pure copper model is the same as it was last time – the performance benefits of pure copper are minimal, but the added weight adds unnecessary torque. The CNPS7000B-AlCu is actually within the official AMD and Intel spec.

[image]

Arctic Silver 5

Our choice of thermal grease also remains unchanged from the last article: Arctic Silver 5. Arctic Silver 5 is one of the easiest thermal greases to apply with unparalleled performance. Although there used to be concern about the silver content of the grease, today’s CPUs with large heatspreaders make the concern about the possible conductivity and capacitance a non-issue.

Arctic Silver 5 – 3.5 g (enough for ~15 large CPU cores)
$10
http://www.arcticsilver.com/

Running Total: $1000



SIDEBAR: I ordered new cookware, a medical textbook, and Zalman coolers all at the same time from Amazon.com.

The Motherboard

Page:: ( 5 / 14 )

When it comes to motherboard chipsets, we can think of no other platform we’d prefer over NVIDIA nForce 4 and nForce Professional – even the famed Intel platform division that produced the legendary Triton 430TX, 440BX, and i875P loses in a matchup against “nForce4 SLI for Intel.” The Intel team still has some SATA RAID talent that NVIDIA does not, but I wouldn’t be surprised if NVIDIA’s team ends up recruiting Intel platform engineers too.

Which nForce4 SLI motherboard?

We’ve already done a round-up of most of the nForce4 SLI motherboards on the market. In that article, we gave the DFI LANPARTY nF4 SLI-DR our Editor’s Choice Award and the MSI K8N Neo4 Platinum our Bull’s Eye Award.

Both products clearly would be potential candidates for our ultimate desktop system. The DFI is faster on the benchmarks, however the MSI has more SATA-II ports with the second Silicon Image controller running on PCI-Express, a second GigE controller running on the PCI-e bus also, and an integrated SB Live 24-bit. In fact, the MSI would clearly have been a good candidate for our storage server and given the Tyan Tomcat K8E a run for its money (last year, we used a Tyan Trinity i875P for our storage server) and probably be my choice for a single processor Athlon64 SLI capable motherboard where I knew I’d need the SATA-II support.

That said, for gaming, the extra features of the MSI aren’t that useful and when it comes to raw performance the DFI is our easy top pick. It had the consistent performance advantage over the competition in our nForce4 round-up, and comprehensive overclocking features. Likewise, the LANPARTY nF4 SLI-DR also had the nicest set of “extras” such as the FrontX drive bay, high-quality round cables and UV sensitive cable wraps, as well as a harness for transporting your system.

Japanese Capacitors

The specification list for the DFI motherboard includes a bullet point that Japanese capacitors are used. This is a feature worth discussing – since it’s not designed to suggest that Japanese-made products are better than Taiwanese-made products as some people might assume.

The story behind this is industrial espionage and therefore many of the details are sketchy. The story begins with the recruitment of a Japanese scientist who had previously worked at one of the major Japanese capacitor firms. He had been recruited by a Taiwanese capacitor firm to replicate the electrolyte that had been available in Japan. This scientist was able to successfully replicate the electrolyte, but after the problem was solved, his research staff went rogue and they took the formula and began to sell their own electrolyte at bargain pricing to other Taiwanese companies. Unfortunately, they stole the wrong formula and the result was that several Taiwanese capacitor companies faced the daunting prospect that they had shipped millions of bad capacitors that could fail early.

Of course, now that the problem is known, most capacitor manufacturers are avoiding any shady electrolytes and many manufacturers in Taiwan will important the electrolyte from Japan to guarantee the reliability of their products for their customers. So, while it’s great to see that the capacitors are sourced from Japan, it doesn’t mean that non-Japanese capacitors are necessarily any worse.

Going rogue is usually more fun, but I guess in this case it wasn’t. ;)

DFI LanParty nF4 SLI-DR
$200
http://www.dfi.com.tw

Running Total: $1200


SIDEBAR: Panasonic/Technics used to use bamboo-pulp based separators in their audiophile capacitors.

