The Nvidia Dilemma

I just wanted the 411 on why Nvidia seems to be lagging behind ATI with their graphics cards. Didn't Nvidia use to dominate the graphics card market? When did ATI get so innovative and why can't Nvidia properly imitate it? ATI somehow pushes out cards that use less energy and dissipate less heat but still perform as powerful or even more powerful than their ATI equivalent. After looking at the 5xxx line and their capabilities with DX11 I was impressed. What technology or design is ATI using that Nvidia isn't? Or is it that the next wave of Fermi Nvidia cards will fix these issues?

Yea I know alot of questions and probably complicated answers but I'm here to learn

 

For why, you have to ask the Nvidia engineers.

From my speculation though, it's because ATI is just focusing on their graphics card, while Nvidia is spreading themselves thin dabbing everywhere like Cuda, ion, handphone processor etc.

 

I think its one word: "rebranding".

 

What do you mean by "rebranding"?

 

Nvidia's marketing is... creative. They sometimes take cards and arbitrarily rename them to something with a higher series number. At other times, their mobile cards (e.g. GTX 260M) are not of the same architecture as their desktop cards (e.g. GTX 260).

I don't think there is anything strange about Nvidia lagging behind ATI -- they trade places from time to time. The GeForce FX cards were substantially inferior to ATI's contemporary Radeon R300 (9XXX) series whereas GeForce 6 and 7 were more or less even so it wasn't until GeForce 8 that Nvidia became dominant. They didn't really count on the Radeon HD 4000 series being as good as it was and didn't do anything new for too long a time.

 

Also, neither company completely dominates. ATi has always offered competitive GPUs at fraction of the cost of Nvidia. That has always been ATi's strategy, to provide the best card for the money.

The Radeon 9600,9700,9800 absolutely ruled for a while. The X800 were also dominating.

Nvidia didn't dominate until the 8800 came out. But then 9800 were just 8800.

ATi came back with the 4xxx series, 4870 providing dominating performance far less cost of the GTX280.

It goes back and forth.

Recently, ATi engineers wanted to go big and really crush Nvidia. But the lead engineer fought them and said know. He wanted to go smaller and more cost effective, and his gamble worked.

 

Also ATi learned from the 2xxx fiasco, meeting deadlines are important.

Nvidia because they are so cocky, never learned this. Their G200 series were also delayed by over 6 months. Then FERMI delayed by almost a year. It doesn't matter because their marketing is so awesome, their profits rose during this time, even though the HD5870 was cheaper than the 280/285 and much much better. Nvida fanboys are hardcore fanboys.

ATi on the other hand made a focus on improving their manufacturing. Before releasing the 5XXX series on 40nm architecture, they actually did a test run with a 4XXX series and it was success. They used that experience for the 5XXX series and were able to make the deadline.

 

The reason ATI were lagging behind was because they experienced some financial difficulties, and because they missed their deadline for releasing the 2xxx series. Their aquisition by AMD seems to have helped them a lot.

Despite what they analysts have said then, the aquisition of ATI by AMD seems to have worked out well for both of them.

The overheating problems that nVidia had with its mobile 6/7/8 series didn't help its reputation. I don't understand why they run hotter than ATI equivalant products?!

 

Their architecture is fundamentally different, plus for most of their products, they're still using a larger manufacturing process. The 40nm nVidia products are on par with ATI's 40nm products in terms of power and relative performance, but this hasn't applied across nVidia's line yet.

 

I also think the merge of ATI and AMD was a good boost for both companies.

 

I don't think merge of ATI and NV would be a good thing for consumers, anyways, I doubt anti-trust laws will allow it anyways.

 

ATI is owned by AMD so there is no point on asking/considering the merge of nVidia and ATI

 

lazy-to-type form of nvidia.

 

The two companies are fundamentally different also.

I think it's good for those who want to spend thousands to squeeze out the last 5% of performance, they have Nvidia.

