Why NV30 beats 9700 Pro?

[Switch]

Linkage (http://www6.tomshardware.com/graphic/02q3/020819/radeon9700-03.html)
Anyone looked at this before? The 9700 Pro only handles one texture per pipe. Now even the Ti4600 does two. Should the NV30 follow along with Nvidia's 2 textures (or possibly more) per pipeline, wouldn't this give a pretty large increase in performance?

[-=DVS=-]

Lol you proof nothing , noone knows how much TMUS NV30 gonna have, ATI could have followed they Radeon 8500 road but they did not same may happen with Nvidia :rolleyes:


But sure it would be cool if NV30 had 2 or more TMUS per pipe :D



TMU = Texture Unite ? :rolleyes:

[Switch]

I was simply asking a question. I wasn't trying to prove anything.

[jbirney]

The real question is do they have enough real bandwidth to feed an extra TMU? And dont quote that BS 48 GB number as that is effective under special cases...

[Fotis]

I wish nvidia would release nv30's specs!!!
Radeon 9700 has 110million trans and nv30 has 120million.Don't flame me I'm no expert but can you fit 8 more TMU's in 10 million trans?

[legion88]

Originally posted by Switch
Linkage (http://www6.tomshardware.com/graphic/02q3/020819/radeon9700-03.html)
Anyone looked at this before? The 9700 Pro only handles one texture per pipe. Now even the Ti4600 does two. Should the NV30 follow along with Nvidia's 2 textures (or possibly more) per pipeline, wouldn't this give a pretty large increase in performance?

It will be faster (assuming everything else is equal) but "large increase" is stretching it due to bandwidth limitations. I'm assuming you are comparing NV30 to R300 when asking the question.

[Switch]

Originally posted by legion88


It will be faster (assuming everything else is equal) but "large increase" is stretching it due to bandwidth limitations. I'm assuming you are comparing NV30 to R300 when asking the question.

Yes, I was comparing NV30 to R300. What kind of bandwidth would be required to run something like this? Surely it's not too much more than what the R300 has?

[ZenOps]

When Nvidia releases the NV30 it will be faster, it just makes good marketing sense. If they made a NV30 today and it was slower than a Radeon 9700, Nvidia would never release it.

Nvidia has always prided themselves on being the fastest if not having the best visual quality or being the cheapest.

I don't think Nvidia was planning on the Radeon 9700 .15 micron chip being clocked at 325Mhz, I think most people were guessing closer to 200 to 250Mhz (about Pahrelia speeds)

The specs for the Radeon 9700 and NV30 are very close, both are DX9 floating point parts.. Almost indistinguishable actually, except that the NV30 is a .13 part and requires DDRII both of which are very immature technologies right now.

A chip running 325MHz at .13 is not any faster than a .15 chip running 325 mhz (although it will be cooler and consume less power)

Just a guesstimate, but I'd say the NV30 core has to be at least 400Mhz if Nvidia wants to be assured a solid speed crown over the Radeon 9700. Its gonna be tough, Nvidia for the first time playing catchup.

And ATi can move the 9700 to .13 and DDRII anytime they need to. They are probably just waiting to see what Nvidia's response is.

[StealthHawk]

you actually think nvidia would miss a 6 month time period of having no new card out just because their card wasn't faster? that is not a good business practice in the least, and i'm sure investors would lose much faith(more than they already have :p )

we do know NV30 will have at least 8 texture pipes

[Lamentation]

"And ATi can move the 9700 to .13 and DDRII anytime they need to. They are probably just waiting to see what Nvidia's response is."

Wow, can the leap tall buildings in a single bound too? Why don't they just skip .13 and go to .05.

[jbirney]

The R9700 supposedly can all ready support DDRII. However the move to .13u is not easy yet. As time goes buy the kinks of the .13u will be worked out and any other complex chips built will be likely to have less issues then being built while the .13u process is still "new".

[ZenOps]

Yes, it can leap tall buildings too..

But really, NVdiia has locked itself into using a very immature .13 process. They cannot go back now and make a .15 NV30 part to try and make a feasable number of chips at a reasonable price, its too late for that (It would take months for a .15 design, an intolerable delay time).