Memory

Page:: ( 6 / 14 )

Since our desktop is supposed to be about lots of bling, an obvious choice was to go with the new XMS Xpert series with the giant programmable LED display. If you recall, Corsair was the original manufacturer to put the usage LEDs with XMS Pro. Conventional wisdom is that these gimmicks should compromise stability – especially the XMS Xpert which allows you to custom program the LED display through the regular memory bus interface… However, the point is that Corsair does not implement these features until they are sure that it does not compromise stability. We know this because we’ve been using XMS Xpert modules in a production system for several months and no matter how aggressive we are, the XMS Xperts keep on ticking without missing a beat. It is clear that Corsair’s design team understands that when considering the balance between performance, “bling,” cost, and reliability, that it’s really a question of performance, bling, and cost and that there’s never an option for anything other than 100% reliability.

The LED panel is a large 10-digit alphanumeric display which shows real-time voltage, temperature, and frequency and allows you to custom program any message you want. It’s cool enough to have these features, but it’s even more amazing when you realize that this feature works without compromising stability. The Corsair 1GB Twin-X XMS Xpert 3200XL modules feature matched 512MB DIMMS with 2-2-2-5 timing running DDR-400. Corsair’s XL modules with 2-2-2-5 do run hotter than standard 2-3-2-5 low latency modules. While extraordinary cooling is not needed, we would not be comfortable recommending those modules in a fanless system where the modules are only cooled passively. Any normal system with a basic system fan will work just fine.

But DFI has custom-made PC5000 RAM from OCZ Technology available. These modules are rated to as high as 625MHz by running at 2.85V (standard is 2.75V)! Going with these modules would increase our ability to overclock the RAM, however the latency would also be increased to 3-4-4-10. One catch is that although these memory sticks are qualified at 625MHz, it’s not as easy as plugging the modules in and running your applications. Running the Athlon64 memory controller at 312.5 MHz can stress the CPU itself.

[image]

Power Supplies

Page:: ( 7 / 14 )

The power supply used to be an incredibly overlooked component of the system. At the turn of the century, most people did not understand why anyone would want to spend $50-60 on a power supply when a whole new ATX case with power supply could be bought for the same price. Back then, 250W was the plenty to go around. It really wasn’t until the Athlon and P4 that people started to notice the effects of poor power and started to care about the available current on different voltage rails.

Fast forward to 2005, and most people see 350W power supplies as the bare minimum, with most gamers opting to go to 400W and up. Making sure +12V lines have enough amps isn’t as much of an issue, and I’d say that most people are returning to days of shopping by watts, with only the handful of die-hard power gurus considering things such as cross-loading (the idea that what you have on your +3.3V and +5V affects your +12V rail).

Another interesting thing is that very few systems are drawing that much power over any extended period of time. That’s true even with today’s CPUs and SLI GPUs. You really won’t be able to run out of watts until you start adding large multi-drive RAID arrays. On the other hand, the need for better and cleaner power supplies cannot be underestimated. Without question, we are all buying more powerful PSUs than we need when it comes to wattage. What has really been the case though is that PSUs capable of handling higher current are also PSUs that end up being more stable and reliable. Again, we’re not really learning anything new – we’re just rediscovering why the old super-stable SGI workstations would engineer 747 W power supplies when their CPUs were drawing just 17W and the whole system might only draw somewhere between 200-300W from the wall.

Eating Healthy and Exercising

The issue about investing in a good power supply is that it’s not a benefit you can see immediately. It’s not like spending more on a faster CPU or a bigger HDD. With a PSU, you’re just buying insurance and reliability. It’s like eating healthy or exercising – you’ll probably feel great eating the KFC all the time, and putting yourself on the Burger King Angus Diet up until the point you have your heart attack.

I can get an Athlon FX-57 with dual GeForce 6800 Ultra SLI’s and a pair of Corsair 2-2-2-5 RAM to successfully run stressful NVIDIA tech demos using a no-name generic 350W power supply rated with only 10A on +12V. The 12V rail will drop to precipitous +11.4V under load, but it works. However, just as you may begin to doubt if the PSU quality issue is overblown, the system will spontaneously reboot or otherwise crash.

You can tell when you’re grossly underpowered, but knowing if your PSU is perfect can be more subtle.

So which power supply?

