Charlie's thoughts on the fermi derivative parts

[Dark4ng3l]

In 3 or 4 months 5750s could be priced under 100$ with 5770s close to the 100$ mark. ATI already made money to cover their fixed costs on these things in the last 6 months they won't hesitate to drop prices to compete.

Anyways we are speculating that the performance of GF104 would be similar vs the die space of GF100 parts. But when you combine a full 256 shader part with potentially 10-20% higher clock speeds and better memory clocks it could end up being a good part. The problem is that as bunnyfubbles pointed out in terms of manufacturing costs it would be really close to cypress parts. That makes it doubtful that it can compete with Juniper in terms of price point.

 

[Madcatatlas]

But it MAY! so Haleluja to that!

 

[pmv]

Except the 5850 is like $300~ and performs about the same as 470 making it irrelevant.

In this country the 470 is priced the same as the 5870 - £330 approx $500, making it (the 470) totally pointless.

 

[Leyawiin]

At least the frothing-at-the-mouth vitriol has been toned down to something that resembles reporting.

 

[Rezist]

I also agree that the GF104 should have better clocks and yields and probably better performance per mm^2 compared to standard fermi. I hope they get a part that with a good OC is within 85% of the 5850.

 

[v8envy]

There's still a big hole in the ATI lineup between the 5770 and 5850. Yes, I know the 5830 fits there pricewise, but at about 10-15% faster than a 5770 it still leaves the performance gap untouched.

A higher than 480 clock (800 mhz or so) 256 shader part should fit midway between the $140 5770 and $290 5850 and sell well for low $200s. The performance mainstream segment is huge, and so far ATI seems content to ignore it.

 

[OCGuy]

Which market?
(NV results for their last 2 financial years according to their SEC filing available from the NV website. Numbers in brackets indicate losses)

Most of 2009 was a loss for both AMD graphics and nV due to the economy, it had nothing to do with product.

Now that the GPU market is rebounding, nV pulled a profit in Q4 2009, despite no DX11 part, and GT200 being "eol'd" according to paid rumour sites.

http://www.tomshardware.co.uk/nvidia-financials-GPU-quadro-tegra,news-32820.html

 

[BenSkywalker]

But nvidia's not getting paid the retail price, is it?

Is ATi getting paid retail price for the 5830? If anyone thinks a company isn't turning a profit selling a video card for $500 using a chip that costs $250 they do not understand this market at all.

But I wouldn't be surprised if they're not making a penny on them given that the upper/high end cards are never really there to make huge profits

That is precisely what the high end market does, makes huge margins. Overall profits are quite a bit lower as there are orders of magnitudes more low end parts being sold, but raw profit for the high end is markedly higher then the other segments(outside of workstation/HPC of course).

So?

Started at $200, DDR3.

What about yields?

And don't ppl say AMD didn't do money with RV770?

We are talking about parts that are going to launch in the $200 range, they are about the same die size and comparable PCB complexity to parts that sold millions of units in the ~$100 price range. Saying nV can't manage to manange something on the financial end that their competitor was doing a year ago is just flat out stupid. RV770 didn't make AMD a ton of money largely due to the fact that it only sold 1/2 of what the green team sold. A huge portion of the real costs on these parts is R&D, scales of economy are huge factors when dealing with R&D costs approaching $1Billion dollars. Yields are likely to be directly comparable to any other part around the same size. They are almost certain to be quite a bit higher then GF100 parts out of the gate simply due to being roughly half the die size. Given that they are a bit over half, it seems like nV is planning on having redundant units built in to compensate for yields meaning their potential yield rate improvement over the GF100 can reasonably be expected to be *significant*.

In 3 or 4 months 5750s could be priced under 100$ with 5770s close to the 100$ mark. ATI already made money to cover their fixed costs on these things in the last 6 months they won't hesitate to drop prices to compete.

I fully expect ATi to adjust prices to compete and feel rather confident that nV is going to be comfortable retaining a price premium over ATi as the market has proven repeatedly that they are willing to pay a 'green tax' for nV. It is rather dumbfounding that anyone is going to attempt with a straight face to say that nV can't make parts with these specs for a profit when ATi managed to do so with prices a full 50% lower for quite some time. This isn't a borderline or close call point. It is utterly assinine to think that a company is incapable of making profits on margins that huge. The question revolves around volume, can nV produce and sell enough to recoup their R&D costs. The raw margin numbers are looking to be quite high based on the breakdowns.

 

[Lonyo]

Most of 2009 was a loss for both AMD graphics and nV due to the economy, it had nothing to do with product.

Now that the GPU market is rebounding, nV pulled a profit in Q4 2009, despite no DX11 part, and GT200 being "eol'd" according to paid rumour sites.

http://www.tomshardware.co.uk/nvidia-financials-GPU-quadro-tegra,news-32820.html

You must have some different ATI numbers to those filed with the SEC then, because that's not what I see from ATI's results.

