Technical questions
PH!: would you please explain your multi-chip technology called BitFluent? It seems there should be some sort of order between GPUs, at least at display data path. What about at rendering level, how do you allocate rendering jobs between GPUs?
XGI: For a more in-depth information, you check check on XGITech.com under the product section, there shall be a white paper explaining clearly how it works. Simply put, the two GPU works as follows: The master GPU is connected to the AGP interface and the Slave GPU is connected to the master GPU. Each of the two GPUs has its own local frame buffer, which is what we called DRAM. Basically, the master GPU will process frame 1, 3, 5, ... and slave GPU will process frame 2, 4, 6, ... and so on.
Hirdetés
From our studies, this is the most efficient way of putting two GPUs together. The overall efficiency is about 1.7* the performance of one GPU.
XGI Volari V8 Duo
PH!: still BitFluent: how is the GPU's memory access? Each GPU has its own memory area or they can see each others'?
XGI: That's correct, each has one set of a 128 bit wide, indepedent memory bus that it can access for highest efficiency.
PH!: Some 3DMark 2003 scores have been circulating on the web (originating from Anandtech's forum), regarding Volari's performance. They are:
- Volari Duo V8 Ultra - 5600+
- Volari Duo V5 Ultra - 4000+
- Volari V8 Series - 3000+
- Volari V5 Series - 2000+
- Volari V3 Series - 1000+
Can you confirm these numbers? If so, has any driver optimizations been taken for 3DMark 2003? If yes, can they be considered "legal" (meaning: no image quality degradation, no pre-rendering or calculating methods, etc.) techniques?
XGI: As the drivers get more mature, I believe there will be some changes to these performance figures.
XGI Volari V8 Duo
PH!: do you implement bandwidth-saving techniques in Volari-series chips? How effective is BroadBahn, how does it compare to solutions from ATI and nVidia?
XGI: ...
PH!: important question: how much power is needed for these chips? How much is the typical power consumption of the dual-GPU chip/board? Is there an external power connector like on high-end Radeon and Geforce FX cards?
XGI: The typical power consumption is about 70W. Actually there are two external power plugs needed for this board. This is to push the revolutionary 16 pipeline architecture.
PH!: how much is the transistor count, - at least roughly - in V8, V5, V3 respectively? What manufacturing process is XGI using to produce these chips?
XGI: It's based on UMC's 0.13 um process.
XGI Volari V8 Ultra
PH!: we have done several detailed image quality testings with GeForce FX and high-end Radeon cards in regards to FSAA and anisotropic filtering. In the end, ATI's true multi-sampling 6X FSAA was clearly the winner, but anisotropic filtering was also dominated by ATI. Please describe Volari's FSAA and AF techniques and options. Is Volari using pure multi-sampling, or mixed multi- and super-samping FSAA?
XGI: ...
A cikk még nem ért véget, kérlek, lapozz!
