nvidia just can't stop

I know, less than 5 minutes after I posted that I got no less than 3 PMs informing me of just how incorrect that statement-o-mine was!

Sometimes I just hate hanging here, it's one of the few places left on the net that can just make me feel stupid as hell all the time.

 

Then maybe you should wait a bit and try to confirm your story before you post

 

Now where would the fun be in that?

 

Except why would the PCI Express chip have more silicon there? Regardless, performance tests should make it clear in the end. Hopefully we'll get some nice bandwidth tests early-on.

 

IIRC agp controllers are not on the die as of current, so if they decided to make a native pci-e and have it on the die, it would have more silicon

 

Memory makers have dual channel kits, and I'll bet people will wish motherboard makers start putting out dual motherboard kits for their LGA775 boards. That way, when you bend one of those tiny tiny contacts in the socket with an errant breath or a stray air current from the vents in your house/apartmen, you'll have a backup.

LGA775 is just another reason to go AMD.

 

Um. That would mean there'd be another chip between the GPU and the interface on every graphics board. Last I checked, there isn't.

 

Since performance is the only way to judge if the first generation of ANYONE's PCI-Express solution will matter (at least for gaming) I'm not sure what all the hubbub is about. The FUD and company sniping certainly gets tiring... :?

 

Can the DMCA be applied against companies for "snooping"?

 

That's the impression I got. Good thing I know about this site. I thought there was an actual 'scandal' going on after the news I've been reading.

 

Probably, but only if they've broken any kind of encryption mechanism, which doesn't seem to factor in this case.

I do think it's pretty bizarre that Nvidia are x-raying competitors chips, and then posting those pictures to the web with their own spin. Nvidia should really be talking up their own products, instead of trying to do other people's products down - that just smacks of fear and desperation.

I wonder if ATI can take any kind of action in the form or breach of copyright/design theft/damage to business, especially if Nvidia is wrong about some kind of internal bridge chip?

 

Didn't ATI say a bridge adds latency? That would apply to a PCI-E to AGP bridge too.

Doesn't everyone check their competitor's products out?

 

That probably would still be an issue with a bridge that way around, however everything else isn't so much of an issue as effectively PCI Express is a superset of AGP. I've heard talk that ATI may only offer its higher end SKU's of each new part as PCI Express only, in an effort to boost the platform - although it may be a little early for their plans to have been finalised.

Pretty much, yes. But I'm not sure that they are necessarly able to to determine exactly what the cells do (beyond Idenfying logic and cache areas) and I've not actually seen any documents from any others using those images as proof of one thing or another in what I suppose are "sales" documents.

However, I do wonder if there was some ring of truth to this at the start, but not necessarily now. I'd heard some talk that suggested that the initial RV380 variants were limited to 8 lanes and they may have done something quickly in order to get the test platform up and running. The initial ones seem early last year were A11 silicon - ATI usually numbers usually go A11, A12, A13 etc for minor revisions; the RV380 that is coming to the market is A21 indicating a major (metal change, I believe) revision. Given these X-Rayed images started surfacing some time ago it may be the case that these images are from the initial A11 version, but not the A21 version going into production and the A21 may have further alterations than their test platform than just the move from FSG to low-k as ATI state.

 

I really don't see why latency would be a big deal for a graphics card. There's not much that the graphics card has to tell the rest of the system to operate. It pretty much just accepts commands. As long as it gets them at a high rate, it really doesn't matter if they're a few microseconds later.

 

Very true. Latency is practically irrealitive considering the commands are queued in a ringbuffer which is DMA'd accross at the GPU's leisure. So the more commands, the larger the buffer, and the larger the latency. Doing PIO whould probably incur more latency than any bridge chip could produce.

 

I just love how many places turn to Nvidia when they have questions about ATI's architecture. Oddly, I don't see the same places turn to ATI for questions about Nvidia's architecture. It just doesn't make sense to me, why turn to a company that misrepresented their own card's architecture (eight-pipe GfFXs anyone?) when you have questions about their competitor.

As to ATI's future bridge plans-- accepting a performance penalty from bridging to allow backwards compatibility for older systems is totally different than using a bridge on your highest end products. By the time that happens people won't be expecting ultimate high-end performance from AGP, that's what PCI-E will be for. Those products will be expected to be less than the best, so people will take the performance penalty for the ability to upgrade without moving to a whole new system. It's throwing a bridge chip in the highest end PCI-E cards that will lead to problems, as if AGP were as good as PCI-E there woud have been no need for anyone to ever shift to it.

A native implementation is always going to be better, regardless of how any bridging is done, but there are some cases where bridging is more defensible than others.

 

Interesting

 

I get similar results when using TexBench (a few hundred MB/sec more than an AGP card), but I didn't include them in my review of the platform since the results were slightly erratic.

Interesting indeed.

Rys

 

I'll take a look at the performance of both on the PCIe X800 XT.

 

just 3d?