Parhelia needs .13 micron

[[Ch]amsalot]

Hi everyone, this is my first post to your fine forum.

I'm extremely pleased to see Matrox introducing a power 3D card after several years of inactivity. This is a very exciting time for the 3D gaming segment, as Matrox, along with Creative Labs, ATI, and NVIDIA are all gearing up for new releases.

The screenshots that have been posted here and elsewhere on the Internet clearly show that the Parhelia has exceeded the current competition in terms of visual quality, but my question to you is, at what cost?

My question should be taken literally. At $399 for a 128MB card, Matrox is effectively keeping the Parhelia out of the hands of the masses and even out of the hands of most die-hard gamers. Assuming $299 for a 64MB version (which will certainly take a great performance hit at high-resolution games with many effects), the price is not acceptable. Fine image quality and acceptable framerates can be had with sub-$200 GeForce4 Ti 4200 cards.

In a stagnant market, the Parhelia would do very well. But we all know that the video card market is extremely fast paced. How soon will NVIDIA and ATI adopt their own form of 16xFAA? How soon will these 3D card giants introduce parts that are twice as fast as the Parhelia with even more advanced feature sets (e.g., fully DX9 compliant pixel shaders)? How soon will your $399 investment be rendered obsolete?

In my NSHO, Matrox needs to move Parhelia immediately to a .13 micron process, both to increase clock speeds as well as to drastically reduce cost. If not, Parhelia may be only slightly more successful than another extremely high-end (and very high-priced) board that never saw the light of day --- 3dfx's Voodoo5 6000.

[Ch]amsalot
Maximum PC Magazine Commport

http://forums.delphiforums.com/maxcommport/start

[TdB]

well, I don´t expect parhelia to hold the performance crown for a long time, but right now, it seems to be competetive. however i don´t think a card is rendered obsolete just because there is something better out there(there always is ), it all depends how long YOU will be happy about the card, and how long it will be usefull to you, personally i expect parhelia to keep me satisfied in 1,5-2 years, as most high-end stuff can do that.

and I agree that the 128mb parhelia is aimed at a relative small market(as most high-end stuff is), however i think matrox will get alot of OEM-deals with the 64mb parhelia to keep them afloat.

regarding .13micron, I think matrox is working on that already, and since the .15micron parhelia most likely will be clock-limited, i think the .13micron will be a great performer.
the parhelia has a lot of potential.

[Beezer]

Re: Parhelia needs .13 micron

Originally posted by [Ch]amsalot
Hi everyone, this is my first post to your fine forum.

In my NSHO, Matrox needs to move Parhelia immediately to a .13 micron process, both to increase clock speeds as well as to drastically reduce cost.
[Ch]amsalot
Maximum PC Magazine Commport
http://forums.delphiforums.com/maxcommport/start

I'm curious how moving to .13 micron would reduce cost? Getting that narrow would require at minimum a deep UV process or even ebeam. This past year my company implemented a DUV process (we make disk drive heads). For just two coat/expose/develop tools we spent $20 million. The room to put them in was another $5 million. I doubt the company that makes the chips for Matrox would be "lowering" the price.

[Wombat]

In my NSHO, Matrox needs to move Parhelia immediately to a .13 micron process, both to increase clock speeds as well as to drastically reduce cost.

God, we've gone over this soo many times already. Why do people think that smaller processes are automatically faster & cheaper? Hell, why do so many people think that 130nm is even ready?

You need to to educate yourself a lot more before you're qualified to have a <I>N</I>SHO.

If you think the price of Parhelia is "unacceptable," then don't buy one.

[[Ch]amsalot]

Smaller process gives more processor per wafer, thereby increasing yield while keeping the costs of raw materials the same.

Matrox shouldn't be bearing the cost of upgrading whatever company is currently making their gpus. Rather, Matrox needs to start getting involved with the same foundry that NVIDIA intends to use, or is currently using for NV30. I don't know if that's still TSMC or not, since NVIDIA was not happy with the initial output of their .13 micron dies.

-[Ch]amsalot

[Wombat]

Smaller process gives more processor per wafer, thereby increasing yield while keeping the costs of raw materials the same.

You're forgetting about things like the <I>huge</I> investment to upgrade equipment, plus since Intel is just about the only company to have 130nm down, More die x lower percentage of good die = no gains.

Matrox shouldn't be bearing the cost of upgrading whatever company is currently making their gpus.

Why not? What company doesn't pass its costs of operation on to its customers?

Also, my Parhelia will be on my doorstep within a month. NV30? Riiiight.

[[Ch]amsalot]

Wombat,

Your 'tude is neither necessary nor appreciated. Are you this antagonistic to all newcomers to your forum or am I just special?

All other factors being equal, smaller process means less resistance, less heat, and therefore more stability at faster clock rates.

