NVIDIA GeForce FX 5200 & 5600

From the standpoint of the user or the programmer? Thing is, there tend to be more configurable settings to ease up on fillrate (hence pixel shader) and GPU-bandwidth demands (resolution, AA, AF, tri/bilinear, texture quality, etc.) than geometry and CPU demands (although variable geometry detail is becoming somewhat common and UT2003 pioneered (AFAIK) variable physics quality to change the CPU load). And it certainly seems (to one who reads website reviews often) that many of today's games are madly CPU-limited, but perhaps this is just a function of website reviews dealing primarily with ultra-high-end GPUs while the performance range of current GPUs is much wider than that of current CPUs.

How would this help things? The developer will still need to support arguably as many different featuresets, because of the need to support GPUs from up to 18-24 months ago (not to mention different IHVs). This would only seem to complicate things: now the developer can't assume that feature support has any correlation with performance.

I will say that F-buffer (or another means of achieving similar functionality) seems to point the way towards the Holy Grail, at least as far as pixel shaders are concerned: where the developer writes one shader, and the hardware/drivers decide how many passes it needs, across all product lines from all (well, both) IHVs--and future GPUs that can run the same code in fewer passes actually do it automatically. In about a year from now when this functionality moves its way from high end to low end...

Or will developers have to have a certain instructions per pass limitation in mind to get optimal performance with an F-buffer (to avoid looping over pass boundries, for instance)? And are shader instruction limits starting at 160 per pass really such an important issue where games are concerned?

 

I agree with your position here. Indeed if ATi can get mass market card of the RV350 at a low price(eg 9600 non pro version) it would be a far better card for end users. But nvidia cheaped out on the Geforce FX 5200 so they can bring the price down and still get some margins on the bugger. Never mind it won't perform DX9 applications well and end users will likely have to upgrade their card to get decent frame rates out of their PC.

I am hopeful though that ATi will be very competitive with the Radeon 9600 non pro version and hopefully OEMs will take it over a Geforce MX 5200. It is clearly going to be a better performer and competitively priced and would give developers a better low end DX9 platform to aim at. It may even perform well enough to use software developed for higher end DX9 products such as the Geforce FX5600, Radeon 9700 pro and non pro, Radeon 9500 pro and non pro. This would be a far better end user solution then the Geforce MX 5200 that will bog down a PC by relying on the CPU more.

 

I'd be very suprised if NVIDIA, or any other fabless semiconductor with outside of China is addicted to "low end margins".

You simply can't make money that way.

 

Russ, the vast majority of the market is made of bulk OEM deals. Nvidia makes most of its money off its low end/low cost products.

 

I work for a fabless semiconductor company and "low end margins" is not something that our company, or any company in the same field wants or can find acceptable to continue operating.

50% margins (selling the product for twice the manufacturing cost) is the goal.

40% is ok

25% is marginal

less than that is plainly unacceptable. You MUST recoup R&D costs, manufacturing costs, pay your engineers and also provide a return for your investors. Why do you think ATI came up with the RV350? It costs less than the 9500 and performs the same function so they can sell it for the same price.

Taiwanese and Chinese companies like SiS, Realtek, Trident, etc can sell for less than 25% margin and still make money because they can pay their engineers dirt cheap and they get R&D and investment money from the government either directly or though research grants (or from violating patents). (SiS also has its own fabs, so part of the profit equation gets trimmed there)

Now, if you want to say "low cost", I'll completely agree with you. But low cost does not imply low margin.

So quit your eye rolling.

 

You are arguing semantics. I meant by low end margins obviously I was saying nvidia low end product margins. Makes no differance to me though how you want to nit pick.

 

Sorry, it just helps avoid confusion if you're using the wrong term for what you want to talk about.

 

IMHO;

it could be possible that the R9600 is priced as high (or low) as the FX5200.

Both cards use 128MB 200MHz DDR-SDRAM
The RV350 is the smaller chip ( according to DaveBaumann )

So only the yield at 0.13µm is an unknown.

Otherwise both cards can be produced at the same cost, and if the R9200 PCB is any indication about the PCB for the R9600 then even the cost for the PCB could be lower, cause as I remember, the RV350 and the R200/RV250 are said to be pincompatible.

 

I agree the real contest is between the 9600 and the 5200, but is anyone as moved as I am by the passive cooling? The RV350 may kill the NV34 in terms of hardware capability, but seeing as I had an 8500LE whose fan died within *six months*, that stationary hunk of metal means a lot more to me now.

I would hope it's possible for RV350 to be passively cooled as well, but I'm not holding my breath.

-- L.J.

 

Going from our knowledge of the transistor count, I doubt the RV350 is smaller than the NV34.

NV34 is supposed to be 40Million? RV350 is 70 million? (vague recollections, too lazy to hunt the real numbers down)

The ratio between .13^2 and .15^2 would put the "same area" at about 53 million. (33% more transistors in .13 for the same area)

Assuming my math is right and the estimates are right, it would seem the RV350 die would be at least 30% more expensive, assuming the same yield and same silicon area cost. We know that .13 yield isn't that hot, which would make the RV350 cost more per good die, and also that a .13 wafer costs more.

With all that, I wouldn't be suprised if the RV350 die was currently 50%-100% more expensive than the NV34.

 

Hrm, what part of the equation do I have that is wrong?

 

That is great news to me. I assumed that the .13um would be more expensive yet. I really hope that OEMs don't buy the NV34 it won't drive developers to push the hardware at all IMO the RV350 would be so much better for developers to consider.

Heck maybe I will even buy a Radeon 5600 it is looking that much better from what i have seen and if it is going to be as cheap as the NV34 then the RV350 is a no brainer.

 

News? this was only my speculation !

 

Little defensive there, aren't we?

Getting their mid-level products to 70mil. trans. from 107mil. should provide ATi with a nice boost in profit, not to mention their .13u yields should only get better with time. C'mon, release those benches already....

BTW, a 5200-level product would be nice integrated into the next nForce (guaranteeing people would buy it with faster CPU's, like with OEM bundles).

 

Another thing in the Anandtech review

Ehh the 5200U is gonna be in the pricerange of the R9500 that does offer DX9, and probably and a much better speed. Sad that he didnt include that card in the graphs :/

The non Ulta 5200 he is right, sub $100 ATI havent got a DX9 part, but how much use is it?

 

Was this Thread hacked? My comment above is ~2 days old.

I just read the Review from Firingquad. Link :
http://firingsquad.gamers.com/hardware/geforce_fx_5200_ultra/default.asp

the PS2.0-performance of the FX5200U is only around 5% of the PS1-performance :



Despite this the FX5200 manages to perform quite good in 3DMark03. Even in the Mother Nature test the frame-rates are quite good (@640x480)