NVIDIA shows signs ... [2008 - 2017]

Status
Not open for further replies.
popcorn.gif


Digisan offers it around to anyone else waiting for the main event to start
 
I know there is a die price calculator somewhere on teh internetz where you just enter your wafer size, cost per wafer, die size and yield and off you go.. saves you some CALC.EXE

p.s. the numbers will be semi-accurate.

I would also love to see a comparison of GT218 versus RV710 and see which chip loses out on price, power draw and costs.

digi, sugared or salty?
 
Both are being made on the same process, TSMC's 55nm. What are the relative sizes (484mm^2 vs 256mm^2 or so from memory)? How much does high bin GDDR3 cost? (~$3.25 chip @ 1GHz) How much does GDDR5 cost? (~$3.50/3.80 for 1Ghz/1.25GHz respectively). How much is a TSMC 55nm wafer? (~$4000). What do you think yields are for each product? (roughly proportional to the die size at this point of maturity). How many board layers does the G200b use vs the R770? (hint: very dependent on the memory interface width) How much does the PCB cost? (see last Q) How much do the other components on the board cost, like heatsink, fan, VRMs, passive components, connectors etc? (about the same for each card). How much does assembly cost? (more for G200, more components and a more complex PCB, but not by much). How much does packaging cost? How much do add-ins on the package cost, like dongles, cables, CDs, games and manuals? Shipping?

If you look all that up, it is called research. I did it. I then wrote it up, that is called reporting. If you doubt my numbers, you can go do it yourself, it really isn't that hard, it just takes a bit of time and phone calls. You will end up with teh same conclusion that I did here:
http://www.semiaccurate.com/2009/10/06/nvidia-will-crater-gtx260-and-gtx275-prices-soon/
When forced to price relative to Juniper, NV can't sell G200b parts at a profit. If they don't price it relatively, the parts will sit on the store shelves. $400 GTX285, or $259 HD5850? Your money, do what you want with it.

So, rather than running around whining, why don't you actually go and do the research like I did? Then you will have your answer, to whatever degree you feel is necessary to document the problem or as you posit, lack thereof.

Now, you can extend the exercise to Cypress and Juniper. The one key number is that a TSMC 40nm wafer is ~$5000. If you start at the silicon level, how much does 181mm^2 of a $5000 40nm wafer cost, and how much does 484mm^2 of a $4000 55nm wafer cost? From there, the only real difference between the two is ram costs, call it $3.25 * 8 for NV, $3.50 * 8 for ATI, and off you go.

When you have parts totals and assembly totals, subtract that from the selling price. Start with Juniper, but *HINT* don't take your numbers from Fudzilla, or at least the numbers published from them so far. On one side, you will end up with a positive number, the other negative.

Your homework Trini is to figure out which is which. Then you can consider yourself edumacated, and have someone put a gold star sticker on your forehead. This is not a path off the short bus however, that will take a little more time.

The next time you don't believe what I write, go do the research, you will look a lot smarter. Denials followed by a statement that can be paraphrased, "but I really don't have a clue about the real numbers" is not a way to make yourself look good.

You now know what to do. Clock is ticking. Go!

-Charlie

:LOL:

man your posts are a riot, oh just for your research, your figures are off, you aren't taking into concideration volume discounting at all, not to mention subsidizing by TSMC. You also didn't concider the fact before the 9800 series every generation from nV had shortages of the high mid and high end models, before launch of the next gen line.

and oh yes you don't have a clue about the real numbers. Please go do more research when you get a chance.
 
:LOL:

man your posts are a riot, oh just for your research, your figures are off, you aren't taking into concideration volume discounting at all, not to mention subsidizing by TSMC. You also didn't concider the fact before the 9800 series every generation from nV had shortages of the high mid and high end models, before launch of the next gen line.

and oh yes you don't have a clue about the real numbers. Please go do more research when you get a chance.

it's very possible you are indeed correct however attacking the validity of another's comment without posting factual evidence to the contrary and simply saying "you are wrong", in fact makes you side appear weaker (and with equal possibility incorrect). I'm not defending Charlie just that he did (apparently) construct a valid opinion based upon his research where as others (such as yourself) only rebuttal is that his info is wrong without any substantial counter.
 
he can't make those claims without knowing what I stated, I can't say how much those are but just simply taking into the fact nV has 2 times the marketshare they will need twice the amount of wafers, its pretty easy to see what the outcome is. No amount of research will anyone in the outside of the management teams will really know the exact figures. There is also kickbacks as well. Its not something that is even accessible to him or me for that matter.
 
http://www.xbitlabs.com/news/video/...a_of_Creating_Shortage_on_the_GPU_Market.html

Unless it has been posted already and I've missed it. I've no idea what is true but it puts a different light on the story and excuse me it distances itself from what one or the other side's sock puppet has to say.

Didn't AMD already state to virtually all reviewers that Rv770 and Rv790 were going to be EOL'd due to the impending release of the 47xx series?

So basically a product is being phased out in the face of it's direct replacement for AMD.

