Although we almost always tend to focus on the high-end, the GeForce4 Ti 4600s, of the market it's quite the opposite from the perspective of ATI and NVIDIA. While it's always nice to have a performance leading part from a PR and brand-recognition perspective, those high-performance GPUs are often only found on cards priced between $200 - $400. If you were to look at a breakdown of the add-in graphics card market you'd find that over 80% of the market won't pay a penny over $199 for their graphics card, making cards like the GeForce4 Ti 4600 very low volume products.

Then why produce them? Since they are low volume products it is easy to outfit them with extremely fast memory (325MHz DDR), a very large PCB (almost the size of a Voodoo5 5500) and a complex die also running at a high clock speed (300MHz). The culmination of all of these efforts satisfies the very small percentage of the market that is composed of people like us. In turn, we brag about our incredibly fast video cards and help to educate the rest of the masses to buy one brand over another. When everyone else goes out to purchase a video card they don't spend the $320 on a Ti 4600, they buy whatever NVIDIA card they can for $199 or less. Well if you can't get a GeForce4 Ti 4600, then a GeForce4 MX will do just fine, right?

There isn't a single knowledgeable member of our community that hasn't expressed extreme distaste with NVIDIA's GeForce4 MX line. Little more than a glorified GeForce2 MX with a GeForce4-derived memory controller, the GeForce4 MX has been called everything from NVIDIA's first major screw-up in years to 99% marketing and 1% product. From NVIDIA's perspective, the GeForce4 MX is perfect. The entire line of MX cards falls below the $199 mark making it a product for the masses, the GPU is noticeably less complex than the GeForce4 and thus boasts a smaller die size making it cheaper to manufacture, lower memory clocks give NVIDIA the ability to offer ample supply to the market and it carries the GeForce4 brand which is associated with the very highest performance graphics cards around. In terms of performance, we've proven that the cards run the majority of today's games very well so what's all the fuss about? In the end, the biggest compliant from our community and developers alike is the use of the GeForce4 brand name on a part that has the same amount of DirectX 8 features as a GeForce2 thus deliberately misleading the consumer. In fact, had it not been for ATI the sub $200 add-in card market would probably be dominated by non-DX8 compliant cards.

Just days before NVIDIA's GeForce4 and GeForce4 MX launch, ATI introduced the Radeon 8500LE 128MB which would be priced below $200 and offer full DX8 compliance. NVIDIA of course knew very well of ATI's plans (little is kept secret in this industry) and was forced to produce a mass-market GeForce4, the Ti 4200. Neither the Radeon 8500LE 128MB nor the GeForce4 Ti 4200 were available when both companies were talking about them but it's the recent if not soon-to-be availability of both of those cards that makes this article ready for introduction.

If the vast majority of the market spends less than $200 on a video card then we had better round up as many sub $200 cards as possible. The only criterion for this roundup was that the card would carry a street price of $199 or less. Before we get to the benchmarks themselves lets have a quick look at the two new contenders that we haven't talked about in much detail before.

ATI's Radeon 8500LE: Let the people have DX8 compatibility

Click to Enlarge

Click to Enlarge

Unlike the Radeon LE that preceded it, the Radeon 8500LE is an officially sanctioned, retail part from ATI that only differs from its bigger brother in its clock speed (250/250 vs 275/275). The 9% drop in clock speed doesn't hurt performance too much and allows ATI to target the critical sub $200 market with the card. In fact, during ATI's initial press briefing of the Radeon 8500LE it was placed head to head with the few specs that were known about NVIDIA's GeForce4 MX line. The goal was to offer the only DX8 solution in the mainstream market and beat NVIDIA where it counted the most. The Radeon 8500LE would also become the only board in this sector to offer 128MB of DDR SDRAM on-board using BGA packaged memory.

The Radeon 8500LE is currently available and is priced at $199.

NVIDIA's GeForce4 Ti 4200: ATI, look what you made them do

NVIDIA would have been quite happy supplying the mainstream market with GeForce4 MX cards but because of the Radeon 8500LE it became necessary for NVIDIA to also have a DX8 card in this sector; and thus the GeForce4 Ti 4200 was born.

