New Gigabyte MB: 4*NV GPUs; also 32*ATI WHOA!

karlotta said:
Entropy said:
...Come on, spell it out!
W O R K S T A T I O N
Stopped caring once the high end vector terminals bit the evolutionary dust. Now those had real framerate. And they didn't need any AA either like those newfangled raster abominations. Ugh. ;)

Seriously, workstations are tools. Hot and noisy computers belong in a different room from people. Access only as needed. Cost is largely irrelevant of course at these price levels but ergonomics matter.
And workstations weren't the issue under discussion as far as I could see. The technique does have applications there, but it's pretty damn narrow even in that niche.
 
Entropy said:
And I suspect this reluctance is because stating the actual benefit in real terms makes it bleeding obvious that noone in their right mind would find it particularly worthwhile to spend $500 or so plus hassle and noise to be able to play DOOM3 at 1600x1200 4xAA instead of using 1280x1024 4xAA.
Who the hell gives a shit about something like that? Now really?

So you'll think that you'll be able to play UE3 games at 1280*1024 with 4XAA with today's GPU's, and at a steady 60 fps ?

Sure, a quad GPU system is of course a super niche product but i definitely think that there's a market for it. Think about game developers f.e. And all the 3D Mark .. "i'll put my 4000$ computer in the fridge so that i'll be able to get 500 more points" will certainly buy it :)
 
Entropy said:
The obvious question then being - where does it make sense?
Driving ultra-high res displays (or multiple displays, provided the IHVs implement a multi-display mode), supersampling fanatics, offline or preview rendering, developers planning for the future, GPGPU algorithms that are not suited for networked distributed computing.

Sure it's not aimed at the ordinary gamer, but hey, this isn't "3D Game Technology & Hardware" ;)

The nice thing about graphics is that you can just up the resolution or AA and be GPU limited again.
Sure, but - why? What's so grand about being GPU limited?
The only thing multi card rendering buys you is pixels. Not features.
Go dual and you can reaise resolution one notch, if you're completely GPU limited. Go quad and you can raise it another step. That's it. Sure it's a benefit but hardly much to get excited about, now is it?
I didn't mean to imply that being GPU limited is a good thing, just that even with a "slow" CPU you can put more graphics performance to good use.

Yes, it buys you pixels, not features. The same applies to high end cards compared to midrange cards. Same featureset, just more performance. Would you say high end cards don't make sense? ;)

Multi card rendering only help with very particular limitations. Not "rendering" in the broader sense by a far stretch.
True enough, I should have said fillrate (and limitations due to RTT).

What bugs me about the proponents of SLI/AMR/whatever is that they steadfastly refuse to discuss the benefits in real world terms. You are just the most recent example Xmas, and you are more than bright enough to be aware of what you're doing. What you say is mostly correct, but the analysis of what it actually means in real world terms is lacking.
And I suspect this reluctance is because stating the actual benefit in real terms makes it bleeding obvious that noone in their right mind would find it particularly worthwhile to spend $500 or so plus hassle and noise to be able to play DOOM3 at 1600x1200 4xAA instead of using 1280x1024 4xAA.
Who the hell gives a shit about something like that? Now really?
Going SLI might also buy you some cred points in some very narrow and none too bright circles, and if you belong to such circles, then fine. Or if you always dreamt about having a Voodoo SLI setup but could never afford it, and now that you have a job that pays OK you decide to splurge just because you can. That's fine as well.

But take the benefits from the purely theoretical "you can gain benefits in fill rate limited cases" to actual examples which show just how much better your real life gaming experience gets. To some extent I really do wonder, because I have 80 or so titles standing behind me on the shelves, and I just can't see my gaming experience getting one whit better by going SLI. Come on, spell it out!
I'm not a proponent of SLI in the sense of "I want to buy a SLI setup", be it two or four cards. In fact I don't. I can't afford it, and if I could I knew a thousand ways to spend that money in a better way. However, if I had still some money left after that other thousand things...
See, you and me feel it is a small gain at a high cost. But for someone else maybe the cost isn't high at all, or the perceived gain is bigger.
I don't dismiss an idea as useless just because it doesn't suit me, if I can se how it could be useful for someone else (and it doesn't hurt me).
 
rusty said:
yes, but what benefits are you actually getting there... monitors can only display so much resolution, games only have textures with a ceiling resolution in mind. AA if you're pushing the higher resolutions becomes less of an issue anyway, and there's a point after which anymore AA is not needed.
In theory, there actually is no such point. ;) And practically there's still a long way to go until more AA won't make a difference.

the main point of my argument is that Intel 17 said "heh, ins't the pc market supposed to be dead and gone to the consoles? :rolleyes:" and i'd say if this is what their answer to the consoles is then yes, they really are in trouble. It is an overly expensive solution of questionable benefit....
I guess I misunderstood jimpo's point a bit, sorry.