There are few unanimous comments in computing. You won’t be able to get everyone to agree on which CPU to get or even which memory manufacturer is the best. When it comes to power supplies, however, the best of the best is unanimous – PC Power and Cooling. We’ve been using PC Power & Cooling even before high-end PSUs were the cool thing to do. Indeed, two years ago ,we used a Turbo-Cool 475 XE in our Dual Athlon MP and a Turbo-Cool 510 ATX in our Pentium 4 Storage Server. The PC Power and Cooling PSUs are really that good, and it’s amazing that the design continues to be the reference PSU that has yet to be matched. For the desktop our pick is the PC Power and Cooling 510-SLI and for our workstation we’re building tomorrow, we’ll with a PC Power and Cooling 850W SSI. It’s hard enough to stress a PC Power and Cooling 510W PSU, so I’m not even sure there’s a good reason to get an 850W power supply. [cue mysterious foreshadowing music]

[image]

Chassis

Page:: ( 8 / 14 )

A good chassis layout and design will do more for system cooling than throwing a bunch of noisy fans into the mix. While looking at the various system cases on the market, we knew that our ultimate chassis would be found in the BTX-style Silverstone Temjin-06, the Lian-Li V-1000/1200/2000, and Enermax MaxFlow CS-718. All three designs use a BTX-style motherboard mounting system where the motherboard is mounted on left side rather than the right (when looking at the case straight on). Inverting the motherboard allows better separation between the motherboard and the HDDs and PSUs, and places the CPU at the bottom of the chassis rather than the top.

[image]

Lian Li or Silverstone/Enermax?

The differences between these cases are in the layout of the power supply and hard drives. The Lian Li V-series places the hard drives and power supply at the bottom of the case. In doing so, the airflow across the hard drives is optimized because the power supply exhaust fan will also be moving air across the hard drives. The disadvantage is that the CPU is in close proximity to all of the major heat producing elements of the system (the HDDs and PSU). There is a funnel of heat below the CPU with a warm aluminum shield. Likewise, although the HDDs will have superb cooling, the PSU is under greater stress from all of the heat from those HDDs.

Tread Lightly…

With the Silverstone and Enermax design, the motherboard is inverted but the power supply remains at the top of the chassis. This approach is designed to place the CPU as far away from the PSU/HDDs as possible. Dedicated twin 120mm fans provide a continuous stream of cool air. The hard drives are mounted at the rear, at the very top of the chassis. The chassis engineers envisioned gamers using PSUs with twin fans such that the PSU’s second intake fan would also move air across the HDDs. The HDDs face the rising heat of the GPU, and so the stress on the PSU is also significant

If you haven’t realized it yet, one of the key design flaws of the Silverstone Temjin-06 and Enermax MaxFlow CS-718 is exposed when the case is used with a power supply that only has a single exhaust fan. While single fan PSUs may represent bottom-of-the-line units, it also reflects flagship PSUs where the unit is filled to the max with circuitry and components. None of the PC Power & Cooling units, nor the Silverstone Zeus ST65F, nor the OCZ PowerStream feature the bottom fan.


So, with a flagship PSU, the result could be disastrous with the HDDs enclosed in what is essentially a pizza oven with radiated heat coming from the PSU above, the GPUs below, and no horizontal airflow! Silverstone feels that the HDD temperatures will still be acceptable even without direct cooling, but does say that PSUs with a bottom fan will be required if you would like to keep your HDDs cooler than 35 C.

So…

To summarize, both the Lian-Li design and Silverstone/Enermax design work to improve cooling over the conventional ATX tower design. The Lian Li V-series does a better job keeping HDDs cool and is overall the easier system to work with, but the Silverstone/Enermax will be better at keeping the CPUs cool. It’s just as effective at cooling HDDs if you’re using a power supply with a second intake fan, however with a flagship PSU, you may need to make modifications. Other considerations include the fact that the Silverstone case offers a reset button whereas the Lian Li does not. The Lian Li also requires the use of specialized screws and tools (which are included), which can be a problem if you lose the accessories and the V1000 and V1200 can have trouble with large power supplies.

Choose wisely. We initially planned to build one of our systems using the Temjin SST-TJ06B and the other using the Lian Li PC-V1200, however our Temjin TJ06 disappeared during shipping. Fortunately, Newegg.com stepped in and was able to support this article with a Lian-Li V1200.

