2002: A Year in Review December 25, 2002 Jakub Wojnarowicz

Overview

Page:: ( 1 / 7 )

Looking back

Well, with 2002 over and done with, we’ve got quite a few things to look back on, don’t we? This has been a year with a few major surprises, and many expected events.

ATI

Did anyone see this coming? For the first time recent memory, ATI delivered a killer product without a peep of marketing hype preceding it. One day we had the superb GeForce 4 standing on a completely different level from everything else in the market, and then it was crushed by ATI’s Radeon 9700 line. At the time, with GeForce “5” only a few months away, this was shocking enough. 9700 was on the market several months early and we thought it was a coup for ATI just because it gave them some time to build up momentum before the launch of the next NVIDIA part. The situation was very analogous to the launch of the Athlon by AMD, which caught Intel with its pants down as the Athlon leaped past the beleaguered Pentium III, and the P4 wouldn’t be ready nearly in time.

What seemed a finesse move by ATI in getting to market early, has become a crushing blow to NVIDIA by their failure to produce a contender at all. Let’s get into a few background details first. ATI designed the 9700/9500 chips to be built with the .15 micron (150nm) design process – a tried and tested method. NVIDIA, doubtless reassured by TSMC that it would be ready, banked on the .13 micron process being available in 2H 02. The GeForce FX chip was ready a long time ago and would have been on the market sometime in the third quarter, if not for manufacturing delays.

So while technically the fault doesn’t like directly with NVIDIA for the delay, in reality there is some blame to disperse there. We wouldn’t be having this argument if ATI had made the same mistake and trusted TSMC’s claims about .13 micron being ready. Since both ATI and NVIDIA rely on outside manufacturing (namely: TSMC) to produce their chips, either the ATI engineers convinced everyone their chip could run on .15 micron, or ATI found some reason to doubt Taiwan Semiconductor’s claim. In short, that means someone at NVIDIA failed to do the same for their company.



SIDEBAR: I almost wrote ‘the upshot of all this’ as the link to the next page, but then I realized I how odd ‘upshot’ sounds.

Consequences for NVIDIA

Page:: ( 2 / 7 )

Ultra Titanium

As we all know, NVIDIA loves to stick by the 6 month product cycle. Typically they release a chip, then 6 months later a faster version of the chip, and another 6 months later a completely new chip. The internal codenames for the chips are predictable like clockwork. NV10, NV15, NV20, NV25, etc. The cards themselves are the GeForce 256, GeForce2/Ultra, or GeForce3/Ti and GeForce4.

In the codename scheme, any of the ones with numbers at the end divisible by 10 imply a completely new chip, or at least a great departure from the original design. They offer new features and dramatically improved performance. The codenames ending in 5 are tweaked versions of the original, with faster memory and a faster core. The chip might have a few changes in silicon to improve clock speed (NV10 -> NV15) but nothing like the introduction of pixel shaders (NV15 -> NV20).

So, what does this have to do with anything?

Let’s think about it for a second. If NV30 (GeForce FX) arrives in numbers at the end of January or beginning of February, NVIDIA has basically missed a whole product cycle. It will have given ATI not just two months, but almost half a year to establish itself – that’s bad enough.

What’s worse is that NV30 will have missed its window of opportunity. Doubtless, the engineers at NVIDIA have been hard at work at NV35 for a long time and it would be ready for production in Q1 03. Unfortunately for NVIDIA, the manufacturing delays at TSMC have put NV30 into the same quarter. When GeForce FX hits stores, it will be comparable to the Radeon 9700 Pro, more or less. NV35 (GeForce FX ‘Ultra’, or ‘Titanium’, or whatever NVIDIA’s marketing decides to dub it) would have been the card with a significant speed advantage.

It’s perfectly possible for NVIDIA to release GeForce FX for merely two or three months and then follow it up with NV35. That will be a financial hit as far as they’re concerned, since they wouldn’t be able to extract maximum value out of the GeForce FX. Indeed, unless it offers a significant performance or feature advantage over the Radeon 9700 Pro, they won’t even be able to price it at a premium since ATI will surely drop the price of their cards. ATI has already been getting its money’s worth from the 9700 cards and by then the probably would have already implemented price cuts to pave the way for R350.



SIDEBAR: It’s pretty crazy when games are looking way better than the 3DMark demos.

