Slashdot Mirror
Reviews Arrive For nVidia GeForce 6600GT AGP
bhtooefr writes "The Tech Report got their hands on a reference board of the nV 6600GT AGP, and did some benchmarks. Interestingly, even with a slower memory clock on the AGP card, it was FASTER in some benchmarks than the PCI-E card. Tests performed were: Doom 3, CS:Source, Far Cry, 3DMark05, Rome: Total War, and Xpand Rally (the last two tested with FRAPS)."
pacmanfan contributes links to more reviews at
Extreme Tech, Hard OCP and PC Perspective.
Actually a 9800pro runs about $200 which is about what the 6600 costs. And the 9800 was not keeping up with the 6600.
http://www.kubuntu.org/
On the other hand on some of the benchmarks I have seen that card is being run with 4x or even 1x PCIe bus, without any serious performance degradation, so, rather logically, this isn't the issue jet.
The lack of PCIe lanes to accomodate a lot of cards without tricks and headaches, yet is. But this is configuration issue, not bandwidth.
that's one of the main reasons, but it's not the main reason. the biggest reason that PCI Express is a lot cheaper to produce, which everyone loves (PCI Express is serial, AGP is parallel, AGP requires more connections between the card and the chipset on the motherboard, making it a lot more expensive than PCI Express for motherboard manufacturers). second of all, 3D hardware manufacturers love it because it isn't a bidirectional bus like AGP, where you have a single bus that typically moves data in a single direction but can move it in the opposite direction with a huge performance penalty. PCI Express is two unidirectional buses, so there's no penalty for reading from the video card. this means you can do all sorts of nutty effects and use the GPU as an extra processor in some applications a lot more effectively than you can with AGP.
your logic doesn't hold up, considering the Athlon64 has no PCI Express motherboard quite yet. sure, they've been announced, but they do not have any in retail. PCI Express was an Intel-led push, along with DDR2 and BTX (although we haven't really seen the last yet). it is simply much cheaper and much easier to manufacture than AGP. I mean, SLI was theoretically possible with AGP3.0 (introduced AGP8x, but it also had support for multiple AGP devices on a single motherboard). there were absolutely no motherboards, to my knowledge, that supported multiple AGP cards, certainly not in the consumer space. given NVIDIA's recent SLI push and ATI's forthcoming SLI chipsets, both would have hopped on AGP-based SLI if it were available. I'd guess that it was simply too expensive to make motherboards with multiple AGP slots more than anything else. with PCI Express, this limitation is gone.
The 6600 cards are pretty reasonably priced, so picking up two of them and getting 180% performance of a single 6600GT AGP is pretty attractive...
No, it isn't. According to reputable benchmarks, dual 6600 GTs ($200 x 2) typically perform slightly worse than a single 6800 GT ($400 x 1).
Why would you accept the undoubtedly higher power consumption of dual 6600 GTs versus a single 6800 GT, when the price is about the same?
Those who can afford a new motherboard (and probably a new CPU) just for the SLI capability won't be stooping to 6600 GTs; they'll opt for something better.
Erlang.org: wow
The parent is spot-on. Current AGP 8x bandwidth is 8x(266*8, 2128MB/sec), but the performance difference between 4x(266*4, 1064MB/sec) and 8x is negligable(around a percent, within experimental error). Considering that x1 PCIe is 250MB/sec, PCIe and AGP are effectively running at the same speed given the same multiplier. If you take in to consideration that we just said that we aren't making use of anything past AGP 4x yet, it's a logical assumption that PCIe x4 should also be enough, and that x8 would be enough for the next generation of cards that would somehow need the doubled bandwidth.
But getting back to the point, the current PCIe graphics standard is x16, which at 4GB/sec(and this is each way, BTW - PCIe is full duplex, AGP is half) is far more than we need. The current solution of dividing up the 16 lanes from that single slot in to 2 groups of 8 lanes for 2 PCIe x8 slots(though using an x16 connector for power issues) still results in each card recieving more bandwidth than it can effectively use. With a single x16 slot, PCIe is future-proof enough that bandwidth won't be an issue for some long period of time, and than the x8 SLI solution won't be bandwidth limited for some shorter, but still long enough period of time that it's not going to be a realistic issue until at least the 3rd or 4th generation PCIe motherboard chipsets are released, at which point they can be built with more lanes.