Click to Enlarge

Click to Enlarge

The GeForce4 GPU is considerably more expensive to manufacture than the GeForce4 MX so costs had to be cut as much as possible to make the Ti 4200 a profitable offering. The GPU is clocked at 250MHz, making it the highest yielding NV25 GPU coming out of TSMC. As far as memory goes the card can be offered with 222MHz DDR (effectively 444MHz) or 250MHz DDR (effectively 500MHz) SDRAM. The difference in memory clock speeds arises from the fact that the cards will be offered with TSOP packaged memory and not the new BGA packaged memory we've seen on the higher end GeForce4s. In order to offset the costs of the 128MB cards the slower memory is used while the 64MB cards get the benefit of faster memory.

Sticking to TSOP memory allowed NVIDIA to keep to an older board design and you'll notice by the above pictures that the power circuitry is significantly reduced. The Ti 4200 PCBs are also 6-layer boards instead of the much more expensive 8-layer 4400/4600 boards.

The end result is a GeForce4 that is cheap enough to produce that it can be sold for $179/$199 (64MB/128MB). Availability is very scarce at this point, most board manufacturers are waiting for chips from NVIDIA. Most cards will begin to appear in May. One thing that should be made very clear is that the overclocking characteristics of the reference boards being sent out for review now are in no way representative of how final Ti 4200 boards will overclock. NVIDIA's reference boards have historically overclocked much better than shipping cards which also explains why NVIDIA usually encourages reviewers to try overclocking in their reviews.

The Test

You'll notice that this roundup is done a bit differently from graphics roundups in the past. We benchmarked each card at three resolutions, 800x600, 1024x768 and 1280x1024. For those tests we used the fastest available Athlon XP CPU on an KT266A platform (2100+ on an EPoX 8KHA+). Obviously not everyone looking for a sub $200 video card will be running the absolute fastest Athlon XP processor so we benchmarked every card under every game at 1024x768 at 15 different clock speeds.

We kept the FSB speed and CPU architecture constant so we could accurately measure how the various cards scaled with CPU clock speed; while this does produce some unofficial CPU speeds (e.g. 800MHz/133MHz FSB Athlon XP), you can get a good idea of where your CPU will limit your performance with any of these cards.

At the end of each game benchmark section you'll find a CPU Scaling graph that depicts exactly what we've mentioned above. We spent countless days running the CPU scaling tests so we hope they come in handy for you.

Unreal Performance Test 2002 - Build 918

We originally debuted this test months ago but it has been healthily evolving since then. The premise behind the Unreal Performance Test is to take the current build of Epic's Unreal Engine and test today's graphics cards using it. This not only better prepares you for what to expect when games featuring the engine come out (e.g. Unreal Tournament 2003) but it also helps the guys at Epic during the development stage as they can work out driver bugs with the hardware vendors long before the engine's public debut.

The engine we're testing with now is up to build 918, up from build 848 we originally tested with back in January. The benchmark today is much more stressful than even what we used in our GeForce4 review and according to Epic, it's even more stressful than most game scenarios will be but it makes for a wonderful benchmark. A number of optimizations for both ATI and NVIDIA's architectures have also been implemented into the engine. The only thing to keep in mind is to not focus on the actual frame rates themselves but the performance standings as things will be a bit less stressful when UT2003 comes around.

CPU Scaling under Unreal Performance Test 2002

To kick off our CPU scaling tests we use the Unreal Performance Test 2002 with our barrage of CPUs. The graph is fairly self explanatory so we'll only point out interesting trends.

If you look at the slowest CPU speed (800MHz) you'll realize that with the exception of the GeForce3 and GeForce3 Ti 200, the GeForce4 Ti 4200 can offer a decent performance improvement over any of the competing cards in future games based on the Unreal Engine.

It's also interesting to note that with the exception of the GeForce4 Ti 4200, the rest of the contenders don't really scale well with CPU speed. After the 1.2GHz mark pretty much all of the cards are running on flat scaling curves meaning that they've run out of juice.

Luckily we won't see this strenuous of a situation for some time to come, let's have a look at some currently available games.

Serious Sam 2: The Second Encounter

The publicly available Serious Sam 2 demo was used for this test. We ran the Extreme Quality add-on to override any autodetected graphics settings, then set our resolution and disabled anisotropic filtering for the benchmarks. We picked the first demo as our benchmark level of choice. We recorded the average frame rate without excessive peaks.

CPU Scaling under Serious Sam 2

Here we have noticeably flatter CPU scaling curves; after the 1GHz mark even the GeForce4 Ti 4200 begins to flatten out indicating that with all of the visual features turned up at 1024x768, the Serious engine is running into video card bottlenecks. Here we have to look at acceptable frame rate levels and pretty much everything below the GeForce3 Ti 200 and Radeon 8500LE are a bit lower than we'd like to see.

devmap ffa_bespin
timedemo 1
disconnect
demo jk2ffa

Finally at 1280x1024 things get a little more interesting but it's clear that Jedi Knight 2 is far from GPU limited, let's have a look at the CPU scaling graphs to see exactly what's going on.