Lian Li PC-V1200
$210 at Newegg.com

Silverstone Temjin SST-TJ06B
$150
http://www.silverstonetek.com

Running Total: $1940




SIDEBAR: If it seems like I keep bringing up SGI’s design in the Octane/Octane 2 workstations in all my system building articles, it’s only because it’s truly that special to me. My Octane After all, the Octane was built at a time when people would spend $40k on a workstation. Those truly reflected no-compromise solutions. Today’s SGI’s aren’t built quite to the same level as the older ones, although the Tezro or Prism both have well-thought out cooling approaches superior to today’s PCs or Macs.

Hard Drives

Page:: ( 9 / 14 )

The fastest performing single desktop hard drives will be the WD Raptor 10,000 rpm drives despite the lack of S-ATA features such as NCQ. If we were looking for the absolute fastest performance and could live with the reduced drive capacity, they would be good choices. On the other hand, I never seem to have enough hard drive space.

The true ultimate setup would therefore be a pair of Raptors in RAID-0 as the main program drive, a third Raptor dedicated for swap file and temporary files (a 36GB would be fine, but the 74GB is quieter with FDB motors), and then several large capacity 7200 rpm SATA drives for data. Placing the swap/temp files on a dedicated drive will speed up virtual memory accesses, and with a Raptor, it would be that much faster. The striped RAID-0 would provide superior read and write performance, and the remaining 7200rpm drives could be the place for video, audio, and all other large media. This isn’t a luxury we can all afford.

One of the key secrets to the performance of the Western Digital Raptor is its low seek times. However, with NCQ optimization the differences in seek time under real-world conditions narrows. Where the Raptor still holds the crown is with throughput. This is where RAID comes into play.

In a RAID-0 Striping setup, data is read and written to alternative drives and as a result you can double the throughput both when writing and reading. In a RAID-1 Mirroring setup, data is written simultaneously to two drives. While this can actually slow down write speed, the added level of redundancy protects data against a physical crash of a hard drive. Most RAID-1 Mirroring devices do not accelerate the read, but more advanced RAID controllers will be able to read-stripe by requesting alternating blocks from each drive.

What are the best 7200 RPM desktop drives for RAID-0?

On the desktop, our favorite 7200 RPM drive for RAID-0 is the Hitachi Deskstar T7K250. These 250GB 7200 rpm SATA-II based drives have a peak bandwidth of 300MB/sec and NCQ support. Most importantly, Hitachi has built these drives using 2 high-density platters each having 125GB. This increased density should improve reliability, but it will also increase throughput and reduce the seek times. Seagate also produces high-density platter drives (the highest in the industry with 133GB/sec). While this also helps Seagate improve reliability, the design philosophy between Seagate and Hitachi is substantially different. As a general stereotype, when comparing the two, Hitachi drives are engineered for lower-latency whereas Seagate drives have been engineered for quieter operation. That’s just the stereotype, and everyone will be able to find exceptions, however that’s just a design philosophy. No manufacturer will ever claim that their drive is anything but reliable and there is not a very good way to evaluate HDD reliability at this time.

[image]

SATA (cont’d)

Page:: ( 10 / 14 )

Why Serial ATA is more than just about transfer rates

When you ask most people what makes Serial ATA so interesting, the answer is usually going to involve the super-thin cabling and the frustration of new power plugs. Some people may add that SATA 1.0 can support 150MB/sec transfer instead of a peak 133MB/sec of IDE, but we are all wise to the fact that most hard drives won’t saturate an ATA 133 bus. The cabling issue seems trivial, but it’s just as important as some of the additional underlying features. Having a single cable for each drive produces a point-to-point topology in which each drive can provide maximum throughput to the controller. With IDE, or even traditional SCSI, each additional hard drive can reduce the total throughput available to the system. Likewise, if you look at the SATA cables very carefully, you’ll see that they’re staggered. This is done intentionally so that when you hot-swap a drive, power is sent to the drive in a controlled fashion to limit the inrush of current to a drive. The drive will never get power before the pre-charge and grounding have occurred.


Serial ATA moves some of the SCSI-like technologies down into the affordable desktop region. There are 2 elements worth highlighting. The first is CRC checking. With traditional IDE drives, the data is checked for accuracy but that’s it. The commands themselves that are being sent are not checked for accuracy and you can imagine that if the directions are off, it’s garbage-in-garbage-out. Serial ATA adds the error checking for the commands being issued -- the same ability that was found in Ultra160 SCSI and newer.