Nvidia is phasing out a series of products with what as replacement?

Regards,
SB
 
man your posts are a riot, oh just for your research, your figures are off, you aren't taking into concideration volume discounting at all, not to mention subsidizing by TSMC. You also didn't concider the fact before the 9800 series every generation from nV had shortages of the high mid and high end models, before launch of the next gen line.

and oh yes you don't have a clue about the real numbers. Please go do more research when you get a chance.

OK, I posted something very close to the exact numbers, and you say they are off, but won't say why or what the real ones are. Wow, talk about short bus candidates advertising their deficiencies.

As for volume, do you really think that ATI and NV are in different volume pricing tiers? Oh, what, that would require you to think and do some bare minimum logic. You demonstrated how good you are at that when you tried to register for my forums and did not put in a real email address for your registration to go to even though directly instructed to do so. I'll bet you are still trying to figure out why "I am silencing your voice" or whatever.

So, since I don't have a clue about the real numbers, enlighten us. Really. Just think about the alternatives, you could spend the time here writing up some brilliant thesis on silicon costs, the foundry model, and volume discounts, or go back to work and get many people fries while dreaming of a 9400GT that you may save up enough for.

Put up or shut up.

-Charlie
 
OK, I posted something very close to the exact numbers, and you say they are off, but won't say why or what the real ones are. Wow, talk about short bus candidates advertising their deficiencies.

As for volume, do you really think that ATI and NV are in different volume pricing tiers? Oh, what, that would require you to think and do some bare minimum logic. You demonstrated how good you are at that when you tried to register for my forums and did not put in a real email address for your registration to go to even though directly instructed to do so. I'll bet you are still trying to figure out why "I am silencing your voice" or whatever.

So, since I don't have a clue about the real numbers, enlighten us. Really. Just think about the alternatives, you could spend the time here writing up some brilliant thesis on silicon costs, the foundry model, and volume discounts, or go back to work and get many people fries while dreaming of a 9400GT that you may save up enough for.

Put up or shut up.

-Charlie


They aren't really close, there is a tier a system at TSMC, Charter, and other fabs. The wafer cost is actaully between $3000 - $4000 a wafer (at the higher volumes for the general process), not to mention you have forgotten the different types of 55 nm processes, which also effects the cost of the wafer. There is also depending on yields a break down and reduction of cost per functional chip. Should I go on, or did your research go into even this depth (which is quite shallow)? Heh you want a thesis or should I just do an antithesis and be done with? To pick a general number and work backwards will give you a very rough and limited idea of cost of manufacturing but again that just suits you fine for your goals in your articles.

As for you forums, I decided it wasn't worth the time to post on your forums ;)

for a start here is something I suggest you get

http://www.gsaglobal.org/publications/pricing/index.asp

Process Geometry
Process
Wafer Size
Company Type
Foundry Location
Metal Layers
Poly Layers
Mask Layers
Epitaxial
Development Stage
Volume of Wafers Purchased
Other Services Included in Price

Package Family
Leads
Units Per Week
Drop-In Heat Spreader Used
Wire-to-Package Ball Ratio
Substrate Cost
Substrate Technology
Substrate Layers
Wafer Size (for WLP only)

All of these effect the cost of the chips, so is your analysis correct? Well I guess I put something up.......
 
Last edited by a moderator:
Does anybody know if Graphics includes IGP?

At the end of the day. If profit from chipsets was more than 12mil then old ATI products are in black in that quarter.
IGP is included in the computing solutions breakout.

I know there is a die price calculator somewhere on teh internetz where you just enter your wafer size, cost per wafer, die size and yield and off you go.. saves you some CALC.EXE

p.s. the numbers will be semi-accurate.
http://mrhackerott.org/semiconductor-informatics/informatics/toolz/DPWCalculator/Input.html
 
They aren't really close, there is a tier a system at TSMC, Charter, and other fabs. The wafer cost is actaully between $3000 - $4000 a wafer (at the higher volumes for the general process), not to mention you have forgotten the different types of 55 nm processes, which also effects the cost of the wafer.

Sigh, here we go again. Must remember to use small words.

There are different prices for different processes? Really? Wow, I never would have guessed that one! You forgot to add that metal layer count also adds to the cost, but I won't think any less of you because you missed that, mainly because it isn't possible.

So brainchild, do you think NV and ATI use the same process for their chips, or different ones?

Also, what tier do you think both NV and ATI are in? Do you REALLY thing that with the volumes they run, either isn't in the top tier? Come on now.

In the pricing scheme, we also agree! Wow. Go read the original post here:
http://forum.beyond3d.com/showpost.php?p=1344910&postcount=1500
where I said, "How much is a TSMC 55nm wafer? (~$4000)". Now, a 55nm wafer that costs, by your numbers, $3000-4000 is likely modified down due to volume, and up due to layer count.

So why did you post that part? Yelling at the wall?

There is also depending on yields a break down and reduction of cost per functional chip.