CPU Scaling under Jedi Knight 2

Just as we mentioned on the previous page, under Jedi Knight 2 the performance of the majority of these cards is governed primarily by CPU speed. The Radeon 7500 tapers off after about the 1.53GHz mark and the GeForce2 MX 400 is stuck at its 57fps regardless of CPU speed but other than those two the performance scales very linearly. If you're a big Jedi Knight 2 fan you'd be better off with a more powerful CPU than you would be going after a faster graphics card.

Comanche 4

For years we've searched for a good flight simulator benchmark and we've always run into the same problem, we've never been able to find one with built-in benchmark capabilities. Luckily, NovaLogic released a benchmark demo of their DirectX 8.1 Comanche 4 flight sim. We ran the publicly available demo (which you can get here) with all options set to high and sound disabled.

Much like what we saw with Jedi Knight 2, Commanche 4 seems to be very CPU bound at 800x600. Let's see how things fare as we increase the resolution a bit.

CPU Scaling under Comanche 4

Under Comanche 4 it's clear that the game is very CPU dependent but it's also interesting to note that as CPU speed does increase, the performance difference between the Ti 4200 and the rest of the contenders grows considerably. It grows to the point where at 1.73GHz there is a noticeable performance difference between the Ti 4200 and the rest of the pack, a performance difference that is relatively nonexistant at lower speeds.

With the GeForce2 MX 400 thrown in primarily as a reference point, it's clear that if you're running that slow of a video card it's definitely time for an upgrade before you even touch your CPU.

Even at 1280x1024 the performance of almost all of the contenders is well above respectable, only the old GeForce2 MX 400 is unable to keep up at this high of a resolution.

CPU Scaling under Quake III Arena

You can see that after a certain point these cards begin to run out of juice under Quake III Arena. Even the Ti 4200 exhibits some leveling off while the rest of the contenders are clearly flattening their CPU scaling curves either just after the 1.26GHz mark or well before it.

But with frame rates almost all above 138 fps is there any reason to complain at all?

Return to Castle Wolfenstein

Another game based Quake III Arena engine, RtCW proves to be a much more stressful implementation of the engine. We tested with the game's High Quality defaults. It should be noted that with the texture detail set to high (part of the High Quality defaults), performance on cards with only 64MB of RAM become much more erratic than those with 128MB. There was a considerable amount of swapping which was noticed during the tests.

CPU Scaling under Return to Castle Wolfenstein

The CPU scaling chart is very similar to what we saw under Quake III Arena with the exception that the scaling curves are not nearly as smooth. As we mentioned at the start of our RtCW tests, this is mainly due to the swapping that occurs when texture detail is set to high.

Final Words

If it isn't clear by now this should help considerably, NVIDIA's GeForce4 Ti 4200 is by far the best card under the $200 mark for those purchasing in the very near future. It does very well against ATI's Radeon 8500LE and manages to perform well even on slower CPUs. Our CPU scaling charts can help you pick out exactly what would offer you the best bang for your buck depending on your CPU speed but in most situations you'll get very good performance out of the Ti 4200.

Provided that availability isn't an issue, the 128MB GeForce4 Ti 4200 cards should be sought after instead of the 64MB cards. While you should be able to overclock the 128MB cards to the 64MB speeds, you can never add additional memory to a 64MB card which will hurt you down the road. We are already seeing that in some of our tests today and games will only become even more demanding on memory sizes going forward.

Let us not forget about the Radeon 8500LE however as it is the primary reason that we have the GeForce4 Ti 4200 today. If it weren't for the Radeon 8500LE we would be left with a DX7 compliant GeForce4 MX to tend to the mainstream market. Unfortunately for ATI, the card that was designed to put the GeForce4 MX to shame ended up resulting in the pending release of the Ti 4200 which is a bit more than the 8500LE can offer.

This is the perfect example of where competition truly brings us better choices in a market, it'll be interesting to see what happens if a capable third manufacturer returns to the scene. With DirectX 9 on the horizon the major hardware firms are diligently at work on their next-generation GPUs, from R300 to NV30 to something else being shown now behind closed doors - as usual, the graphics market always keeps us busy. Stay tuned.