The second new feature is something that’s become very popular in tech circles: Native Command Queuing. NCQ allows the drive to reorder data requests in an optimal manner for data transfer. Imagine that you have two pieces of data on the inner track of the HDD called A and B and data on the outside track called C. If a program requests file A, then file C, and then file B and you’re running a standard IDE drive, the HDD will read the first file, move the magnetic head across the entire platter to read the second file , and then back to read the third file. With NCQ, even though the computer has requested A then C then B, the drive is smart enough to know that it should read A and B at the same time and then move to C, speeding things up.

That’s cool.


SIDEBAR: 80-pin SCA had the interesting design of integrating power and data on a single connector. Western Digital has a cable that has a single physical connector that plugs into SATA power and SATA data.

Scalable Link Interface

Page:: ( 11 / 14 )

Normally, I’d take the time to ponder the ATI versus NVIDIA issue. After all, the X850 line has proven to be a top-notch GPU that most of us would love to own. Still, NVIDIA stole the thunder from ATI. There’s no debate since NVIDIA beat ATI to dual modern dual PCI Express GPU support and those of us looking for top-of-the-line performance can get a pair 6800 Ultra’s today, while it’ll still be at least a month before ATI will have their new Crossfire GPUs available.

Is SLI a waste of money? What about the GeForce 7800 GTX?

An obvious question is why we aren’t going with a GeForce 7800 GTX in our article. Actually, the easy answer is that this article is being written in May (yup, we had our FX-57 then), a time when the 7800 GTX was not announced yet. This will also help us delineate the issues that arise when buying a “flagship GPU.” Does it make sense to buy the flagship GPU or wait for the “next generation” GPU to come out? This seems like a tough question, but the real answer is straightforward – go with your budget, and buy as early as you can.

Most would argue that a pair of 6800 Ultra GPUs does not reflect the best-bang-for-the-buck either, due to the simple fact that the 7800 GTX should be out soon (and by the time this is published, fully available). I wouldn’t disagree with that. However, an important fact is that flagship GPUs can make an immediate difference in today’s games. While it’s true that you’re paying a price premium, you are also getting the full benefit of the performance from the get-go. This is different from the 512MB 6800 Ultra’s which were announced in March in which there was little benefit.


Another important point is that when the 6800 Ultra’s came out, they were $500 each. Up until the very moment that the 7800 GTX was released, they continued to be around $500. What that means is that the GPUs were “relatively” cheaper at the time of launch than they are now because when they first came out, it was “that much faster” than everything else. In other words, if I bought the 6800 Ultra when they first came out, I would get the same performance I get from buying a 6800 Ultra now, but one option gives me a full year of experiencing high-performance gaming.

Not everyone can afford SLI 6800 Ultra GPUs, but there’s no question that it’ll provide a better experience in today’s games. With the 7800GTX, the top of the line has changed. They’ll be priced at $600 each. If you had $600 to spend on a video card, or better yet, $1200 to spend on a pair of 7800 GTX’s, the right approach is buy the 7800 GTX as early as possible. Think of it like a car – by the time they start dropping in price, it’s not really a bargain – it is depreciation. I would only buy flagship GPUs when they first come out – this doesn’t maximize your performance/price ratio, but it maximizes your “hours of flagship performance/price” ratio. If you can’t afford the initial GPU pricing, just stick with what you can afford – trust me, it works out.

So which 6800 Ultras?

For our GPUs, we ended up selecting a pair of BFG GeForce 6800 Ultra OC PCI-E 256MB. These cards feature a 1.1GHz memory clock and a 425MHz core clockspeed a small bump over the standard 400MHz. We like BFG graphics cards a lot as they have quickly grown from being the “start-up” board manufacturer (they came to the market in the GeForce 4 era) to probably one of the most consistently performing manufacturers to date. In a way, they are like the new Canopus of graphics cards, only with more reasonable pricing. We would anticipate BFG’s GeForce 7800 GTX’s to be just as good.

