AGP 8x is here
This one goes to all the smartasses who claimed AGP 8x was ways off still:
No longer: Via just released the P4X333 chipset:
DDR 333, AGP 8x, USB 2.0 etc.
Only weak point seems to be the bus between the North and the southbridge: 533MB/sec ain't that great for a bus that has to handle the I/O of 6 PCI slots, 6 USB 2.0 ports and all the old school connectivity.
For details go to: <a href="http://www.tomshardware.com"; target="_blank">http://www.tomshardware.com</a>;
G-News
No longer: Via just released the P4X333 chipset:
DDR 333, AGP 8x, USB 2.0 etc.
Only weak point seems to be the bus between the North and the southbridge: 533MB/sec ain't that great for a bus that has to handle the I/O of 6 PCI slots, 6 USB 2.0 ports and all the old school connectivity.
For details go to: <a href="http://www.tomshardware.com"; target="_blank">http://www.tomshardware.com</a>;
G-News
Comments
<strong>Can current cards even come close to saturating an 8x AGP bus?</strong><hr></blockquote>
Considering that the new Matrox card has 20 GB/sec of memory bandwidth, the answer to that is a resounding "yes".
<strong>
Considering that the new Matrox card has 20 GB/sec of memory bandwidth, the answer to that is a resounding "yes".</strong><hr></blockquote>
The Amiga guys leaked info about the MATROX turned out to be pretty good. Does that mean MATROX plans to support PPCs with its new card or are the amiga folks just trying to get attention?
Btw: the mentality "oh do we need that, it's not going to be of interest for the next 2 years" is so fataly wrong, it almost hurts.
In the computer industry newer = better, we don't need the Apple "sufficiency" tactic here.
If you can get AGP 8x, you want to get it, same with ATA 133 vs 66 in a 3000$ machine.
G-News
[ 05-16-2002: Message edited by: G-News ]</p>
Oh well.
GeForce 6 on an 8x AGP slot in a 1.6 GHz dual PM with 333 DDR RAM and a 566 MHz bus...
G-news
Do they double the bus width or do they double the actual clock? How the heck could they send 8 bits per clock? 1x, easy. 2x, 1 bit rising 1 bit falling, also pretty easy. 4x same as 2x but using two out of phase waves, OK not too hard. 8x, I don't think they're using 4 out phase waves (or one every 90 degrees) wouldn't this cause problems? Or is modern logic control so good that this isn't a problem anymore?
programmer, where are you? someone please edumacate me.
<strong>The 20GB/sec is for the on card texture memory only, I haven't seen a published throughput for on/off card traffic. HUGE difference there. Matrox's card looks just a tad nicer than 3DLabs as it can do 3 monitors out of the box, but I think they were playing press release wars this week. Lets see shipping hardware running real applications for starters.
I also have a sneaking suspicion nVidia shouldn't be considered dead until they reveal their next chip (ATi ???). They promised radical improvements and a different architecture too, just no details. I think August will be interesting.</strong><hr></blockquote>
The question was whether modern graphics cards could fully utilize AGP 8x. Since they can fully utilize 20 GB/sec texture memory, then its a safe bet that they could also fully utilize an AGP 8x bus if they were updated to support AGP 8x properly. Note that a seperate question is whether system memory could keep up with AGP 8x.
Only a fool would consider nVidia or ATI dead just because Matrox announces (but does not ship or even announce pricing) on a DX9 card. Even if they were shipping today at $199 they wouldn't conquer the market because they simply can't build them fast enough. And does anybody think that the industry leaders don't have the next 2 or 3 generation GPUs already in development? Fear not, this year will represent a revolution in GPUs just like the last two or three years and all the major players will be there in force.
I have seen the future, and it is awesome.
<strong>programmer, where are you? someone please edumacate me.</strong><hr></blockquote>
Heh, well since you asked by name I guess I have to at least try.
Tom's Hardware says AGP 8x is 32-bit 66 MHz octuple pumped, giving bandwidth of 2.1 GB/sec (which matches 266 MHz DDR nicely I might point out). Intel has also defined a 64-bit extension to AGP 8x, which presumably could double the throughput to 4.2 GB/sec.
As to how they achieve octuple pumping, I don't actually know. They haven't added any more lines because it is backward compatible with AGP 4x (only, not AGP 2x). There were a couple changes to support "the modified signaling". I'm a programmer not an electrical engineer, so I'm afraid I can't explain the signalling details (my electronics knowledge is limited to basic circuits). Obviously they are generating an extra transition, but I can't even guess what magic they pull to do that.
You'll often notice that both Agp and PCI cards often don't use all the possible pins, also it's plausible that from one standard to another, a pins function might be changed, from say ground to a signal.
But I don't know exactly either, but it's nice to know its there
G_News
<strong>Only weak point seems to be the bus between the North and the southbridge: 533MB/sec ain't that great for a bus that has to handle the I/O of 6 PCI slots, 6 USB 2.0 ports and all the old school connectivity.
</strong><hr></blockquote>
Actually, the max bandwidth for 6 PCI slots is exactly the same as the max one for a single slot. That's how buses work.
Bye,
RazzFazz
133MB or mroe from the PCI, up to 133 from the ATA, over 50 from USB, most certainly actually 3x 50...it's sufficient probably, but it's tight.
G-News