NVIDIA Bridging It (PCI-Express)

[Dave Baumann]

NVIDIA PCI Express Technology at CES[/url]]PCI Express significantly increases bandwidth between the central processing unit (CPU) and graphics processing unit (GPU) by enabling a balanced distribution of bandwidth to those applications that require it the most. NVIDIA's PCI Express bridge chip, an industry first, affords the flexibility to instantly adapt a variety of the Company's award-winning GPUs to the new PCI Express bus technology with the full capabilities and throughput offered by PCI Express.

Although its a quite route to ensuring you are PCI-Express compatible without transisitioning your products all at once, this seems like a slightly odd move - you'll loose the benefits of the PCI-Express bus (as you are still operating at AGP8X - which makes the first line of that quote a little disengenuous in relation to what they are talking about) and your lower end parts have to include a second chip which must raise costs. NVIDIA had a full range of AGP8X complaint parts on the market when there was a transistion to AGP8X, it seems a little odd the are building a bridge for PCI-Express.

 

[KimB]

What we may be seeing is a renewed focus on their core technologies. They don't want to be blindsided by ATI again, and it was probably a question of priorities. After all, the only reason to really go fully PCI express right now is for cost reasons. The performance difference will be nonexistant.

 

[Rockster]

Nvidia Sucks!

 

[Tim Murray]

are ATI going the native PCI Express route, or are they going to use a bridge chip as well? I've heard about the supposedly PCI Express-using R423, but I don't really know what that implies in terms of native or bridged.

And considering the lowest-end products are PCI (for both major companies), wouldn't the lower-end chips probably be AGP? I can't see HP and Dell jumping on the PCI Express bandwagon for their $800 machines, or at least not yet.

Last, would the extra bandwidth provided by PCI Express actually benefit the next-gen cards from ATI and NV, or is it just another marketing gimmick?

 

[Rockster]

Dave, are you implying that chips currently in development at NVidia won't natively support PCI Express?

 

[KimB]

The extra bandwidth never benefits video cards from the start. I don't see why that would change now.

 

[John Reynolds]

The extra bandwidth never benefits video cards from the start. I don't see why that would change now.

The increased bandwidth would certainly benefit their ability to write back to main system RAM.

 

[Bouncing Zabaglione Bros.]

IIRC, ATI have said that they will not be bridging, and will be using a native implemention, probably on RR423. If Nvidia bridge, then they will be blindsided by ATI again. As soon as these benchmarks arrive that test the new interface, Nvidia's numbers are bound to be far, far lower than ATI's, which will once again put ATI in the lead, and Nvidia following on behind, in performance, technology, and marketing.

 

[Dave Baumann]

What we may be seeing is a renewed focus on their core technologies. They don't want to be blindsided by ATI again, and it was probably a question of priorities. After all, the only reason to really go fully PCI express right now is for cost reasons. The performance difference will be nonexistant.

Actually, I've heard performance differences up to about 10% in current apps with PCI-Express, from numerous different vendors and parties. Proof is in the pudding, but I wouldn't be surprised to see a slightly greater difference with PCI-Express than previous revisions of AGP - largely because the communication is handled differently between AGP and PCI-Express. Another issue is that if you add a bridge, you also add another layer of latency.

are ATI going the native PCI Express route, or are they going to use a bridge chip as well? I've heard about the supposedly PCI Express-using R423, but I don't really know what that implies in terms of native or bridged.

ATI have already demonstrated an "9600" style product with native PCI-Express, which would probably give you a fair indication that they already have a solution for the low end.

And considering the lowest-end products are PCI (for both major companies), wouldn't the lower-end chips probably be AGP? I can't see HP and Dell jumping on the PCI Express bandwagon for their $800 machines, or at least not yet.

They'll be a transition period, but from the larger OEM's you'll probably see a range of SKU's offering PCI-Express when Intel flicks the switch. You might find that PCI-Express may be cheaper for manufacturers in some respects (at a guess).

Dave, are you implying that chips currently in development at NVidia won't natively support PCI Express?

I don't know exactly what NVIDIA are doing wrt PCI-Express and their roadmap, but you can certainly see that a bridge is the quickest route to ensure your entire range of products are compatible with the bus.

 

[Arun]

*cough* NV36-X with PCI-Express Bridge *cough*
Whether the NV40/NV41/NV42 use this bridge, I don't know, though.


Uttar

 

[Reverend]

Another issue is that if you add a bridge, you also add another layer of latency.

May add another layer of latency.

 

[cthellis42]

Proof is in the pudding,

The proof of the pudding is in the eating.

At any rate I'm not sure why nVidia's bridging it either. It would seem to be a relatively painless way of getting a good speed boost (not to mention a marketing point), so letting it slide for a card release or two while ATi will be taking advantage of it seems strange.

Granted most people won't be upgrading to motherboards that can use it immediately, but it will certainly be tested by the reviewing community, and the high-end gaming system OEM's will certainly make that step as quickly as it's available, and going with whatever makes them fastest and most furious.

 

[jb]

May add another layer of latency.

Its not possible to add in any thing into your path with out it effecting some of your timings through that path. However your right as we don't know what the impact of going through the bridge chip will be... heck they could even speed it up some how

 

[KimB]

The extra bandwidth never benefits video cards from the start. I don't see why that would change now.

The increased bandwidth would certainly benefit their ability to write back to main system RAM.

Which would be less of an issue for current games, and more of an issue for the professional market.

 

[KimB]

Actually, I've heard performance differences up to about 10% in current apps with PCI-Express, from numerous different vendors and parties. Proof is in the pudding, but I wouldn't be surprised to see a slightly greater difference with PCI-Express than previous revisions of AGP - largely because the communication is handled differently between AGP and PCI-Express. Another issue is that if you add a bridge, you also add another layer of latency.

I would typically expect a slight performance drop with the first batch of native PCI-Express boards. It will take a bit for the hardware designers and driver writers for motherboards and graphics boards to get it right.

 

[KimB]

Another issue is that if you add a bridge, you also add another layer of latency.

May add another layer of latency.

No, it will add another layer of latency. The question is whether or not it will be important. I don't see why the added latency should be important for a graphics board, since the graphics board largely just receives commands. Texturing from main memory may be slowed, but that is slow anyway.

 

[YeuEmMaiMai]

I see it as another example of Nvidia got caught with their pants down....


It looks like they are late to the next tech party and people with cutting edge machines will most likely go for a solution that features natice support.

Nvidia better hope that NV40 is at least equal to R400 performance or they are in the toilet for sure......

it would suck for them if NV40 did not have native support and it was slower than R400

What we may be seeing is a renewed focus on their core technologies. They don't want to be blindsided by ATI again, and it was probably a question of priorities. After all, the only reason to really go fully PCI express right now is for cost reasons. The performance difference will be nonexistant.

 

[Hanners]

"To create a sixteen lane PCI Express bridge chip that works without performance degradation is a huge technological advancement," said Dan Vivoli, executive vice president of marketing at NVIDIA. "The use of a PCI Express bridge chip gives NVIDIA a decisive competitive advantage in what is the most exciting graphics bus transition in recent history."

Is it just me, or does crowing about migrating to a faster bus but not losing performance sound weird?

 

[kyleb]

actually it sounds about normal coming from nvidia pr.

 

[Vince]

Does anyone know the logic differential between implimenting AGP & PCI-Express? Just a rough estimate as I've been unable to find this.