-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
ATI cuts 6950 allocation
- Thread starter An Exception
- Start date
HD6950 - 1440 shaders? HD6970 - 1600 shaders than. Sounds like 5800 refresh
If the top card is 1920 SP with 80 cores per SM then I'd have thought the 6950 would have been more like 1600 SP.
It bothers me a bit because 1536SP is still coming up for the 6970 but I can't work out if its actually 1536 or a mis-calculation of 1920SP based on the architecture changes.
Last edited:
It definitely doesn't sound right at all, that's what 1440 shaders for HD6950 would mean though.
Associate
- Joined
- 10 Nov 2010
- Posts
- 115
What does Wikileaks say about it all?
What does Wikileaks say about it all?
That's probably the secrets contained in their Insurance.aes256 file...
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
UNfortunately this is again you twisting the argument.
AT no stage have I ever, or have I ever seen any other AMD fans suggest AMD are selling unrealistically low prices as a market share grab or purely for volume, stop changing your argument EVERY SINGLE TIME you post.
This is where you seem to be completely misconscrewing everything, probably on purpose.
Here, lets say you get 100 480gtx candidates per $5000 wafer, and you're only pulling 30 gpu's with all salvaged units off a die, to break even you'd have to sell every gpu to AIB's at $166, this would leave you 10's to 100m's in the red due to R&D and other costs, you need to make profit on every single gpu, ideally you want to sell those gpu's at $300, but lets call it double for parity's sake. 30x$166=5k profit on 30 cards.
Now take the 5870, you get 140candidates per same cost wafer, you're pulling 90 or so minimum, thats $55 per core. to make the SAME(this is what you seem to be missing) profit per wafer you would only be required to sell each core at $105, thus leading to a each core costing 1/3rd to AIB's.
Here's the bit you keep missing, AMD can add on $166 per core if they want, for a total price of $221 per core to AIB's, and still come in MUCH CHEAPER, yet at that price over 90gpu's they'd be pulling in 90x$166 profit, or 3 times the profit. So for every $5 wafer, you're looking at, if they take the same profit per gpu, about 3 times the overall profit, for a card thats STILL over $100 cheaper to the AIB's, if they take a similar profit per wafer, then they can destroy Nvidia pricing, if they go somewhere inbetween they end up significantly cheaper and highly profitable.
You keep excluding these factors, AMD to make the SAME LEVEL OF PROFIT can sell there cards much much much cheaper and thats the bit you keep completely ignoring.
No one, no one amd side is talking about selling at cost, at silly low pricing, to gain market share, what we're talking about is when the core is 40% smaller, and the yields are hugely higher because they are working WELL within the limits of the process, then by pricing MUCH lower they are only making the SAME profit as Nvidia.
If AMD sold a Cayman at the same cost as a 580gtx, they'd be making significantly more money per sale, but they'd be dramatically limiting sales by making it a £450 card.
By selling it at a far more sane £350, they will most likely be able to make a higher margin than a 580gtx can, with far larger supply and because there is a larger supply there would simply be too many on shelves, at too high a cost.
The fundamental problem is not AMD's benevolance, an idea you seem to put across, nor is it AMD selling at cost, an idea you keep putting across, nor is it a market share grab, they simply make cores with similar performance that are FAR smaller than Nvidia meaning for the same profit levels they can sell significantly cheaper than Nvidia can. To sell at the same price would be hurting their own sales, for no reason.
The 5870 has a FAR healthier profit at £300 than a 480GTX had at £400, if the 5870 had sold at £400, it would have sold significantly less cards, it really is that simple.
Just to point out, this strategy that you insist we think AMD are running, is EXACTLY the strategy Nvidia has been forced into.
Without their professional gpu market, they'd be DEEP in the red, the 470gtx loses cash per core, the 460gtx is being sold basically at cost, probably a loss on the 768mb ones, the 260gtx was sold at a loss, the 275gtx was probably sold at cost and the 280gtx/285gtx had to be sold way below what Nvidia wanted.
Last edited:
So your not suggesting the 6970 is gonna be in the £200-300 bracket?
Jesus, can you even read?
I am beginning to think you are trying to just wind people up and troll for no reason, you make a single sentence remark about a well thought out and well written post then you chose to ignore it and make a comment which was not even inferred in the post.
So yeah I'm calling Troll.
I thought it was a common knowledge that it's definitely going to be outside this bracket?
I was merely trying to qualify something before responding to the rest of it.
If you go back to his post here
http://forums.overclockers.co.uk/showpost.php?p=17942622&postcount=301
drunkenmaster said:
hes suggesting £300+ for the 6970 would be silly... so I'm a little confused as to where hes actually placing it unless I'm mis-reading it and he means both over £300 hence why I'm trying to qualify it.
If both 6950 and 6970 fell into £300+ bracket, a huge range where most profits come from, would remain untouched by AMD's offerings. We're talking £180-300 bracket here, as DM seemed to suggest.
I think you misread his post right there
As regards to me you best have link in what your referring to.
It always amuses me when people start talking about what pricing strategy will work best, and how cards should be priced to make the most money... In reality sales prices are determined by extremely complex mathematical models, with many different competing factors. There is no simple "right or wrong" answer, and those thinking in these terms are vastly underestimating the complexity of the problem.
In mathematical terms, sales pricing is incredibly complex: It's a multi-dimensional optimisation problem, where the number of dimensions (variables) is defined by the number of different competing factors (sales volume, manufacturing costs, marketing costs, market saturation, interaction with other products, competetor market share, product awareness/anticipation, value perception etc etc). The primary constriant is maximisation of overall revenue for the company (over ALL products), with secondary constraints related to consumer satisfaction and similar longer-term factors.