 

[BenSkywalker]

You must have some different ATI numbers to those filed with the SEC then, because that's not what I see from ATI's results.

The ATi division posted a $53 million profit in Q4, they posted a $50 million dollar profit for the year. Through Q3 they were at a $3 million dollar loss for the year. His statement is accurate and agrees with the SEC numbers.

http://phx.corporate-ir.net/External.File?item=UGFyZW50SUQ9MjcyODB8Q2hpbGRJRD0tMXxUeXBlPTM=&t=1

 

[v8envy]

Saying nV can't manage to manange something on the financial end that their competitor was doing a year ago is just flat out stupid.

While I agree with everything you said I do have a nitpick. The 4850 was presumably made with defective cores unsuitable for the higher margin products (4870, 4890). So any $ received for the otherwise unsellable 4850 cores was "bonus" income. If it weren't for the higher end SKUs I doubt we'd have seen $100 pricing on 4850s for long.

In comparison, the GF104 will have to stand on its own. These will not be defective Tesla cores otherwise worth $0. And they're being manufactured on a demonstrably flawed 40nm process, whereas to my knowledge 55nm yields were not constantly blamed for persistent part shortages.

I have a very hard time swallowing $500 video cards being a loss leader myself. It smacks of bullshit even more so than Fermi being limited to a single run.

 

[Stoneburner]

I fully expect ATi to adjust prices to compete and feel rather confident that nV is going to be comfortable retaining a price premium over ATi as the market has proven repeatedly that they are willing to pay a 'green tax' for nV. It is rather dumbfounding that anyone is going to attempt with a straight face to say that nV can't make parts with these specs for a profit when ATi managed to do so with prices a full 50% lower for quite some time. This isn't a borderline or close call point. It is utterly assinine to think that a company is incapable of making profits on margins that huge. The question revolves around volume, can nV produce and sell enough to recoup their R&D costs. The raw margin numbers are looking to be quite high based on the breakdowns.

You speak in declaratives. You do not speak in "evidentiary burdenese"

Seriously, you need to support assertions with evidence. It helps people follow your train of thought.

 

[BenSkywalker]

In comparison, the GF104 will have to stand on its own. These will not be defective Tesla cores otherwise worth $0. And they're being manufactured on a demonstrably flawed 40nm process, whereas to my knowledge 55nm yields were not constantly blamed for persistent part shortages.

If the GF104 was supposed to be ~275mm I would agree with that take, but in the ~325mm range it sounds to me like they have build redundancy in and are planning on one core being bad on every single chip and still being sellable as a top bin part. The product whole they have in their normal lineup, the 460 part, seems like it could end up being further castrated GF100 core offerings or perhaps GF104 cores that work @100% depending on how the yields play out. Seems to me like they have compensated for yield issues with the GF104 fairly decently if all the current info holds up. Yes, it is going to be more expensive to manufacture then the 5770/5750 parts, but it likely will cary a decent price premium anyway(and unlike the 480/5970 caliber parts, chip costs are a smaller factor in overall pricing by a decent margin in this segment).

You speak in declaratives.

Only on being able to make a profit on a $500 graphics card with a $250 GPU cost. I absolutely speak in declarative terms in that instance. In terms of why I can so strongly state that the cost assumption is absurdly wrong- go check out the price of a motherboard and 1GB RAM- both being sold for a profit. The cost of building a graphics card PCB is quite a bit lower then a mobo, but even if we assume it is the same it would still be pretty hard to come remotely close to losing money.

 

[cbn]

These derivative parts will have the "double precision" taken out of them?

If so, will that make them faster per mm2 silicon in games?

 

[BenSkywalker]

These derivative parts will have the "double precision" taken out of them?

Not likely, that would require some significant reworking of the shader cores.

If so, will that make them faster per mm2 silicon in games?

Smaller die size, less power consumed, less heat generated- combined those should spell higher clock rates which appears to be their best bet for increasing per mm performance based on what we have seen for scaling to date.

 

[Soleron]

Here's Charlie's '$250' logic in more detail.

"If NV hits 20% yield (for both cards combined), and assuming all chips are '480s', then the cost for each good die is $5000 / (104 * .2), or ~$240. After testing and packaging, that number is well over $250. This is just for the chip mind you.

Add in $25 for the PCB, $25 for the cooler, $35 for the GDDR5, $25+ for everything else on the board, and $25 for shipping, the box, dongles, and software, and you have a card with a cost of $385.

Lets just say that NV ships 2 470s for every 480, so 20K 470s vs 10K 480s theoretically at launch. It doesn't matter how much they 'charge' for the GPUs themselves, it all averages out to $250 or so whatever they put on paper. That means that ASP for the boards is (($349 * 2) + $499)/3 = $399.

Cost ~=385, retail ~=$399. Can you say HUGE loss? Really. Retailers aren't going to take zero margin for the privilege of selling GTX480s. Distributors aren't going to take zero margin for the privilege of selling GTX480s. That means the AIBs, who aren't going to take zero margin either, will sell it for no less than $350 average. NV, who sells the boards whole, isn't likely to ask more than $300-325 either.

This means NV looses $50-75 per GTX4x0 they sell. A sane company would not try to make this up in volume."

Also:

"I usually lowball that kind of number to give NV every benefit of the doubt. Realistically, the cost of the parts is higher than I said, the yield is definitely lower, PCBs cost more, and coolers cost more.

I just shoot low to avoid people bickering about things like, "You said the cooler was $25, but it is really $24.07, so the entire story is bullshit". "

 

[TheMeanestGuest]

If the supposed prices and performance for these parts are correct, that is unfortunate. It is doubly unfortunate that people will probably buy them regardless without considering this.

I was actually kind of intrigued until I found out the prices. I suppose I'll be waiting until the next iteration of cards to upgrade, as I don't find anything particularly appealing over my 4850 as things stand today.

 

[cbn]

Not likely, that would require some significant reworking of the shader cores.

Does this mean we will likely see less competition for ATI then?

What can Nvidia do to force ATI to lower prices? How can they make their chips faster in games per dollar spent?

It seems to me focusing on HPC is causing Nvidia to reach the floor sooner where selling chips as gamers don't make sense.

Will this decreased amount of competition give Intel enough wiggle rrom to enter the discrete GPU market?

 

[badb0y]

In this country the 470 is priced the same as the 5870 - £330 approx $500, making it (the 470) totally pointless.

What about the GTX 480 compared to the ATi 5970?

 

[tommo123]

the cheapest i've seen (OCUK) is:

5970 = £530
480 = £445

 

[OCGuy]

Does this mean we will likely see less competition for ATI then?

What can Nvidia do to force ATI to lower prices? How can they make their chips faster in games per dollar spent?

It seems to me focusing on HPC is causing Nvidia to reach the floor sooner where selling chips as gamers don't make sense.

Will this decreased amount of competition give Intel enough wiggle rrom to enter the discrete GPU market?

I think you are still missing the point. nV doesnt NEED to win price/performance. In Q4 2009, with nothing to compete against 5XXX in the "high-end", they still turned a nice profit.

You are comparing ATi and nV as if they are on equal footing. nV has better marketing, better brand recognition, and people are willing to pay for it.

480 and 470 + whatever cut-down chip comes out of Fermi will extremely well.

 

[bunnyfubbles]

There's still a big hole in the ATI lineup between the 5770 and 5850. Yes, I know the 5830 fits there pricewise, but at about 10-15% faster than a 5770 it still leaves the performance gap untouched.

A higher than 480 clock (800 mhz or so) 256 shader part should fit midway between the $140 5770 and $290 5850 and sell well for low $200s. The performance mainstream segment is huge, and so far ATI seems content to ignore it.

The GF104s are still ~2 months away at the earliest, and you're also assuming AMD is going to keep their current pricing.

Again, as far as resources go, there's not much keeping AMD from pricing a 5850 par with the highest end GF104 - they're price gouging right now due to lack of competition.

 

[crazylegs]

I think you are still missing the point. nV doesnt NEED to win price/performance. In Q4 2009, with nothing to compete against 5XXX in the "high-end", they still turned a nice profit.

You are comparing ATi and nV as if they are on equal footing. nV has better marketing, better brand recognition, and people are willing to pay for it.

480 and 470 + whatever cut-down chip comes out of Fermi will extremely well.

Yes, but that legacy will not last indeffinitely, its only going to work in the short term.

If AMD/ATI can keep winning the midrange, priceerformance crown, then NV are going to need a good response in order to maintain their footing in the market.

 

[v8envy]

For now. 3DFX was more established and had loyal fans willing to pay more for less. Look where that got them. This is generation 2 where ATI has the better price/performance product, and even the dimmest bulbs are starting to catch on.

I don't think any of the NV top brass or stockholders are ecstatic with nv's position in the market or currently retailing products. Which is why I think those products will rapidly improve. If history is any guide then Fermi II should be absolutely awesome.

 

[OCGuy]

Again, as far as resources go, there's not much keeping AMD from pricing a 5850 par with the highest end GF104 - they're price gouging right now due to lack of competition.

Isn't that was what Zoners hate Intel and nV for doing when they are in the position to? (Although Intel has been kind to us recently even though they are leaving AMD in the dust, but that is a different subject).