AAlchemy 8464DB
Quantum3D’s third generation, open architecture family of advanced, scalable PC-IGs that combines sixteen VSA-100 chips, 1 GB of dedicated graphics memory and 42.5 GB/sec graphics bandwidth per channel to deliver up to 890 Megapixels per second at resolutions of up to 1600 x 1200 pixels, with trilinear filtering and T-buffer™ full-scene, sub-pixel anti-aliasing enabled. AAlchemy features the essential capabilities required for realtime training applications-- including high-resolution output with multiple pixel formats and full-scene anti-aliasing, advanced texture compression for enhanced on-board texture storage and texture paging performance and precision, hardwarebased inter-channel synchronization (SwapLock™). Available with optional external source synchronization (QSync™) and the nVSensor™ sensor post processing subsystem, AAlchemy is an open architecture PC-IG that supports the 3dfx® Glide™ API as well as SimGL™, Quantum3D’s vis-sim optimized graphics API that is based on the published OpenGL® API (but which is not an implementation that is certified or licensed by Silicon Graphics, Inc. under the OpenGL API). Both Glide™ and SimGL™ are widely employed by industry leading realtime 3D scene managers and high-level APIs, including MultiGen-Paradigm Vega. AAlchemy’s open architecture, unrivaled performance, image quality, key simulation features and low cost has enabled the deployment of advanced synthetic environment training applications on PC-based systems-- which results in tremendous savings in acquisition, maintenance and development costs over proprietary alternatives. Consequently, AAlchemy™ has been adopted by leading training systems integrators including Lockheed Martin, Thales, Alenia Marconi, Indra, Honda, Mitsubishi Precision, Adacel, Raytheon, Boeing, Soft Reality, Raydon, SDS International, Aerobel and others.



About nVSensor
Developed for precision training and simulation applications that require high-fidelity night vision goggle (NVG) and/or infrared (IR) synthetic environments, the Quantum3D nVSensor is a programmable realtime image processing graphics subsystem that adds physics-based special effects, symbology and reticles to imagery rendered by the AAlchemy graphics subsystem. Designed to preserve the precision of the AAlchemy graphics subsystem, nVSensor interfaces with AAlchemy via a 30-bpp digital pixel port and provides user selectable color space conversion to RGB values or gray scale, with support for white-hot and black-hot object definition. For integration flexibility, nVSensor provides both 24-bpp digital (DVI) and 30-bpp analog RGB outputs. nVSensor employs a programmable, multi-stage pipeline architecture that enables the subsystem to provide the most advanced sensor effects available in a PC-based product today without affecting core AAlchemy rendering performance or adding latency or transport delay to the overall graphics pipeline. Special effects currently supported by the nVSensor Effects Library include blur, noise, gain, bias, reticles and symbology, AC coupling, automatic gain control, polarity inversion and white-hot/black-hot settings.

Correct!

No penalty for FSAA, Free FSAA. 16 X SLI VSA-100s

So much Power, it's kinda Ridiculous!! 

gamma742 wrote on 03.07.08 at 17:39:03:


The statement "No penalty for FSAA" is not exactly true.  I have spent a significant amount of time experimenting with my Q3D SX system that has a single board 8232SB card in it.  The "FSAA hit" the user experiences really all depends on two things:

1)How fast your cpu is?  The fastest SX system Q3D produced had dual 1 Ghz Pentium 3 CPUs.  I don't know of any games from that era that could utilize dual CPUs reliably.  So, you are essentially running with a 1 Ghz CPU.  To get the maximum performance out of the AA5 architecture, 1 Ghz is way too slow.  It just can send the T&L data to the card fast enough.

2)How you configure the drivers.  I discovered a little know secret with the Q3D drivers.  If you want to see maximum performance, hard select no FSAA (as in 0X).  Don't mess with using the toggle key for this test.  The Q3D drivers basically won't dynamically reconfigure the chip/FSAA architecture on the fly.  This means if you select 8X FSSA, it will only run as fast as a single VSA-100 chip can go, even if you toggle FSAA off.

OK, so what I did is run a little test on my custom 8232SB system at 1600 x 1200 running Quake 3 1.32 point release.  I say custom because I ditched the dual Pentium 3 Tyan mobo and installed my Pentium 4 2.8 Ghz mobo/CPU combo with 1 GB of DDR SDRAM.  Here are my results:

Quake 3 (v1.32, MAX detail settings, "four" demo)

No FSAA:   86.0 fps
4X FSAA:   32.0 fps
8X FSAA:   13.3 fps

This clearly shows that FSAA isn't free.  You just might not see the difference because of the driver config and/or CPU speed.  When properly configured, the board will indeed turn into a 3dfx speed demon.  Even with the original Pentium 3 setup, I could achieve frame rates in the 70 fps range at high resolutions.  It's all about getting the card T&L data fast enough.  If I totally turn off all the detail settings and lower the resolution, I can get about 132 fps with the card.  I'm not sure if a faster CPU beyond the 2.8 Ghz would result in much more gains in speed.

Remember, my setup is a single 8232 card.  I have no idea what an SLI config will do.

A few more factoids about the AA5 I discovered:

-I couldn't get the drivers on my system to turn off vsync.  No matter how I set the config panel, it wouldn't turn off.  I managed to hack the registry after hours of hunting and experimented.  I later discovered uninstalling and reinstalling the drivers also fixed the problem.  The downside is it totally broke the OpenGVS demo software.  I could never get it to completely work again.  I'm actually running now with a completely clean install of W2K.  I didn't get the OpenGVS demo disk with my system, so I haven't been able to reinstall it.

-I've noticed the AA5 numbering system is often misinterpreted by folks.  Often, the second number in the string is considered the number of boards in the "stack" by folks.  It is in fact the number of pixels per clock cycle the board can render.  The number of boards is indicated by a trailing "SB" or "DB".  All this info came from the Q3D SX owner's manual.  Here are a couple examples:

8132SB (1 pixel per cycle, 32mb/chip, single board)
8232SB (2 pixel per cycle, 32mb/chip, single board)

I'm not sure of the exact configs that were sold, but it looks like gamma742 scored the ultimate setup.  I'm actually very curious to know how well the SLI config will run because I have a second functional 8232 single board w/power board and housing that I wouldn't mind adding to my system.  I just need to find a set of SLI connectors that go between the cards.  Anyone have a spare set;)

Sigtau

Surprisingly, it's not as heavy as I thought. I think my Mercury "Brick" is heavier that the 8464.

Quote:


You're right about that. You've done your home work.

I haven't had time to pull my 8232SB out of my AAlchemy to test the
8464DB yet but I will say the 8232SB is much faster than the Mercury
System and both perform FSAA extremely well.. IMO.

Bob~

Gratz man! Finding a setup like that is a very nice find. Let us know how that thing comes along as I'm sure every one here is interested in finding out what that thing can really do.

no picture ?

Okay my friends, Pics as promised.

I installed and tested the 8464DB:




In the pic above, I am trying to show the two SLI connectors



In this pic I am showing additional memory on the cards near the heatsinks.



In this pic you see the 8464DB next to my 8232SB (removed)



You may need to click this to max res to see it detected as an 8464DB



This pic show it all!! The 8464DB running the Quantum3D GVS Demos
with my 8232 pulled out to the side and my signature 2 dogs napping
on the couch in the background.  

Now I'm not an expert but I thought the second number in 8x32 or 8x64 stood for the amount of cards?

If so, should your system not be called an 8264 system?
I'm only seeing two cards.