The question is not whether .13 micron is ready, but rather, will Matrox be ready to make the switch when the technology becomes available. More important, will Matrox be willing to bring the cost of the card to levels where it can compete with high-end offerings from NVIDA and ATI?

-[Ch]amsalot

[[Ch]amsalot]

Wombat asked,

"What company doesn't pass its costs of operation on to its customers?"

My point is that Matrox should not bear this cost alone, but rather use the same foundry that NVIDIA, ATI, Transmeta and others use. That way the cost of the technology is shared by many, not just by Matrox alone.

-[Ch]amsalot

[Wombat]

Are you this antagonistic to all newcomers to your forum or am I just special?

This "'tude" only goes out to people that come here and post their uninformed opinions when the same issue has already been discussed in-depth five times already. If you look around, you'll find that there are some people here that really know their shit. ALPBM and Oboy work in Intel's fab, and help make sure that they have 130nm processes running, which nobody else in the world has. I work in a processor design lab (was on McKinley, now onto newer things), and can tell you that it's a non-trivial amount of work to switch from one process to another. You come in here, and like those who fell before you, proclaim that 130nm is undoubtably faster and cheaper than 150nm, with no facts, examples, or other ability to back up your statement beyond the drivel found in video-gamer magazines. At least go through the other parts of this message board where we've gone over the merits of this debate multiple times.

[[Ch]amsalot]

Wombat,

You're right... I should support what I say with facts. As an example of the lower costs of .13 micron process, I'll use AMD's Athlon XP as an example.

When the Athlon XP 1800+ (.18 micron) was introduced, it sold for $252 (1000 units lot). The newer and faster Athlon 2200+ (.13 micron) starting selling point was $241.

Presumably, AMD has shared in the costs of upgrading the fab, as is somehow reflected by the [lower] price of the processor. Hmmm... perhaps inflation works backwards in AMD's universe?

As for speed, you could argue that the 2200+ is tantamount to an overclocked Palomino, and I'd be inclined to agree with you at this point. Let's reserve judgment until we see performance figures for the lower voltage T-breds. I stand by what I say about smaller die size = faster possible speeds.

Your system specs sound so familiar...
--------------------
Abit KG-7 Raid w/ T-bird 1.4@12*133 with AX-7, 512MB Mushkin CAS222 PC2100, Leadtek GF4 Ti 4400, Philips Accoustic Edge, Pioneer 16xDVD (unplugged), Plextor 40XMax, Plextor 12/10/32S, WD 80GB 7,200 (2MB), Sony GDM-FW900 + Samsung 170T (DVI), Win 2k (gotta avoid crashing every hour).

[GT98]

Originally posted by [Ch]amsalot

When the Athlon XP 1800+ (.15 micron) was introduced, it sold for $252 (1000 units lot). The newer and faster Athlon 2200+ (.13 micron) starting selling point was $241.

I dont think the whole $10 saved by switching to .13 micron..theres more here at work then shrinking the die size down to a smaller one to make it cheaper..

[[Ch]amsalot]

I would agree with you, however, Wombat requested a specific example of a processor that was successfully moved from a .15 micron process to a .13 micron process; that was faster and less expensive. The AMD XP 2200+ just happened to be a handy example [Edit: woops, the Palomino was .18 micron]. Another example is GeForce2 Ultra to GeForce3. NVIDIA was able to get a much faster part out to consumers for basically the same or lower price, thanks to switching from .18 to .15.

-[Ch]amsalot

[Oboy]

First of all welcome to the forum [Ch]amsalot.

OK going from .15 to .13 will not save money! When you go down to .13 your die yield goes down, and lets think about this, going form .15 to .13, the die will not be downsized by that much, where only talking about a .02 difference in line width. That’s Fen small, a human hair has a diameter of about 70 microns. Switching to Cu and not using ALwas a real performance gain. Where you will save money is going from 200mm to 300mm wafers. That’s a BIG savings! The amount of die per processed wafer goes WAY up, so you have more die to sell. But die yield is still a big concern.

And since we are talking about this does anyone have a link to what the DX9 standard is? I have looked and have only found ppl says GPU's will need to use .13 process and nothing actually stating that.

Oboy

[[Ch]amsalot]

Switching to larger wafers is always a good idea Die yield goes down initially but as the process is fine tuned you get more good chips per wafer.

DirectX 9. I can't seem to find too much on it, but I'll continue to search:

DirectX 9 press release:

http://www.microsoft.com/presspass/press/2002/May02/05-30DirectX90PR.asp

Tom's Hardware:

http://www.tomshardware.com/business/01q3/010719/

Here's some more specifics, but it isn't in english:

http://www.3dcenter.de/artikel/2002/03-25.php

-[Ch]ams

[[Ch]amsalot]

BTW Oboy,

Thanks for the welcome.

I noticed that you have not disputed that, all things being equal, going to .15 micron process to a .13 micron process will result in higher maximum speeds.

By any chance to you have a background in electrical engineering?

-[Ch]amsalot