Intel and AMD had huge headaches moving to .13 and it looks like its no different for Nvidia.

ATi can sit back a bit, wait for the technology to mature, and jump in whenever they want, there is no pressure or rush to get a .13 chip out until there is some semblance of competition from Nvidia.

Interesting read: http://www.lostcircuits.com/video/nvidia_nv30/

[SurfMonkey]

When ATi swap over to .13u they'll hit exactly the same problems that everyone else has. They have the same problems with design and re-design, they'll have the same chance of thigs not working. The only thing they may enjoy over other companies is a better yield at that point.

I can't really see nVidia telling ATi what pitfalls there are moving to a .13u process. And that's why in the long term nVidias choice to move to .13u now will pay off. The six month lead that ATi will enjoy will be destroyed when they move prcesses and nVidia will already have a mature line of chips that will only need refreshing not redesigning.

In the short term ATi is on a winner, in the long term nVidia has given themselves the edge. If they can survive the delay now.

[PreservedSwine]

Originally posted by SurfMonkey
When ATi swap over to .13u they'll hit exactly the same problems that everyone else has. They have the same problems with design and re-design, they'll have the same chance of thigs not working. The only thing they may enjoy over other companies is a better yield at that point.

I can't really see nVidia telling ATi what pitfalls there are moving to a .13u process. And that's why in the long term nVidias choice to move to .13u now will pay off. The six month lead that ATi will enjoy will be destroyed when they move prcesses and nVidia will already have a mature line of chips that will only need refreshing not redesigning.

In the short term ATi is on a winner, in the long term nVidia has given themselves the edge. If they can survive the delay now.

Why would you think that? ATI doesn't fab the board, TSMC does.

[Bigus Dickus]

Originally posted by SurfMonkey
When ATi swap over to .13u they'll hit exactly the same problems that everyone else has. They have the same problems with design and re-design, they'll have the same chance of thigs not working. The only thing they may enjoy over other companies is a better yield at that point.

I can't really see nVidia telling ATi what pitfalls there are moving to a .13u process. And that's why in the long term nVidias choice to move to .13u now will pay off. The six month lead that ATi will enjoy will be destroyed when they move prcesses and nVidia will already have a mature line of chips that will only need refreshing not redesigning.

In the short term ATi is on a winner, in the long term nVidia has given themselves the edge. If they can survive the delay now. And for the opposite viewpoint: ATI will be making the transition with a proven and tested chip design, not a completely new and radical architecture (i.e., NV30 and R300). It's easier to attack a difficult problem a piece at a time, so that you know when that variable is taken care of and you can move on to the next one.

With a new chip + new process, there are many variables that interact and influence the chip behavior, and that may not be trivial to sort out.

Besides, the point that TSMC is doing the fabbing and they will have a better grasp on the .13u process is a valid one. It's not like NVIDIA and ATI fab their own chips and NVIDIA isn't going to share information with ATI: the information will neccessarily be shared because of the common fab.

[ZenOps]

As mentioned, neither Nvidia or ATi has any control over the actual final production of the .13 chips, as neither has a .13 chip fabrication plant to call their own.

Nvidia is using a TSMC fab, which has been quoted very recently as saying that their .13 process is currently only at about 15 percent yield (or 10 to 20 percent depending on who you ask)

At that rate, and for what a typical high-end videocard sells for, Nvidia will actually lose money on every chip they make until the yield gets into the 40 percent range. Nvidia more than likely pays TSMC for the .13 reject chips in addition to the good ones.

But even if they do decide to start making boards, there won't be very many because right now, .13 chips coming out of TSMC are less than 1 percent of total production.

Nvidia took a gamble to try and leapfrog ATi with an unproven technology, and Nvidia is getting burned early. There is no telling when .13 will be a profitable or a mass-production process, but it looks very far off (maybe even late 2003/early 2004)

Ati probably already has a .13 design die shrink for its fully functional Radeon 9700 nearing completion, but won't even think about fabricatingl it until it is possible to actually make money from it.

[Megatron]

I think a better Title for this thread would be why the 9700 beats the Nv30.

The answer would be simple..the R300 is a real product here NOW....the NV30?...well...you know....its coming i guess?..