Because Nvidia exists, I think it's great ATi says, hey if you don't need that last 5-10% of performance, here is something that is 90% of Nvidia for 50% of cost.

 

True...I really always liked Nvidia mobility chips over ATI's mobility...but now that I have the 5870M, I am hoping it changes my mind So far the heat vs performance vs potential on this single mobility 5870 is unmatched and producing good stuff

For some reason, Nivida mobility chips just always ran smoother to me....like the laptops the Nvidia chips were installed on said "I know you... and oh ya baby, you can run like butter on my rig"......BUT to date, Nvidia has nothing close when it comes to mobility chips and the 5870! I'm a proud owner! Now my rig says "SHeeeeiiii^^^^...I know you 5870...lets start doin work!"

 

For some reason Nvidia chips always fried on me with O/C on default settings

 

nVidia was late with Fermi cards for two main reasons. First, they redid how their entire GPU architecture works, from the ground up. This caused many issues that took time to work out. Secondly, their 40nm manufacturing process returned them with very few working samples, delaying everything even more.

If one of those problems had not existed, they would have been out months ago. However, they believe (and I see their logic), that the delay caused by reworking their GPU architecture will pay off in the long run.

 

Ya, Fermi is a new architecture since the Geforce 8 series.

ATI, AFAIK, is still on their old Radeon R600 architecture-base (from the HD2000/3000 days) but with lot of improvements and tweaks of course. Which is likely why the Evergreen (HD5000) cards are quite powerful as well as efficient due to a very mature hardware.

Rumor has it that ATI is aiming for a late 2010/ early 2011 release of their own new architecture as well dubbed the Northern Islands with possibly a 28nm process.

 

woops i meant ati and AMD

 

No it's not. You may want to redo your research for ATi again.

 

so the evergreen series isnt based on a prior architecture?

 

It's complete from ground up.

RV770 was not a RV600. RV600 as a disaster so AMD created a new architecture then too. This article explains they did not create another RV770.

ATi is not Nvidia rebranding constantly.

http://www.anandtech.com/show/2937

 

This is the main reason why ATI is winning now...

I think its better boost for AMD since their CPU division is making losses even now while ATI is having bumper profits.

 

did ATI not assist with the creation of DX11?
i cannot remember hearing where, so i will need someone to back me up on this before anyone can take it as any kind of truth, but ATI helping M-soft with DX11 meant it had a bit of an advantage over nvidia with regards to creating a new DX11 chip. Isn't this also the reason ATI had dx10.1 chips for a good while before nvidia aswell, as dx10.1 is essentially dx11 with only a couple of things missing.... (need back up on most of this)

 

DX10 and DX11 was a collaboration among ATi, MS and Nvidia. After initial talks for DX10, ATi went ahead and supported all of the features of what we now call DX10.1/some DX11. When the DX10 came to be finalized, Nvidia refused on pretty much everything which is why DX10 initially for most people, was like what's the point, stick to DX9.

The fact MS let Nvidia get their way, many believe is the reason why Vista was a failure for gaming, there was no reason to upgrade to Vista for DX10.

It wasn't until years later that 10.1 was implemented, which was what DX10 should have been to begin with, and DX11 is pretty much the leftover of what wasn't implemented.

Direct Compute and Tessellation was on the table for DX10 and Nvidia said FU to that.

Now that DX11 is finally here and Nvidia actually decided in their good graces to support it, Windows 7 has been the BEST SELLING OS for Microsoft for initial sales EVER. Windows 7 upgrades destroys XP, Vista, 95 initial sales.

Put it this way, more people went out and bought Windows 7 in like a month than all sales for Apple's Mac OS X from 2002 until present. Which is kind pathetic for Apple if you ask me.
- This I think is largely because everyone was screaming, OMG Windows 7 is the best PC gaming platform evah!

Windows 7 immediately took 30% of the PC market share within few months, and well, that is almost 3 times the market share for Apple, ROFL. Apple market share last I read was about 9%

 

This is already old. I forgot the reason BUT Northern islands is gone. It will be replaced by Southern Islands which uses SOME of the NEW Architectures that was supposed to be designed in Northern Islands and they will be in 40nm.

