Opinions needed on this Interview

[JoshMST]

Hey guys, just thought I would shoot this one over (since Dave never posts front page news to other site's articles- though I do link to him on occasion ) and see what you think.

http://www.penstarsys.com/editor/tech/graphics/nvidia/sli_inter/index.html

Chris Daniel took the time to answer some of my more obscure questions, as well as cover their thoughts on their solution vs. a supertiling architecture.

I thought it was interesting, but then again... I am biased to my own work (but not to NVIDIA or ATI mind!).

 

[Geo]

Wow, Josh, that was as sweet a hatchet job as I've ever seen. It would take the CSI boys to identify the victim. No fingerprints, no name, no id at all.

Until a citizen walked by, took one glance, and said "What, are you guys kidding me? It's ATI. Oy!"

Edit: Which, btw, is not to say that NV isn't entitled to their opinion, and that the community shouldn't hear what their expertize on this matter tells them, while the rest of us weigh in our own minds how much their self-interest is skewing that message --but, come on, own up!

 

[JoshMST]

Unfortunately, NVIDIA essentially answered the questions they wanted to. I tired to get them to talk about possible disadvantages, as well as previous products that supported AFR and how successful they were (ATI MAXX). They simply refused to answer any of those. So my choices were: publish the interview as is with NVIDIA not answering the questions they didn't want to, or not to publish the interview at all.

So, while this is a bit of a mouthpiece for NVIDIA, I thought I did get a few questions in there that did uncover a couple of unknown things about SLI.

 

[Geo]

Unfortunately, NVIDIA essentially answered the questions they wanted to. I tired to get them to talk about possible disadvantages, as well as previous products that supported AFR and how successful they were (ATI MAXX). They simply refused to answer any of those. So my choices were: publish the interview as is with NVIDIA not answering the questions they didn't want to, or not to publish the interview at all.

So, while this is a bit of a mouthpiece for NVIDIA, I thought I did get a few questions in there that did uncover a couple of unknown things about SLI.

Yeah, those kind of choices suck, but thanks for being honest about it. They *do* have a right to their opinion, they *do* have the expertize to give a useful one, and we should hear it and weigh it amongst other sources.

 

[JoshMST]

Well, if ATI gives me the chance, I would certainly like to ask many of those same questions of their engineers and product managers! I think a counterpoint interview to this one would certainly provide readers with a good idea of which company has their ducks in a row when it comes to multi-card configurations.

 

[DemoCoder]

Was MAXX really that bad? Wasn't its achille's heell the impossibility of it working under NT properly, and bad drivers?

 

[JoshMST]

Oh Demo... thats about the same as saying, "Was Jim Jones really that bad? I mean, wasn't his achilles heel that he was insane and could mix up a mean batch of kool-aid?"

 

[trinibwoy]

This was funny :

You have seen our cooling solutions evolve from the GeForce FX heat sink/fan (sorry) to the GeForce 6 heat/sink fan.

I thought it was a decent interview overall. Nobody expects a company to wilfully detail the shortcomings of their products in comparison to the competition.

The comparison of AFR/SFR and super-tiling was interesting - I had never really though about the possible disadvantages of supertiling (just assumed it would be better for some reason due to the inherent load balancing). Also, time will tell if his prediction of the PCIe bus choking with multiple GPU's in future titles is accurate. His claim of zero driver overhead for SFR load-balancing deserves a though.

A lot of it was also discussed here already. Their developer 'SLI education' campaign and their push to deliver more profiles in upcoming drivers etc etc.

 

[trinibwoy]

Well, if ATI gives me the chance, I would certainly like to ask many of those same questions of their engineers and product managers! I think a counterpoint interview to this one would certainly provide readers with a good idea of which company has their ducks in a row when it comes to multi-card configurations.

How do you go about approaching these IHV guys? Did you get a hook-up with Chris or did you just decide to shoot him an email one day and request an interview?

 

[JoshMST]

I went through NV PR first, and they set up the interview.

 

[Geo]

Well, if ATI gives me the chance, I would certainly like to ask many of those same questions of their engineers and product managers! I think a counterpoint interview to this one would certainly provide readers with a good idea of which company has their ducks in a row when it comes to multi-card configurations.

A statement like that would look good at the top of that interview in introduction. My $.02

Come to think of it, why don't you guys on these written interviews just put up the questions you asked anyway? With "No answer" as answer? If I'm watching a TV interview and Mr. Interviewee pulls an "old stone face" at least I know what he was asked and that he refused to answer.

You'd protect your own integrity better that way.

 

[JoshMST]

Yes, watching professional suicide from a distance is a very rewarding activity.

 

[Geo]

Yes, watching professional suicide from a distance is a very rewarding activity.

Wow. What a sick paradigm y'all act in.

 

[DemoCoder]

Oh Demo... thats about the same as saying, "Was Jim Jones really that bad? I mean, wasn't his achilles heel that he was insane and could mix up a mean batch of kool-aid?"

What, what I meant was, was MAXX a failure because of AFR, or because of bad implementation and NT problems? That is, just because MAXX failed doesn't mean AFR isn't good.

 

[JoshMST]

Ah, I see. No, AFR is not intrinsically bad, it is just that ATI implemented it poorly. From what I know, it was a 100% software controlled solution, and as such had some real issues in communicating. Perhaps the ATI guys could explain why it worked the way it did.

 

[Reverend]

I haven't read your interview yet Josh but did you state (in the interview) which questions NVIDIA chosed not to answer?

PR sucks huh? Yeah, and you can tell BB I said that!

 

[Reverend]

OK, just read the interview.

With Supertiling, each GPU is forced to do all of the geometry transform calculations, resulting in no geometry scaling.

I'm not sure if this will be a long term "problem".

Additionally, I wished you had asked him about possibility/probability of single board SLI offerings from NVIDIA in the future.

 

[rwolf]

Nice interview anyway.

 

[Jawed]

Why do I not believe this?:

What developers do you have onboard actively utilizing SLI features in their future titles? How are developers reacting to SLI?

...
By developing future titles on SLI technology, game developers are able to optimize their titles for SLI configurations. Additionally, the performance of SLI today gives developers a glimpse of how tomorrow's single GPU systems will perform. This allows them to tune for all elements of the system including CPU utilization.

Jawed

 

[Dave Baumann]

So my choices were: publish the interview as is with NVIDIA not answering the questions they didn't want to, or not to publish the interview at all.

I don't think there is anything wrong with publishing no comment answers as it least it shows it was on your mind.

From what I know, it was a 100% software controlled solution, and as such had some real issues in communicating.

Does that differ from NVIDIA's AFR solution (not the SLI interconnect, but the rendering scheme itself)?

Of course looking at the one Q&A that I did it seems like its presupposing what ATI will implement when there is a vacuum of knowledge. IMO (And I'll stress, IMO) ATI's implementation will take the course of least drver manipulation from ATI - i.e. the default implementation will work in the maximum of cases with the minimum of fuss, but that may not necessarily be the highest performance option and that may well be Tiling. However, certianly in a bench such as 3DMark05 Tiling will be of very little benefit, much like SFR is, and AFR is the best solution for producing the highest score - given this, and given the "capture the flag" mentaility does it make sense for ATI not to offer an AFR solution?

Should ATI opt to offer more than one solution then I can see a similar situation to NVIDIA with them offering user selection through the game profiling, but having fewer initial predefined titles. And, if this is the case then the comparison should be SFR against tiling.

Another one that may have been interesting to know about is where they feel AFR's limitations are in terms of number of devices.