[SnakeEyes]

Good point megatron. What's available now will always beat stuff you can't get. :D

[Jandar]

Originally posted by Switch
Linkage (http://www6.tomshardware.com/graphic/02q3/020819/radeon9700-03.html)
Anyone looked at this before? The 9700 Pro only handles one texture per pipe. Now even the Ti4600 does two. Should the NV30 follow along with Nvidia's 2 textures (or possibly more) per pipeline, wouldn't this give a pretty large increase in performance?

reread that link...

8 Pixel Pipelines
1 Texture UNIT per Pipe
8 Textures PER Texture unit.

lot more than just 8 textures in a pass....

:shrugs:

[jbirney]

TSMC makes more money when it hits higher yeilds. Throwing silicon away is not good for anybody. TMSC will learn from its common mistakes on the NV30 and apply those to any other IC it fabs on the .13u process. Its smart business.

[ZenOps]

The "paper" specs of the NV30 which has two texture units instead of one is more a function of using DDR2 memory than anything else. Ati can move to DDR2 and the second texture unit once again, anytime they need to or when it becomes profitable (DDR2 memory is very expensive right now)

Snipped from Anands Gigabyte 9700pro overclock to 400Mhz (with that tiny ass stock heatsink too)

"ATI doesn't need a 0.13-micron process to ramp up the clock speed of the R300, but they will need faster memory. Our astounding overclocking successes don't kill the need for a 0.13-micron R300, since ATI will eventually need to outfit each of their 8 rendering pipelines with a second texture unit once the move to faster DDR/DDR-II memory is made."

I think Nvidia bit off way too much, the NV30 is a great chip but if noone has the ability to make it, then what good is it?

I'd say why the R9700pro beats the NV30 too, with an overclock like that, the NV30 will need to debut at least at 450Mhz IMO.

[sancheuz]

Originally posted by SurfMonkey
When ATi swap over to .13u they'll hit exactly the same problems that everyone else has. They have the same problems with design and re-design, they'll have the same chance of thigs not working. The only thing they may enjoy over other companies is a better yield at that point.

I can't really see nVidia telling ATi what pitfalls there are moving to a .13u process. And that's why in the long term nVidias choice to move to .13u now will pay off. The six month lead that ATi will enjoy will be destroyed when they move prcesses and nVidia will already have a mature line of chips that will only need refreshing not redesigning.

In the short term ATi is on a winner, in the long term nVidia has given themselves the edge. If they can survive the delay now.

I agree completely with you. Ati will face the same probelms in redisigning their chipset.

[PreservedSwine]

Originally posted by sancheuz


I agree completely with you. Ati will face the same probelms in redisigning their chipset.

Ever hear of backwards engineering? Not to mention TSMC will have that much more exprierience on the .13 fab process...

[ZenOps]

I'm guessing the first revision for a faster Radeon 9700 pro, which may show up next year won't even be .13 micron. TSMC is going to be too busy scraping the bottom of the barrel to get a handful of NV30's out of their 1 percent manufacturing capability on 15 percent yield. ATi definitely can't go to UMC (even less advanced than TSMC) with a 107 million transistor chip, UMC has always been a little behind TSMC in terms of manufacturing and yield.

It will probably be a slight core revision with a little bit of tweaking to squeeze out a few more mhz on the .15 micron process. It may or may not use DDR2, it depends if Micron or Samsung brings down the price to a resonable level, if not then maybe just using a 333 to 350 Mhz DDR (instead of the 310 they are using now)

The next revision after that would be a .13 die shrink with the second texture unit, and DDRII for sure. The die shrink is necessary with the addition of the extra transistors of the second texture unit, I mean the chip is already physically huge with 107 million transistors now, add a second texture unit, and it becomes wise to move to .13

That timeframe is reasonable and gives ATi and TSMC plenty of time to work out the bugs of the .13 micron process.

Did I mention the .13 micron process is really immature? I think Nvidias NV30 is going to be delayed a lot longer than peeps want to believe.

[PreservedSwine]

Zen- I think you have a crystal ball. That's exactly how things seem to be developing.

So, do you have any stock advice that willprove to be as accurate:eek: