Like some PCI PC-TV cards the Apocalypse cards send all their data over the PCI bus directly to the frame buffer in your 2D card. Using the PCI bus avoids signal degradation problems that can be produced with external link-cables.
The PowerVR chip at the heart of the Apocalypse cards has a number of radically different features to all the current 3D accelerators. The chip's design includes Infinite Planes, a full 32bit floating point on-chip Z-buffer and 'Tile' rendering. For more details see the
The new PowerVR2 adds Bilinear and Adaptive Bilinear filtering, 2D/3D compositing on-chip and higher performance than the original PowerVR1 chip. The problems associated with Matrox graphics cards and the earlier PowerVR1 cards are no longer an issue with the new chip.
I already had the earlier PowerVR1 based Apocalypse 3D installed in my machine and so I was interested to see if the cards would be confused. I decided to simply switch off the machine and swap the cards over. On Power-Up Windows 95 detected the new card and asked for the driver disks - So no confusion. Once the driver installation was complete, the machine re-booted and was ready to go. (If you do not already have DirectX version 3.0 or higher installed, you must install it at this point.
Unlike the drivers for the original PowerVR1 card, the small yellow configuration 'Triangle' on the tool-bar has now gone and been replaced with an additional Tab in the 'Display Properties' window. This can be selected from the Control Panel, or more directly by Right-Clicking on the Desktop and selecting Properties.
The Apocalypse 3Dx control panel allows the user to configure the drivers for either Maximum Performance or Maximum Compatibility and to simply turn the PowerVR2 ON and OFF. The ability to quickly turn the PowerVR2 ON and OFF from the control panel, means that it is possible to operate more than one manufacturers 3D card in the same machine.
If you want to get a little more sophisticated, troubleshoot, or just optimize for maximum performance you can use the advanced button to set individual settings.
You can see that there is also a lot of preconfigured programs.
Some of you may remember the big debate regarding whether VideoLogic was cheating on the D3D benchmark tests. VideoLogic here is proving that what they did for the test can be done for an individual game- you can continue to call it cheating, but I am happy to see the effort to strive for compatibility and better performance without causing everyone unnecessary graphical corruptions.
Direct3D, like all DirectX libraries operates using Compatibility Bits'. The software writers must check these to see which function a 3D card supports and work with them. The Compatibility mode has been added to allow some games to run that assume certain Direct3D functions are present. This compatibility is achieved by emulating/transalting the missing functions. This inevitably leads to lower performance. For all the Benchmark tests the drivers were set to Maximum Performance.
The following benchmark results were obtained with the current Version 4.0 drivers posted on the Videologic Web Site.
The Pentium tests were run on a Giga-Byte 586DX Dual Pentium motherboard (only 1 processor fitted) and the 2D card used was a Matrox Mystique (4MB). All the Pentium Pro tests were run on a Microstar 6103 Dual Pentium Pro motherboard (again only 1 Pro fitted) with a Videologic Grafixstar 600 (4MB) used as the 2D card. Both machines were run at standard PCI bus speed of 30MHz and 33MHz, dependant purely on the processor being tested.
The 'PC Player Direct 3D Benchmark' (vr 1.1) was produced by the German 'PC Player Magazine'.
This first graph shows the 'PC Player Direct3D' results with a wide range of Intel Processors
The Pentium Pro 133MHz results were produced by under clocking a Pro 166MHz 512k processor. The Pentium Pro 233MHz results were produced by over clocking a Pro 200MHz 512k processor.
A few interesting points are shown with this graph. Firstly the Pentium Pros with 512k cache give far better performance than the 256k cache chips (A Pro 180 256k is almost matched by a 133MHz 512k Pro.
Secondly the clock speed of the PCI bus is important with this test. The Pro 166Mhz 256k results are actually higher than the Pro 180 256k. With the Pro 150 and Pro 180 chips the PCI bus operates at 30MHz rather than 33MHz.
The 3D chip manufacturer 3Dfx, produced the Wizmark 3.0 benchmark. The Wizmark benchmark displays a room with textured walls and floors, vases, two suits of armour and various other complex objects. The benchmark result produced, is an average frame per second (fps) reading, for moving around a preset path inside this room.
The Microsoft Direct3D benchmark (called D3Dtest.exe) is supplied with the DirectX SDK. The Direct 3D benchmark measures Polygon Throughput, Pixel Fill Rate and Intersection Throughput. These tests can be run individually or all together, and are calculated both rendering front to back (ftb) and back to front (btf). The results presented here are an average of the 'ftb' and 'btf' values.
The tests are also performed using 'with Updates' and 'without Updates'. The 'with Updates' setting, forces the 3D card to display the test images as they are generated. The 'without Updates' setting leaves a black screen during the tests. This second setting is designed to measure the raw performance of the 3D card, by removing any effects due to the 2D card.
The second graph shows three sets of benchmark results
The Apocalypse 3Dx can only run Wizmark in "Lite" mode, because it does not directly support the particular Alpha Blending mode needed under Direct 3D. The D3Dtest results
do not vary when the program name is changed.
All three graphs clearly show that the Apocalypse 3Dx just gets faster as the processing power increases. Even an Intel Pentium Pro 233MHz (O/C 200) does not appear to drive the card to its limit
Interestingly the Wizmark results appear to track the Direct3D results! This indicates that the Wizmark Lite benchmark is actually just a measure of the cards Polygon Throughput.
The Intersection Throughput is a measure of the 3D cards handling of hidden surface removal and triangle culling
These results show a high dependence on the raw processing power of the CPU, rather than the more usual combination of CPU and bus bandwidth shown in the earlier graphs. The cache size in also far less important, giving only a few percentage points improvement in the results. Again the Apocalypse 3Dx has not reached a limit even with the fastest processor tested.
Pixel Fill Rate
The fourth and final graph shows the Fill Rate results from the Direct3D test
This test is a measure of 3D accelerator's rendering performance. The graph actually shows the Apocalypse 3Dx quickly reaching a limit with a Pro 150MHz 256k.
From the first three results for Pentium processors, a 200MHz Pentium would also reach this same limit of around 40million pixels per second. The slight increase in Pixel Fill Rate shows a small dependency on the Processor, cache and PCI bus speed. (This I would guess is most likely a limitation of the benchmark.
What does '40 million pixels per second' actually mean?
Well take a 1024*768 screen, this contains 786432 pixels. So 40Millions pixels per second would give a maximum frame rate of 50 frames per second. Not bad
In fact assuming pixels are 16bit colour, then this equates to 80Mbytes per second over the PCI bus.
(Unlike some other 3D accelerators, the PowerVR design means that any pixels sent to the screen are
pixels. I.e. the on-chip Z-buffer means that no pixels should be over written when rendering each frame.
A complete table of all the results from the tests
The Apocalypse 3Dx is a very quick card, as measured by these benchmarks. But benchmarks are only part of the test of a 3D accelerator. The final test will come from the games, either though direct ports to PowerSGL (The PowerVR API), or from the more universal APIs; Direct3D and OpenGL.
But so far things are looking great for the new Apocalypse 3Dx.