• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

ATi HyperMemory Thread...

I

insanarchist

[H]ard|Gawd
Joined
Sep 3, 2004
Messages
1,387
This sounds promising, assuming PCI-E (or X, I forget which) has enough bandwidth to make this anything but a "Decent semi-integrated graphics" solution. What is the bandwidth potential on those things anyway vs. agp 8x?
 
If ATi can figure out how to put an inexpensive high speed, approximate 4MB framebuffer cache on the chip itself, then ATi will probably introduce mid and low-end videocards that have no RAM chips on them. And it could be faster than todays desktop cards.

Ditto with Nvidia. Its a prime technology for PCI-E, maybe not now as first incarnation PCI-E is still too slow (Yes, I said PCI-E is still too slow) But this is the better route to go.
 
Local Video Memory bandwidth (X800XT-PE) - 36.8GB/s
PCIe x16 Bandwidth - 8GB/s
AGP 8x - 2.1GB/s

Way too slow to be of interest to me. Cool name though! :)
 
Yeah, but it would also be possible for an onchip memory cache to double the speed of DDR3 memory and have zero or a half wait state, as opposed to one wait state (two inch traces to the memory causes delay)

So... If you could have 4MB of onchip framebuffer cache that runs at theoretical 40GB/sec with no wait-states that is used 100 percent of the time, and the rest of the textures (maybe needing 10 percent access) to go to the slower PCI-E it could still outperform a "normal" card today pretty easily.
 
Good point!

I'd guess we have quite a long way to go before uber high-end "no onboard ram" graphics cards, however I'd say for the integrated graphics chipset market, with the price of ram always dropping (not steadily every day, but at least every couple of months) I would expect that we could see some very decent integrated solutions for the low to possibly even middle-end segment. I'm just waiting for the entire northbridge to be re-integrated into the processor itself. THAT would be some massive bandwidth!
 
ZenOps said:
So... If you could have 4MB of onchip framebuffer cache that runs at theoretical 40GB/sec with no wait-states that is used 100 percent of the time, and the rest of the textures (maybe needing 10 percent access) to go to the slower PCI-E it could still outperform a "normal" card today pretty easily.
Click to expand...
Why not just stick a 4MB on-die framebuffer cache on a GPU and have local video memory for textures? What's so great about dragging hundreds of MBs of texture data across the PCIe bus? Ok, I see the benefits for low-cost video solutions but I'm really not interested in those.
 
Why not just stick a 4MB on-die framebuffer cache on a GPU and have local video memory
Click to expand...

Yup, could and probably will do that too for the high end boards. Depending on how well it physically works, depends a lot on whether we see it first without memory or bundled with memory in a high-end config.
 
i wonder what it would be like to have a video card where u gotta buy seperate video ram to stick in ram slots on the vid card, that would be crazy
 
You must log in or register to reply here.
Back
Top