COMBATSIM.COM: The Ultimate Combat Simulation and Strategy Gamers' Resource.
 
Beating the Performance Curve Part 2

by Shui-Che Lim

Dollars and Sense

I've set up a simple spreadsheet based on some possible configurations and costs of the various upgrade paths that you might consider. The component costs were recently taken off of PriceWatch and represent the best cost if you’re willing to go mail order. For people who still prefer to deal face to face with their vendors, retail prices will probably be about 25% higher.

Component Prices

Motherboards

  • Socket 7 w/AGP..115
  • Pentium II..145

Processors

  • AMD-K6/233...149
  • Cyrix 6x86MX PR233...128
  • Pentium 233 MMX...189
  • Pentium II 233 MMX...260
  • Pentium II 266 MMX...367
  • Pentium II 300 MMX...515
  • CLabs 3d Blaster II.......229
  • Diamond Monster 3D II...249
  • MegaMonster...449

Certain assumptions were made regarding each of the cases. It is assumed that most people are using systems based on Socket-7 architecture and using an Intel Pentium 200MMX or equivalent. This is the mainstream system that software developers are now targeting.

Further, all the price/performance graphs were based on the performance for all systems at 640x480 resolution. Even at this level, a convincing argument can be made about not going to a Pentium II-300 with a Voodoo2 now. If 800x600 mode were used, the price/performance results would have skewed even more in favor of a socket 7 solution.

Finally the price of the MegaMonster is an estimate only. It is assumed that the software bundle that Diamond has assembled will add about $50 to the cost of the package. Therefore, the MegaMonster should only cost about $200 more. Prices for the Pentium II-233 and 266 are for reference only, they were not used since no performance numbers are available for them.

Case #1: Voodoo2 Only vs. Pentium II-300 w/Voodoo2

Case #1: 640x480 resolution performance figures 640x480 resolution performance figures used

The base configuration presumes that you will only add a single Monster 3D II to your system. The relative difference in price and performance to a dual Voodoo2 system and a Pentium II-300 system are compared.

The most astonishing difference in relative performance is seen in the system with the Pentium II-300. Here you can see that the cost differential is 3.7 times the cost of an upgrade to a Diamond Monster 3D II only. However, you only achieve 1.56 times the additional performance under GLQuake. Dividing cost vs. performance gives a relative value index of 2.3. Smaller numbers represent better value (price and performance scaling more linearly).

I need to point out that these numbers are valid for GLQuake only. The reason the MegaMonster with dual Voodoo2 cards scale so badly is that Quake is triangle limited as opposed to fill-rate limited. This means Quake cannot pass the geometry information to the Voodoo2’s triangle setup unit faster than the triangle setup unit can render the triangles in the framebuffer. That’s why adding a second Voodoo2 for Quake doesn’t give that much better performance.

It is important to keep in mind that you cannot directly ascribe performance under Quake to general performance under combat simulations. Quake is a first person action shooter. The creatures, creature animations and the environment modeling is so demanding as to create a bottleneck in geometry transformation and triangle setup. While it is true that modern combat simulations are demanding, it is also true that in most cases, you won’t get mixed up in huge furballs with up to a dozen aircraft unless you’re playing sims that model eras prior to the 1960’s. Most modern air combat is BVR and you hardly ever see all opposing forces at the merge.

The trend in modern simulations is to create a highly detailed player vehicle with the same level of detail for all like vehicles in your wing/squadron/group. However, detail levels for the remaining computer controlled vehincles are toned down, i.e. they don’t have as many polygons, moving parts, or texture maps. The environments are usually out in the open and have simple structures unlike the detailed structures seen in first person shooters.

Typical simulations nowadays can have upwards of 8K-10K polygons per screen. The only other thing which can add appreciably to the poly count is the use of virtual panning cockpits. However, Quake/Quake II could easily have twice that number.

3Dfx claims that two Voodoo2’s in SLI mode could double the triangle and rendering performance, assuming that the application in question is more fill rate limited than geometry or triangle limited. I believe that this is the case with modern combat simulations. That said, Voodoo2’s could potentially double your performance and it could also potentially give better performance than a system with a Pentium II-300 with a single Voodoo2 card.

Case #2: Upgrade your Socket 7 CPU and add Voodoo2


Case #2:  640x480 resolution performance figures 640x480 resolution performance figures used

Even if you add the cost of a new Socket-7 CPU whether it’s based on AMD, Cyrix or even Intel architecture, you will see that at most the Pentium II-300 system with a single Voodoo2 card will score 3.17 times the performance vs. the Cyrix 6x86MX.

Based on the price performance index, you get the best value with the AMD K6-233 since the relative price/performance vs. the Pentium II-300 tracks very closely. This means that for the performance that you could get, the AMD-K6 tracks very closely to what you would expect to pay for the CPU. If you lean more towards performance, the Pentium 233MMX looks good since the Pentium II-300 system still costs over 2 times more but only offers an additional 1.56 times performance.

The most interesting point to note here is that the performance of a Pentium II-300 system equipped with a single Voodoo2 card averages just over twice the performance over the three socket 7 CPUs. With a high performance Socket-7 CPU and a dual Voodoo 2 setup, you could potentially get nearly equivalent performance of a Pentium II-300 system with a single Voodoo2 for simulations.

Even with the added expense of the MegaMonster package, you still come out ahead in most cases by at least $271 if you stick with your current motherboard and upgrade to a Pentium 233MMX. Even if you upgrade to the newer 100MHz host bus AGP Super 7 boards, you’ll still come out ahead $156. You save even more if you use CPUs from AMD or Cyrix. While you won’t always equal the performance of the Pentium II-300 system, dual Voodoo2’s give you a potential to get really close in many cases while saving money at the same time.

A Final Note on Performance

The GLQuake numbers for a Pentium II-300 with a single Voodoo2 at 74.4FPS while impressive and interesting is just plain overkill. The real key to having an enjoyable combat sim experience lies in keeping rock steady 35-40 frames per second. The human eye really cannot perceive anything beyond that for simulation use.

Given the dominance of 3Dfx in the marketplace, you can also reasonably expect that combat sims will come with native 3Dfx Glide support out of the box. This will keep all the naysayers at bay regarding the large performance difference of Direct3D on Socket 7 vs. Pentium II-300. Native Glide support means that the difference in performance between an up to date Socket-7 machine with even a single Voodoo2 vs. a Pentium II-300 will be kept to probably less than 30%.

Simulations now coming down the pipeline are not even going to challenge the performance of a single Voodoo2 card let alone an SLI enabled dual Voodoo2 system. We are at least 6-8 months away from the software industry fully utilizing the power of a single Voodoo2.

When that day does come, you can inexpensively increase performance by upgrading your CPU from cost effective mainstream models instead of the pricey high-performance ones. Since Voodoo2’s performance scales with the performance of the CPU, you can upgrade to faster and faster CPUs after they’ve fallen in price. Waiting an additional 6 months or so to go to a Pentium II-300 and a new PII motherboard could reduce your total cost by 50% or more. This is what I consider a smart upgrade path

Go to Tech Central


Main


© 1997 - 2000 COMBATSIM.COM, Inc. All Rights Reserved.

© 2014 COMBATSIM.COM - All Rights Reserved