The future and consequences for ATI

Page:: ( 3 / 7 )

R350

R350 is to ATI what NV35 is to NVIDIA – a faster version of the basic (R300) chip. At least, so the rumor mill holds. If those same rumors hold up, you can expect it to be called the Radeon 9900, since ATI is is expected to reserve the lofty ‘ten thousand’ range for their next-generation products. Just a few days ago, the Inquirer reported that ATI and TSMC are working together to make the RV350 on a 130nm process. For the confused, R350 is the big brother, RV350 is the younger sibling, like 9700/9500.

What’s really interesting about R350 is that it will continue to be manufactured with a .15 micron process, meaning that ATI will remain sheltered from any further fallout from the .13 debacle, should it occur. This does translate into slightly higher production costs based on the price of silicon, but that may be less expensive if yields remain better than at .13 micron. Of course, if R350 ramps up the clock speed too much, it may have yield problems of its own due to heat issues.

ATI’s status

ATI’s recently announced fiscal Q1 03 results were very encouraging. Revenue was up 34.5%, or almost $81 million and profits were $5.0 million, compared to a loss of $32.2 million the quarter before. Numbers also improved compared to the year before, ATI having reversed a trend towards ever-greater setbacks.

Given the current economic conditions and the brief time that its 9700 cards were on the market, this is very encouraging news for ATI. Their situation will continue to improve, at least until GeForce FX hits the market. After that, ATI will stabilize at a comfortable level, unless GeForce FX does something unexpected.

Unexpected? That’s right, there are some questions about GF FX’s performance. NVIDIA claims it is 30% faster than the Radeon 9700 Pro, although theoretically the 9700 Pro has a higher real-world fillrate since it has more memory bandwidth. Most people have taken NVIDIA’s claims with a grain of salt, calling it a bit of FUD to get the masses to wait before buying a 9700 card. Of course, if NVIDIA delivers what it claims – that would be the unexpected kink. ATI would recede on the market a bit and stabilize at a lower level than otherwise expected.



SIDEBAR: Nobody counts theoretical fillrate anymore. It used to be that manufacturers advertised their chip’s performance based on the theoretical maximum fillrate that the chip could deliver. Since memory was never able to keep up, the whole thing got bad press and was exposed around the Voodoo 3 / TNT2 wars.

The future

Page:: ( 4 / 7 )

Questions, lots of questions

ATI’s foreseeable future is going to be very dynamic, and they need to keep gaining momentum. We expect a very aggressive R350 launch, though not necessarily in a heavy-handed marketing way. ATI has experienced success by keeping mum and may try the trick again, just letting R350 shock people for a second time.

NVIDIA’s situation is very uncertain, but not in a bad way. The ball is in their court and as long as TSMC can deliver, so can NVIDIA. Taiwan Semiconductor aside, the biggest question at NVIDIA will be what to do with NV35. They could launch it early (relative to NV30’s launch, which would put it on time with the original 6 month schedule), but we wouldn’t be surprised if ATI hit right back with R350. 9700 already showed that ATI can beat NVIDIA at the 6 month schedule, since the card was launched before NV30 was to hit the market. R350 being ready to counter NV35 is guaranteed.

If ATI launches R350 on schedule, that will of course cut down NV30’s shelf life and we’ll see NV35 right after that, and we’ll be back on the 6 month schedule. How both chipmakers deal with any future manufacturing problems at TSMC is up in the air.

In Taiwan

There is one wildcard in the whole works, and that is Taiwan Semiconductor. Being the largest independent foundry in the business (UMC is a distant second), it has considerable leverage with both ATI and NVIDIA who need TSMC to manufacture their chips. That means it’s doubtful that NVIDIA would pursue legal action against TSMC for the .13 micron delays. Still, it may be that NVIDIA might wring some concessions out of the company, perhaps a discount on future chips or a lump-sum damage payment. This is all speculation however, and we expect any such deal would be kept quiet.

Any further delays at TSMC can put a damper on the whole market. Sooner or later both companies will move completely to .13, then .09 micron processes and if production delays occur then the pace of graphics card development will slow considerably. CPUs have generally used more advanced manufacturing techniques, but graphics chips are rapidly catching up and if their complexity continues to grow (they already have vastly higher transistor counts than the most complex CPU), they’ll need better manufacturing much sooner than a CPU would.