[image]

The Optical Drive

Page:: ( 12 / 14 )

Whereas PC Power & Cooling are the unanimous choice for flagship power supplies, the closest thing to flagship optical drives are those from Plextor. I still remember my 4Plex SCSI CD-ROM with a massive 1MB buffer which came in a box so big you’d think there was a laser printer inside. Indeed, my 12x Plexwriter and UltraPlex 40x Wide SCSI CD-ROM still hold a special place in my heart. Even before Plextor was Plextor and was called Texel, they were always considered one of the best optical drive manufacturers in the world. (I actually skipped Texel drives in my upgrade path. My first single-speed CD-ROM was a Sony drive and then I quickly upgraded to the venerable 2.2x Toshiba XM-3401B. I was an early adopter for CD-ROM technology, paying what probably is the equivalent of buying blu-ray stuff today).

Plextor’s current flagship models are the PX-716SA and the PX-716AL. The SA is a Serial ATA model with 16x DVD+R write performance on 8x certified discs, and 6X write performance on dual layer discs! The AL variant, though being an IDE drive, is a slot-loading drive. We haven’t seen slot-loading drives for sometime (since Pioneer) and this is something we’re looking forward to seeing again. Plextor’s new approach is notable because they claim to be able to use 3-inch CDs without any trouble or extra adapters. It’s unclear if it’ll work with “business card CDs.” For media, I use Taiyo Yuden CD-Rs and DVD-Rs exclusively. The M.A.M. Gold media is better, but the Taiyo Yuden’s are good enough for most.

Plextor PX-716SA
$125
http://www.plextor.com
Running Total: $3425

Keyboard and Mouse

Last year, we recommended the Goldtouch USB adjustable ergonomic keyboard at ~$120. While the Goldtouch remains the best keyboard we’ve used to date for typing, the poor placement of the ALT button makes it a little less ideal platform for a multi-tasking user. We would still recommend the Goldtouch USB to writers or anyone else who spends most of time in the QWERTY area, but we’ve since found a great alternative.


The fixed split keyboard of the Logitech Cordless Comfort Duo ($100) offers great tactile response and many of the benefits of the Goldtouch. Although the full cam-based action of the Goldtouch is better for larger-than-average or smaller-than-average hands, the Logitech design ends up working very well for me. The main disadvantage is that it’s only sold with a bundled Cordless Mouseman Optical. A now-discontinued bundle with the Cordless Mouseman Wheel was sold as the Cordless Freedom Pro.

Our mouse of choice is the MX700 rather than MX1000. While the MX1000 offers the best sensor of any of today’s mice, its size is a little bit too big for gaming. The MX700 and original MouseMan Wheel design are on the upper limit of a what can be considered a good gaming mouse. The MX510 and First Wheel Mouse design are a good ergonomic fit for gaming as well.

[image]

Floppy

Page:: ( 13 / 14 )

You should blame Microsoft for charging you a $15 tax on every new Windows XP system you build. In order to supply the Serial ATA or RAID drivers during Windows XP installation, you need to have a floppy drive. You can’t even use USB flash drives. If Microsoft were like Apple, they would have a least allowed people to swap out the Windows XP installation CD for a RAID driver installation. (More enterprising system builders can slipstream the drivers into a custom install CD to avoid needing to buy a 1.44 floppy.)

Microsoft’s poor design choice aside, the floppy drive remains a critical component of your system. Many manufacturers still believe in DOS-based BIOS flashing utilities and refuse to support Windows based flash BIOS updates on the account of the greater potential for irreversible damage. Newer motherboards are able to update the BIOS via a USB flash drive.

If you really cared about the a quality 1.44 floppy drive, you will need to go with a “sealed” design with a die-cast frame such as the 1.44 floppy from Teac (the original gold-standard in floppy drives) or Sony in which moving parts aren’t exposed to the user’s fingers (as is the case with Samsung floppy drives). USB double-speed read floppy drives are available, but I’m not sure who would actually benefit from that as opposed to a USB flash drive.

1.44 Floppy Drive
$15
Running Total: $3590

Monitor

In our 17” LCD Round-Up, I mentioned that “all things equal,” smaller LCD monitors would offer better gaming performance, even for the same advertised pixel refresh. For that reason we’ve opted to go with a 17” LCD monitor rather than a 19” one. Our monitor of choice for the ultimate gaming system is therefore the NEC MultiSync LCD1770NX. It has the excellent color that will let this system show the full efforts of video game artists, and the brisk pixel refresh that keeps smearing at a level that doesn’t interfere significantly with your gaming. Since our review, the LCD1770NX has actually dropped in price, making it an even better deal. Last time we awarded it 1st place, but we’re making this a Bull’s Eye product.

