A few weeks ago I asked Loyd Case, who is Product Marketing Manager at NVIDIA, if he would be willing to do a question and answer session with us. Loyd said he would be glad to and thought it would be fun. I asked visitors to nV News to send in questions they wanted to ask Loyd and we ended up with 27 questions. It took a few days for Loyd to get back to us, but he ended up answering every question.
I'm glad to have had the opportunity to meet Loyd earlier this year during a trip to NVIDIA. He knows his stuff, is an avid gamer with a killer rig, and best of all is that he might be older than I am!
Let's get started...
nV News: Your official title at NVIDIA is Product Marketing Manager. What is your role at NVIDIA?
Loyd: I like to tell people I fit in the white spaces on the org chart between product marketing, developer relations, the performance lab and PR. I've written white papers, done presentations, helped out with benchmarking, and will also be doing competitive analysis. I'll have the opportunity to do some web stuff, too.
nV News: The early benchmarks from the KYRO chip are impressive considering that the clock and memory speeds are running at 125MHz. Does NVIDIA anticipate moving to tile based rendering in the future or a continued "brute force" approach with a few new tricks?
Loyd: Tiled architectures are one way to reduce memory bandwidth usage. The strategy that tiled architectures employ is to be very efficient at "not drawing". In other words, if there are a lot of pixels in a game or other application that are not drawn, tiled architectures can discard them very quickly. Although this is somewhat useful, it's not really the hard problem to be solving. Drawing high quality shaded and blended pixels is more important that "not drawing" them. Also, accelerating more geometry (better shapes and characters) is a lot more important. In the future, Nvidia may pursue a tiled architecture, when it makes sense. At this point in time, it doesn't seem to make sense.
nV News: How exactly will NVIDIA improve memory bandwidth? Current DDR solutions
are not providing enough effective bandwidth to support high resolution rendering that gamers and developers want so much.
Loyd: Short term, memories will get faster. Longer term, there are definitely some things in the works that I can't really talk about. Faster RAM is a great way to improve throughput, and certainly more elegant than doubling the number of chips just to do the same workload. NVIDIA works with all the major memory manufacturers to ensure that the best memory solutions will be implemented in future NVIDIA products.
nV News: What about T&L performance? Will we be seeing geometry compression coming soon? If so, what obstacles do you see stopping you?
Loyd: I'm not quite clear on what is meant here by geometry compression.
While T&L support can be bolted on to existing game engines, I'm personally looking forward to titles that are designed from the outset with hardware T&L in mind. Already, Evolva, which has been well received by many gamers, does much better on GeForce2 GTS than anything else. There was a lot of stuff announced at E3 that will make serious use of the T&L engine, like Sacrifice, Black and White and Colin McCrae2. Hardware T&L, as done by the GPU, allows art to become one of the defining factor in game design.
nV News: What enhancements were made to the T&L engine of the GeForce2 besides the obvious speed-up?
Loyd: The circuit paths through the T&L engine were tweaked and streamlined. The result is better sustained performance.
nV News: Do you have a practice putting green in your office or are you stuck with cubicles?
Loyd: I don't play golf, but I haven't seen one here - I think they had to remove it to make room for new people as NVIDIA continues to expand and grow. A lot of the marketing guys (including the VP of Marketing, Dan Vivoli, Derek Perez and others) play hockey, though. Me, I stay away from anything that involves full body contact. Put me on a pair of alpine skis, however…
nV News: How important a feature do you believe FSAA is?
Loyd: Anti-aliasing is far too controversial these days and certainly doesn't deserve all of the discussion. All current implementations (ours and others) are extremely simple, basic implementations. In almost all cases, users end up preferring to run in high resolution versus anti-aliased since the performance impact is the same. Even so, GeForce2 MX and the GeForce2 GTS supports full-scene, order-independent, high performance hardware anti-aliasing.
I personally think that anti-aliasing is most useful for older titles that can't run at higher resolutions. I don't find myself playing those too often these days, though.
nV News: Is fill rate still king?
Loyd: No. What's king is balanced solutions that offer users current value today, but gives developers the ability to build the cool games of the future we crave. Fill rate is important, because it does affect so many features. The insanely high fill rate capabilities of the GeForce2 allowed us to really push forward per-pixel shading, for example. But with GeForce2 GTS, we're moving beyond fill rate.
nV News: Word's out that you have a kick-ass gaming machine at home. What hardware are you using? From your hardware geek perspective, and not an NVIVDIA marketing manager perspective, what balance of resolution, texture quality, polygonal detail, and FSAA you think is best, and why.
Loyd: Okay, hardware geek hat on.
What makes you think I have ONE system? I mean, come on… ;-)
In my basement lab, I have anywhere from 4 - 7 PCs running, ranging from a PIII/600 up through an Athlon 850 and PIII/866. All of them have either GeForce DDR or GeForce2 boards. All of them have at least 128MB of RAM, and they're all networked together. A bunch of us played Baldur's Gate all the way through in the same room, and I must say, it was one of the best gaming experiences I've ever had -- much more social than most action games, but much faster paced than pen-and-paper.
I did have DSL for Internet connectivity, but dropped it (I had a business LAN package) for cable service now that I'm not running a business out of my basement. Also, now that I no longer have a real business reason, it remains to be seen whether or not I can maintain that many systems. Note that two of them are actually used by my two daughters for their stuff.
Until recently, my production system (the one I actually wrote all those articles on) was less than bleeding edge -- a PIII/600 with GeForce DDR on a 440BX motherboard. It had lots and lots of disk space, though.
nV News: What games are you playing now? What are a few of your favorites?
Loyd: Right now, I'm slowly playing through Ground Control and Vampire: Redemption. That may all fall by the wayside when Icewind Dale ships, though, as my local net gaming group is yearning for a good dungeon crawl. I used to play a lot of flight sims, but not so much these days. I am waiting for B17-II and Crimson Skies, though. I also just picked up Shogun, but haven't really started it yet.
nV News: Does NVIDIA have a student co-op program? If so, how many students do you take a year?
Loyd: Yes, there's an internship program. This summer we have about 24 students on board working in different capacities throughout the company. Most are here in Santa Clara but a few are in our North Carolina offices.
nV News: What does NVIDIA see as the next large evolution in 3D cards beyond T&L, FSAA and higher order surfaces?
Loyd: With regards to the earlier question about bandwidth, though, there's no doubt that higher order surfaces will help in that arena. The trick is to make sure all those cool new features get implemented. Our developer relations group is the best I've seen in this business. An example of this is what we saw on the show floor at E3, with developers showing upcoming game titles on the technology NVIDIA has been evangelizing.
nV News: How is the load balancing with the GeForce coming? Meaning we see certain cases were the CPU can beat the GPU. Does NVIDIA have any plans to work it out so both are maximized at the same time?
Loyd: There are no game benchmarks that adequately stress the T&L engine to date. The GeForce2 GTS T&L engine, in terms of lit, filtered triangles, can still outperform a 1GHz PIII by a substantial margin. About the best T&L test we've seen is 3D Winbench 2000, and we consistently beat software T&L there on the fastest CPUs available to us. In cases where you may see a CPU supposedly beat GeForce T&L, it's almost never an apples-to-apples comparison. Everyone who uses a pro app like 3D Studio or Maya will tell you that the GeForce -- or better yet, Quadro -- can beat any CPU in preview rendering.
That said, there may be times when load balancing would be useful, but I don't know what work, if any, is going on in that arena.
Loyd Case Q&A: Page Two