SIDEBAR: Speculation is so much fun when you’re playing with other people’s money.

Intel

Page:: ( 5 / 7 )

The tide turns

In what’s becoming a see-saw battle, Intel took the top spot from AMD again with the release of the 3.06GHz processor. In general, the tide has been turning, as expected, Intel’s way for quite some time. Just as the Athlon outpaced the poor old Coppermine with ease, so has the Pentium 4 ran circles around the Athlon. Slowly but surely, the P4’s ease of scaling led it into spheres of performance outside the Athlon’s reach. Yields have been consistently, even surprisingly good on the Pentium 4 while AMD has had to make a few paper launches to keep up.

How good are Pentium 4 yields? Good enough that Intel had the time and confidence to include Hyper-Threading on the P4 3.06GHz even though original schedules had it appearing in 2H 2003 with the 90nm Prescott core. In our review, we also noted that Northwood still has a lot left in its legs, as the chip overclocked to 3.56GHz. That’s a 500MHz improvement, no mean feat for a 3GHz core that started its market life at 2GHz! Whether or not the review part was a golden sample is hard to tell, since so few people own one of these expensive beauties.

Hyper-Threading

The 3 gig part not only claimed the raw speed crown, but also introduced Hyper-Threading, the first major chip feature since SSE. It is also the first improvement aimed solely at improving desktop performance, a sign of just how badly hardware has outpaced games since 3dfx/ATI and NVIDIA, and AMD and Intel pulled out all the stops in their competition.

More importantly, for Intel it’s a way to grab customers’ attention with a new feature. Since the Athlon XP2800+ is available only in negligible quantities, Intel’s only competition at the top is with itself and the shrinking wallets of the market. With less money to spend, people need a good reason to buy a top-of-the-line processor and Intel hopes Hyper-Threading is it. As we’ve seen in the review, Hyper-Threading does work and provides very tangible results in certain situations. It’s not a true replacement for a proper multi-processor box, but with Athlon MP’s available only at the 2400+ speed rating and the Xeon being completely uneconomical, HT is the best choice we have. Besides, it’s ‘free’ performance and will be included on all future Intel chips.

Rambus

Rambus sealed its fate and that of RDRAM by disturbing the balance of power. Like Louis XIV’s France, it gained a position of power, struck out too greedily and too often, uniting its enemies and losing its allies. Defeated in court and creating a public relations disaster, Rambus was not a company with which Intel wanted to be associated with. Although the best Pentium 4 solutions still rely on RDRAM, they are still more expensive than DDR solutions which are almost as potent. Intel has itself finally released competent non-RDRAM chipsets with the 845PE and GE.



SIDEBAR: Even with Rambus seemingly down for the count and DDR-II on the horizon, don’t be surprised if Intel keeps RDRAM platforms around as an option. RDRAM is uniquely suited to the P4, as the P4 is suited to RDRAM.

AMD	Page:: ( 6 / 7 )

Years gone by

There is no doubt that without AMD, we wouldn’t be here. AMD is single-handedly responsible for lower prices, faster processors and better features on those processors. If it wasn’t for the K5 and K6, we wouldn’t have MMX, 3DNow! and SSE. Premium processors would still cost over $1000, and we suspect that they’d be a lot slower. In fact, a ‘budget’ computer might still mean a $2000 box.

It is with some concern then that we observe AMD’s quarterly reports and await the new Athlons. Nobody wants a return to the bad old days, and even Intel would admit that AMD has forced the introduction of better processors at better prices.

Yet, with continued delays and paper launches that resemble Intel’s problems at the end of the Pentium III, things look bleak for AMD. Thoroughbred, a core introduced this past summer, is already on its last legs and good luck finding an XP2800+ even listed on pricewatch. Barton has suffered numerous delays, slipping to the first quarter of 2003 and its silicon-on-insulator plans have been dropped. Rumors hold that the delay is caused by (surprise!) UMC’s 130nm production problems.