Not that your sentence makes sense, but I gather you are trying to say that yield affects cost. I agree. Which is why I said in the original post here:

"What do you think yields are for each product? (roughly proportional to the die size at this point of maturity)"

I guess you didn't actually read what I wrote before you went off in a huff did you?

Should I go on, or did your research go into even this depth (which is quite shallow)? Heh you want a thesis or should I just do an antithesis and be done with? To pick a general number and work backwards will give you a very rough and limited idea of cost of manufacturing but again that just suits you fine for your goals in your articles.

My research was a lot more in depth that I printed, and I will freely admit that I rounded some numbers to protect sources. What is your point. You haven't said anything in your 'rebuttal' that didn't BACK UP what I wrote. Own goal there guy.

I didn't go into huge depth in the article because that wasn't the point, and it was long. To date, the only number you have posted is a vague range that agrees with me. I would ask you to go on, but starting would be a better thing to request.



As for you forums, I decided it wasn't worth the time to post on your forums ;)

for a start here is something I suggest you get

http://www.gsaglobal.org/publications/pricing/index.asp

(stuff cut)

All of these effect the cost of the chips, so is your analysis correct? Well I guess I put something up.......

No you didn't. You put up things that effect wafer cost, but you didn't actually put numbers on it. You can't refute my numbers without numbers. It is like asking how much a drink costs at a restaurant, and you go off on chemistry that affects the color of the glass it is being served in. Try answering the question for a start.

I don't need the publication, I can get the info from the people doing it. This tends to be a much better way of doing things than getting it from a third party report that may or may not have the exact data you need.

So, let me use smaller words. What is the Bill of Materials (BoM) breakdown for the GTX260, 275, 285 and 295, and for the HD4870, HD4890, HD5850, HD5870 and Juniper cards? Also, specifically list what the silicon (GPU/ASIC) costs are for each. If you want a pat on the head as well, how about volume GDDR prices as well.

I know these things. You obviously don't, or you would not have dodged the question while trying to make yourself look intelligent. NUMBERS. Put up or shut up.

-Charlie
 
Volume costs when I gave that price wasn't what you are saying, nor was I talking about yields vs cost per wafer.

Charlie, the stuff you cut out and that link, those guys do it for a living, if you want, I suggest you buy the yearly reports and then look at your numbers, if you can't afford to do so, contact them and see if they would be so gracious to give you last years info or the year before that, until you learn how to communicate and understand where I'm coming from (if you read anything in your life the about semiconductor industry and implicit and explicit cost models and evaluations) then I'll chat and I'll be all ears, at this point its pretty obvious you have no clue what you are talking about ;)
 
Last edited by a moderator:
Who knows what NV will do to pad the time, but they've done it before, when the FX was a disaster. They've got a large portfolio of tech, and lots of internal costs they could cut. I think it is far too early to engage in hyperbole.

Checking newegg I dont see a single valid Nvidia buy (in the upper end) at this time...

5870/50 trumps the 285..when you can get them. The cheapest 285 is $305 AR, $325 normal (at least it's not a sparkle). The next cheapest is $355/335. Etc.


4890's are selling at $185-200. One model is $169 AR. The XFX model for reference is 199 (most people prefer to buy XFX or EVGA for lifetime warranties all else equal). It's competition the 275 is $220 at the low end (a couple models $210 AR), and $249 for the EVGA (dont see any XFX). Plenty of stock.

4870 1GB is $145 for a XFX, $125 AR. Another model is available for $135 AR, etc. Non rebated pricing about $150 and up. It's competition the 260 starts at about $169, including a EVGA model. So that's about another $20 win for ATI. There are a few 260's at $149 AR. Plenty of stock.

9800GTX+ there's one PNY model currently for $99 AR, $129 normal. The EVGA model is $135/120. This card appears to be dying out with only about 4 models. The 4850 starts at $89/99, the XFX is $125. 4870 512 MB also is an option here, starting at $125. Too be fair to Nvidia the reviews I'm looking at show the 9800GTX+ to be faster than the 4850, though I'd assume 4870 512 would top it. At least on paper Nvidia fares best here, as theoretically for $99 you cant get the GTX+ vs $89 for the 4850, and the GTX will be faster. However, if you ignore rebates as I prefer, you're looking at $99 4850 vs $120-130 for the lowest end GTX+'s.

Summation of Nvidia on newegg, doesnt look too good currently...the ATI part usually has a clear price edge. Nvidia also seems to rely more on troublesome rebates to even be price competitive throughout the range. And while there are available lower priced models, it seems a good portion of any Nvidia chip shoots to a high price quickly (ie, while there will be one or two models at the low end, there are several more from various AIB's that are just priced way too high to make any sense). For example, while there is a 285 available at $305 AR, there are several approaching $400 as well..I dont see anybody buying those.


So thats the situation currently, Nvidia may be in flux, so is ATI with Juniper coming, but unless you just really like Nvidia there are few decent Nvidia buys right this second at all.
 
Last edited by a moderator:
Status
Not open for further replies.
Back
Top