I don't think this is an answer to match the consoles at all. Neither is it the way the PC market is going, it is but one of many configuration options and I doubt it will become widespread. Multi-GPU/die on a single card/package maybe, but that would be to overcome limitations in manufacturing.
 
in that case i am not really in dsiagreement with you...

my comments are mainly aimed at intel17's remarks.

and yes of course in theory there is no such point, but diminishing returns has to rear it's ugly head!
 
This is pretty interesting combining the chipset from the Intel and AMD sides but how will these work together? So there are40 total PCI-E lanes right? So does this apply to be 4 8x for 4 cards or 2 16x for 2 dual gpu cards? And why on earth are they not doing this board for 939 and do what Tyan did on the K8WE and inculde the two chipset nvidia created for that platform, they could still get the 40 total lanes. I don't understand why they did this?
 
I'm guessing the reason for Intel-only is that it means they can use an Intel North Bridge and and AMD South Bridge, thus getting two SLI-certified chips on the same motherboard. You couldn't do the same with AMD as A64s don't use North Bridges.

Does nForce 4 for 775 communicate internally using HyperTransport? That would make sense in this situation.
 
From what I gather though nvidia has two different version of the Nforce 4 chipset.

Chipset Details
Tyan's Thunder K8WE utilizes a combination of three individual chips onboard to allow for so much onboard connectivity, the nVidia nForce4 Pro 2200 primary chip, the nForce4 Pro 2050 secondary chip, and the AMD 8131 PCI-X controller hub.
The nForce4 Professional 2200 is the "primary" chip of the nForce4 chipset, as this chip supports 16 lanes of PCI Express connectivity, Gigabit LAN, and four SATA-II/300 ports, along with all the extras which go along with the nForce4 design (USB, PCI, IDE, etc). Alone, the nForce4 Pro 2200 is similar in design to the nVidia's own nForce4 Ultra for the AMD Athlon64 platform.

The nForce4 Pro 2200 can be paired with an additional chip to supplement its abilities, which is the nForce4 Pro 2050. The nForce4 Pro 2050 is very similar in design to the 2200, but is neutered to help keep costs down. The nForce4 Pro 2050 can supply an additional 16 lanes of PCI Express bandwidth, an additional four SATA-II/300 ports, and an additional Gigabit Ethernet port. The 2050 however, cannot support USB 2.0, PCI, or IDE devices. When the 2050 chip is added to a motherboard, you basically double your PCI Express bandwidth, SATA RAID port capabilities, and GigE port capabilities compared to a single chip nForce4 Pro 2200 setup.
- gamepc.com http://www.gamepc.com/labs/view_content.asp?id=thunderk8we&page=3

So why not do this instead?
 
Charmaka said:
Does nForce 4 for 775 communicate internally using HyperTransport? That would make sense in this situation.
I think that's what they've been using for their NB<->SB interconnect since the very first nForce.
 
incurable said:
Charmaka said:
Does nForce 4 for 775 communicate internally using HyperTransport? That would make sense in this situation.
I think that's what they've been using for their NB<->SB interconnect since the very first nForce.
That makes perfect sense then. If HT is used between NB and SB on Intel designs and between CPU and SB on AMD designs, coupling an AMD SB to an Intel NB should be a piece of cake, comparatively speaking, as they moth already use the same interface.


As to that Tyan board, I don't recognise the designations but it sounds like those are server-oriented variations of the chipset. As to why they're not being used, I'm guessing the answer is "cost".
 
Xmas said:
jimpo said:
Well, if this is really the way PC market is going to match the new consoles, it doesn't look too promising. Putting four separate video cards on one motherboard is hardly the optimal way to achieve performance, it smells of a quick-and-dirty kludge.

Well, what is the optimal way to achieve that level of performance, then? "Wait for the next gen" isn't really a solution.

There isn't one, currently. At least not a practical and realistic way to compete with consoles that will cost $500 at most (and possibly as low as $300).

My post was a response to "Heh, isn't the PC market supposed to be dead and gone to the consoles?" and my point was that this solution is very irrelevant when thinking about PC market vs. console market.
 
Damn as someone aptly pointed out...the consoles AREN'T taking over as of yet. I really want to get an SLI or Multi VPU setup with greater than 4 cards. Its not just a matter of bragging rights :devilish: but I just want to have a nice multimonitor setup and play games @ HD resolution at least like the way the consoles are. I read the PS3 can drive 2 HD monitors and all the game content is HD res.
 
Dr Evil said:
Charmaka said:
You couldn't do the same with AMD as A64s don't use North Bridges.


What does that mean? I read many articles that mentioned Northbridge on A64.

Usually a Northbridge handles two things: memory controller and video controller. A64s have the memory controller onboard (which you could count as an onboard Northbridge if you wanted to, minus the graphics), which makes a dedicated chip just for the video controller kinda unnecessary, so it's shifted to the Southbridge instead and the CPU communicates directly with that. I believe the main reason for having an NB in the first place (apart from upgradability issues for mobo manufacturers) is to enable you to link very fast to the NB for memory speed/latency purposes and then have the SB on a slower link as speed is less critical; with memory controller onboard there's no reason to have such a fast link to an external chip. As I understand it.
 
Back
Top