Chris123234
Regular
So fascinating if they pull it off and it is competitive. Even more amazing that it was a 3rd party solution, if so.
LucidLogix Hydra, madness?
- Thread starter Citrous
- Start date
PCPER has a nice article too. Comes with a 20min interview/demo
http://pcper.com/article.php?aid=785
EDIT: Mods please delete. Repost.
http://pcper.com/article.php?aid=785
EDIT: Mods please delete. Repost.
Last edited by a moderator:
Pssst... look a few posts above you. Arty posted it already.
Anyway, from what I'm hearing the MSI Big Bang works pretty well...
hopefully we'll see some indepth reports about it next month...Did I really hear him correctly at about 3:15 into the interview that the chip is 23mm per side? I seem to remember them saying the new chip is in 90nm, but that's >500mm^2. For reference a G71 was ~200mm^2 and R580 was ~300mm^2, both on 90nm.
Edit: actually it looks like the Hydra 200 is on 65nm. If it really is 23mm per side, that makes it about the same size as a the 65nm GT200.
Edit: actually it looks like the Hydra 200 is on 65nm. If it really is 23mm per side, that makes it about the same size as a the 65nm GT200.
Last edited by a moderator:
and GT300, both at 24x24.
I don't get it. There's the distinct potential for different boards, especially from different vendors, to render different pixels for any given frame. That's a huge barrier to it working properly, never mind mismatched performance and the potential for that to pollute the experience.
What am I missing?
What am I missing?
Chris123234
Regular
It's much worse than that, especially for geometry processing. If the 2 chips produce just slightly different rounded results for geometry you can easily have parts of the scene moving around by a few pixels between frames depending on how the workload is split. Also within a single frame you can have clipping errors and holes in the scene where polys don't quite line up like they should.
Nvidia's 3d stereo drivers occasionally have issues dealing with geometry not matching for the left/right variant images, and that's on 2 identical cards on a very controlled pipeline.
If it was simply an issue of slightly different texture filtering implementations it wouldn't be too huge of a deal, but still would be annoying. Geometry differences can completely break the scene.
ChronoReverse
Newcomer
Yeah, I'll reserve judgment for when it's actually available in retail, but it really seems to me that if this sort of thing is actually possible, then ATI and Nvidia have been Doing It Wrong (TM) for the last long while with Crossfire and SLI.
I can assume communication is a lot better on the Hydra200 board. the Chip assumes command of the Graphics Lanes to the Lynfields. Reducing overhead on communication. Especially since inter card traffic over PCIe doesn't need to pass through to the CPU but stays on the hydra.
One way to truly scale performance is if setup and composition happens on the Hydra.
One way to truly scale performance is if setup and composition happens on the Hydra.
If they took another approach, the GPU's are not the solution, but the problem that needs solving.