Now, in general, solving a multi-dimensional optimisation problems isn't too big of a deal. Even with tens of thousands of coupled variables you can obtain good estimate of an optimal value very quickly (using Monte-Carlo simulations etc). However, what makes sales pricing so complex (and interesting!) is that you have a lot of uncertainty. Not only do you have uncertainty in the estimates of the variables (which is easy to describe quantitatively), but you have uncertainty in the definition of the variables and constraints themselves!
Because of this uncertainty, each of the non-quantitative variables and constraints must be given a model for how it interacts with the other variables. These are generated through painstaking analysis of sales data, of consumer buying trends, and of whatever other information can be obtained. This data has a *massive* effect on how the sales pricing models work, and this is the reason why companies value customer feedback (surveys / questionaires etc) so highly in certain fields. Accurate sales analysis models are solid gold for anyone trying to mass-market a product...
Anyway, any large company will have a department of quantitative analysts performing these complex optimisation problems in order to generate the optimal pricing for a product. In principle it's really not too different from the work of quantitative analysts who work for investment banks, except for the differing timescales (share prices change hour-to-hour, whereas products are released or repriced every couple of months). So, if you are under the impression that product prices are plucked out of the air by high-level executives, or are selected based on only a few simple "facts" (such as those that are branded around here), then you massively underestimate the complexity of the problem. But hey - armchair analysts for the win on the internet, right?
In mathematical terms, sales pricing is incredibly complex: It's a multi-dimensional optimisation problem, where the number of dimensions (variables) is defined by the number of different competing factors (sales volume, manufacturing costs, marketing costs, market saturation, interaction with other products, competetor market share, product awareness/anticipation, value perception etc etc). The primary constriant is maximisation of overall revenue for the company (over ALL products), with secondary constraints related to consumer satisfaction and similar longer-term factors.
Now, in general, solving a multi-dimensional optimisation problems isn't too big of a deal. Even with tens of thousands of coupled variables you can obtain good estimate of an optimal value very quickly (using Monte-Carlo simulations etc). However, what makes sales pricing so complex (and interesting!) is that you have a lot of uncertainty. Not only do you have uncertainty in the estimates of the variables (which is easy to describe quantitatively), but you have uncertainty in the definition of the variables and constraints themselves!
Because of this uncertainty, each of the non-quantitative variables and constraints must be given a model for how it interacts with the other variables. These are generated through painstaking analysis of sales data, of consumer buying trends, and of whatever other information can be obtained. This data has a *massive* effect on how the sales pricing models work, and this is the reason why companies value customer feedback (surveys / questionaires etc) so highly in certain fields. Accurate sales analysis models are solid gold for anyone trying to mass-market a product...
Anyway, any large company will have a department of quantitative analysts performing these complex optimisation problems in order to generate the optimal pricing for a product. In principle it's really not too different from the work of quantitative analysts who work for investment banks, except for the differing timescales (share prices change hour-to-hour, whereas products are released or repriced every couple of months). So, if you are under the impression that product prices are plucked out of the air by high-level executives, or are selected based on only a few simple "facts" (such as those that are branded around here), then you massively underestimate the complexity of the problem. But hey - armchair analysts for the win on the internet, right?
I want a go at predicting what will come with 6950/70 so here goes. With increased knowledge of the 40nm process a small efficient dice (fantastic yield), low power usage and increase in clocks.
I think the 1920 shaders is fud and both the 6950 and 70 have 1536 shaders (remember 3850/70 both had 320 and 4850/70 both had 800), both giving gains over the 5850/70 (1440 vs 1600), not NV 580 performance but OK and leaving 6990 to deal with the very top end, pricing as follows;
6950 1GB - £249.99
6950 2GB - £279.99
6970 1GB - £299.99
6970 2GB - £329.99
6990 2GB - £449.99
6990 4GB - ?
Obviously the 6990 will not be out till Jan/Feb and later in the year, March'ish they will introduce a 6930 with reduced shaders to fill the gap between 6870 and 6950.
To answer the original allocation of 6950s being reduced, so good is the yield that they can put out more 6970s than 50s if they wish and are doing so therefore reducing the allocation of 6950s, I don't know any insiders and this has all come from my crazed/maddeningly frustrated mind at how good AMD are at keeping this **** secret till the very last moment, argggghhhhhhhhh!!!
had to have a go
J.
I think the 1920 shaders is fud and both the 6950 and 70 have 1536 shaders (remember 3850/70 both had 320 and 4850/70 both had 800), both giving gains over the 5850/70 (1440 vs 1600), not NV 580 performance but OK and leaving 6990 to deal with the very top end, pricing as follows;
6950 1GB - £249.99
6950 2GB - £279.99
6970 1GB - £299.99
6970 2GB - £329.99
6990 2GB - £449.99
6990 4GB - ?
Obviously the 6990 will not be out till Jan/Feb and later in the year, March'ish they will introduce a 6930 with reduced shaders to fill the gap between 6870 and 6950.
To answer the original allocation of 6950s being reduced, so good is the yield that they can put out more 6970s than 50s if they wish and are doing so therefore reducing the allocation of 6950s, I don't know any insiders and this has all come from my crazed/maddeningly frustrated mind at how good AMD are at keeping this **** secret till the very last moment, argggghhhhhhhhh!!!
had to have a go
J.