NEC Mitsubishi LCD1770NX
$320
http://www.necmitsubishi.com
Running Total: $3910

Operating System
Windows XP 32-bit SP2.
$150
http://www.microsoft.com

Final Cost Desktop: $4060





SIDEBAR: Well, I guess besides PC Power and Cooling, almost everyone can agree that Teac 1.44 floppy drives are the best floppy drives. The difference is that no one cares – I spent the extra cash on Teac floppy drives in the 286 era. Now I just buy the cheapest drive, willing to pay only 50 cents extra for the die-cast frame.

Optional Accessories

Page:: ( 14 / 14 )

Sound Card $35 to $280

Sound cards have really started to drop off in technology. The integrated soundcard on the DFI motherboard with Karajan isn’t bad and is perfect when dealing with games and movies. We can improve the quality and performance with an Audigy2 or an Envy24HT. My preference is to go with the Envy24 as it offers better music performance, although the Audigy 2 should offer better CPU performance. A lot of people will be perfectly happy with the Karajan integrated audio though which his why I’ve left it as an optional upgrade.

Speakers $150 to $$$$

Two years ago, we called the Swans T200A and M200A speakers are our PC speakers of choice. The T200A’s remain very difficult to find, but the M200A’s are available at Newegg. These still remain our recommended speakers when it comes to music. For movies and games, the Klipsch ProMedia 4.1 or 5.1 Ultra is our recommendation. Logitech’s Z-5300e offers the best bang-for-the-buck in the lower-priced range. There haven’t been any new speakers that really catch our attention.

In my own personal system, however, I’ve moved onto running real hi-fi speakers with a dedicated amplifier. This can improve your music quality immensely and 2-channel gaming still works well. I use the Polk LSi 7 on Standesign stands. These bookshelf speakers have remarkable bass, and ring-radiator tweeter is the same design you’ll find in far more expensive systems. The cost? $750 for the pair. That’s not including the receiver or amp to power them. These are 88 db/W 4 ohm speakers and so they need quite a bit of power or else they’ll sound terrible. That is, you need equipment that will typically be used with more expensive speakers. I’m running an old AVR-3600dts (the precursor to the AVR-4800; it’s pre-amp and amplifier section is superior to even today’s AVR-3805) and its quality could still be better. If I won the lottery, I’d get a Sonus Faber Stradivari and a pair of Mark Levinson No. 33H amplifiers.

Ageia PPU?

We actually don’t have a physics processor in this ultimate system build, but if it were available, we’d definitely put one in. You would be right to point out that most games don’t support the Ageia at this time, but physics processing really isn’t the big deal. You see, realistic physics are not entirely necessary in today’s games. Yes, in the future, game developers might start incorporating things like modeling non-playable characters with Denavit-Hartenberg conventions and then using real-time optimization-based kinematics of virtual characters, but today’s “rag doll” physics and hand-animated characters actually work just fine.

The area where the PPU will be better is particle effects. Imagine if that rocket launcher caused chunks of rocks to fly realistic and roll on the ground instead of fading through the ground? The PPU is not so much about improving physical realism as much as it is a way to improve graphics. Think about the tech demos that come out with every new GPU and how good they actually look. A lot of these demos get away with seemingly unbelievable graphics performance, because the CPU doesn’t need to deal with all the traditional overhead and game physics. If dual core CPUs or dedicated PPUs can help game developers achieve “tech demo” like graphics in a real-world environment, I’d be up for that. It’d be like a booster for your GPU.

Conclusion

So there you have it, a starting price of a little over $4000 for our “ultimate gaming desktop” with the sound card and speakers being optional. By going with a “no-budget but don’t waste money approach,” our $4000 price puts us in the same price range as Alienware and VoodooPC, only our system has better components such as OCZ’s DFI Special RAM and the DFI LanParty nForce4 SLI-DR. We’ve also listed MSRP for several items in our build, so you should be able to find lower prices on the components we’ve listed.

[image]