 

Holy crap is that true?! Windows 7 did THAT well? Can I get the source for this info?

And if what you said about the collab between MS nVidia and ATI is true, then nVidia made a bad choice slighting both of them like that. I wonder if the next refresh of lower/moderate/moderate-high level nVidia cards will be up to par with the their 5*** equivalents. Anybody have news of those?

 

I'm sure the nVidia's 400 mobile lineup will be mobile conversion not of their high end fermi, but mid end fermi cards, their high end fermi is simply too hot and consume too much power.

 

I thought Northern Islands was only on hold because TSMC was having trouble with 32nm (or was 28nm?) and Southern Islands was a sort of stopgap?

 

ATi scrapped 28nm and went to 40nm and renamed it to South.

But AMD is one of the biggest investors to a rival silicon company that is working on 28nm now.

So who knows.

 

so i was half right

thats a half win for me

 

Well as far as i know it was supposed to be 28nm(32nm was skipped) BUT with TSMC problems they decided to stick with 40nm and just use some of the new architecture.

So the architecture of the chips is still new.

I think the 7 series will use the 28nm and the WHOLE NEW arch.

I want a 7770 just for the sake of being lucky.

 

ATI HD 8888 would be Asia's lucky number.

 

I dont think that is a possibility(having 8888 maybe 8880) and also maybe you mean CHina's lucky number.

 

My thinking is that Nvidia relied too much on the excellent g92 cores (the 88xx series) that they almost focused just on shrinking their technology. There is also their bet with PhysX and CUDA stuff which we still are waiting to see if there can be real rip of benefits from such technologies.

 

Back on-topic, y'all. This thread will get closed or moved to off-topic if things keep going like this.

 

I hope not. I hope they scrap PhysX and CUDA. The general attitude I personally have read from game developers is that PhysX hardware accerlatered Physics is dumb. It should be moving towards GPGPU with OpenCL. Considering ATi Stream Processors are just as powerful as Nvidia CUDA processors and Intel making a move for GPGPU, why not just unify it under OpenCL? Nvidia can keep pushing for their CUDA, but there is no way ATi and Intel are going to support.

I'm just a fan of unification of anything for industries. It just makes it easier and better for consumers. At leas that's my opinion. I don't know if Microsoft should do it, but a unified API for Physics would be awesome also. Having to deal with PhysX really annoys me.

 

TSMC is such a fail isn't there any other company out there which can produce GPU's? Also good to know AMD is investing in a rival silicon company.. maybe we'll have something way better for ATI cards compared to NVDIA ones when ATI uses this new company and NVDIA uses TSMC...

 

What is GPGPU and TSMC? I think CUDA is pretty cool as it can help out with weak processors. Physx I think is kinda of useless but that's just me.

 

CUDA is GPGPU. But Nvidia holds GPGPU development hostage with their stupid push for CUDA. Intel Larabee was pretty much about creating a graphics card with x86 architecture and massive GPGPU power.

The new ATi HD5870 has 800 Stream Processors, which are GPGPU just like CUDA. On some applications, Stream Processors actually blow CUDA out of the water, not even funny.

Overall GPGPU has a lot of potential if Nvidia decides to stop stalling, kill CUDA and just get on board with OpenCL. CUDA is pretty much a set of software and programming language to use their GPGPU power. OpenCL is the same thing except you know, it's Open for everyone to use.

Imagine more games and applications using GPGPU, but we can't because developers either have to decide to use CUDA or ATi Stream, but wouldn't it be better if there was full support for OpenCL so that anyone, despite whatever platform can use it? BTW ATi is one of the biggest supporters of OpenCL.

 

Nothing stopping Nvidia from using Global Foundries as well.

http://www.theinquirer.net/inquirer/news/1356904/nvidia-talks-globalfoundries

 

its really not that big a deal. the companies often go back and forth in dominance.