Paint it black

The year hasn’t been all black for AMD. The new speed rating scheme seems to have gone off without a hitch. Credit two reasons for this – smart marketing which has implemented realistic ratings that are about 100’MHz’ conservative, and a more educated buying public than the last time this was attempted - in the mid ‘90s - by Cyrix. The poor economy also hasn’t prevented system manufacturers from re-adopting the Athlon platform. In fact, AMD’s lower prices seem more attractive now than they did during boom times. Of course, Intel can afford to take the financial hit while AMD cannot.

Hammer time!

Of course, what everyone is holding their breath for is Hammer. ClawHammer will be the desktop, single CPU variant and now has the official name ‘Athlon 64’. SledgeHammer is the multiprocessing chip with the new ‘Opteron’ name. Both will be produced at AMD’s Fab 30 plant in Dresden and thus not subject to the trials and tribulations going on in Taiwan. That hasn’t prevented speculation from analysts that the 0.09 micron process chips (Athens, San Diego, and Odessa) will be delayed until late 2003.

So what is Hammer? AMD’s climb up the mountain with a pair of steak knives, one of which they’re trying to jab into Intel’s heart. Given the continued problems Intel is having with Itanium and the extreme skepticism that the server market views it with, AMD actually has a shot. Opteron won’t produce the kind of performance that Itanium is promising but it will be an economical option. More importantly for us, Athlon 64 will allow the use of more than 4GB of memory – a barrier which we are approaching at a rapid pace. It’s a safe bet that you know at least one person who already has 1GB of RAM.

The extra instructions per clock will be of far more value for Athlon 64 than the modest clock speed increase it will have over current Athlons. Using a 12-stage pipeline instead of the current 10-stage one, AMD is going to be able to pump out maybe an extra 10-20% MHz relative to the basic Athlon, depending on how the rest of the chip is configured. The on-chip memory controller will be a huge bonus, especially when paired with the on-chip dual-channel 64-bit DRAM controller on the Opteron. Athlon 64s will still use a single channel for their RAM. The biggest improvement will be the 8 extra 64-bit registers, which will double the register count on x86. Taking advantage of all this means at least recompiling code, so hopefully AMD has programmers at work which will make a Hammer compiler for the major languages.




SIDEBAR: Hey, what do you know? Maybe the marketing suits at HP and Gateway finally found the financial projection papers which say what happens to chip prices if only one company is around to make them. (note – I’d fully expect AMD to abuse its position as Intel had ten years ago.)

Games

Page:: ( 7 / 7 )

Looks so good!

We are finally seeing games that can take advantage of modern systems. I was reviewing games for close to two years on my Athlon 1GHz with a GeForce2 – a completely unheard of feat. It took until Unreal Tournament 2003 to provide a piece of software that highlighted the review system’s inadequacies (though Infogrames still foolishly chose a 1GHz processor with a GeForce2 as the ‘recommended system requirements’.) Asheron’s Call 2, a MMORPG of all things, was the latest game to make me thankful I now had a relatively modest Willamette P4 2GHz and Abit 4200 OTES card in my box.

What does this mean? Screw the idiot analysts predicting doom and gloom, we all know what drives the computer market – games! In the next year, expect software that will crush any system not built in the last 6 months (12 months if you used the best components.) Doom III alone will make mincemeat out of more computers than any game since… well… Quake.

Unlike last year, this year we have some serious software to look forward to. Just off the top of my head:

• Doom III


• SimCity 4


• Star Wars Galaxies


• Unreal 2


• Freelancer


• Homeworld 2


• Splinter Cell


• Raven Shield


• Deus Ex 2


• IGI 2


• Sam & Max 2


• Lionheart


• C&C: Generals


• Hidden & Dangerous II


• Il2 – Forgotten Battles


• Temple of Elemental Evil (Troika’s 3rd edition D&D game!)


• Galactic Civilizations

Uh… isn’t this an article about this year?

Looking for killer titles to play this year? Check out our Christmas Gift Games list, and throw in a few obvious titles that I missed while scouring my room for games to play – like NOLF2!

But seriously, this wasn’t the best year for games. That people are blaming consoles for it just shows the depth of their ignorance. Perhaps consoles can be blamed for taking too much focus off the PC (see what happened with Monkey Island 4? *shudder*), but not for poor publisher decisions. Besides, we’re still feeling the effect from Deer Hunter and ilk, which implied that’s the best PCs can come up with!



SIDEBAR: Merry Christmas? Why thanks, why don’t you offer me a comment and Sound Off!