AMD CPU speculation... and expert conjecture

Page 330 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
You seem to believe that using defective silicon always generate revenue, which is not true. Plot the money from reusing that silicon against the money spent on fabricating, testing, packaging, storing, shipping... products. There is a minimum market/price-point beyond which you lost more money by making those products than by just rejecting the defective silicon. That is why AMD is not selling single cores, for instance.

For illustration: consider the extreme case you reuse that defective silicon ($100M) and fabricate product Y ($100M) which nobody purchases. You are loosing $200M, but if you ignore the defective silicon you only miss $100M.
Click to expand...

That's not remotely how binning works.

You create product A at a cost of $100M USD per batch. You would sell that batch at a premium for $200M USD. The batch has 50% yields meaning half the dies aren't good enough to sell for full price. now ordinarily you'd throw away that 50% and take the $50M loss (really $100M after profit is counted). During QA you determine that 80% of that batch is perfectly functional at a lower level, you then sell that defective batch at a discount with a different product name. You end up recouping your original production costs, sometimes even making a profit. Worst case scenario you have to sell it at a loss (under original production cost) and use the money to defray that sunk cost vs just eating the whole thing. Can also use the negative profit margin as a tax write off.

Did you know the FX4xxx and FX6xxx CPU's are just defective FX8xxx CPUs? They are all the exact same die. They come off the exact same production line. The chips with one defective module are labeled fx6xxx series, the ones with two defective modules are labeled fx4xxx series. Anymore then that and the die is considered useless and is scratched. APU's are no different, the lower model APU's are just defective dies from the top tier model.

Another aspect of binning is that often there is significantly more demand for the lower or midrange binned product then the final perfect one. To meet this demand the company will take it's perfect chips and deliberately mark them as the inferior product. Better to make a smaller profit off a chip you are selling then no profit off a chip sitting in the warehouse. Enthusiasts eventually found out about this practice and developed various tricks to unlock unlock the chips deliberately disabled potential. And thus the overclocking world was born, people turning Pentium 166's into Pentium 200's, AMD K6 250's to 300's and all the rest is history. Eventually that become an actual selling point via unlocked processors. They are only guaranteed to provide the rated performance but the manufacturer decided to not lock it and leave it up to the consumer to try for higher then rated.
 


30% over the equivalent part from PD, so 8 core steamroller would be 30% better than 8350, etc.

Intel equivalent, would probably be something like i7-4770k in some areas...8 core steamroller would likely be better than 4770k in many areas though.

If Steamroller is 30% faster than Piledriver, it will be roughly 49% faster than bulldozer....
 


They are talking per core performance at the same clock speed. IE CPU A's "core" can do 1,000 WIGOPS (Widget Operations) per second. CPU B's "core" would then have to do 1,300 WIGOPS per second to have a 30% increase in "IPC".

I honestly doubt we will see a 30% increase in IPC, more likely a 15~25% depending on work load. We could also see a clock speed bump though that remains to be seen.
 


15% increase in CPU performance is much more than you are giving credit for:

if PD is behind Intel IPC by 30%, that means it cut the gap in half.

If AMD can keep maximum overclock frequency high and Broadwell ends up taking another big bite out of overclocking headroom like we've seen since SB to IB transition, it should be feasible to have an AMD chip that can compete with an Intel chip in all workloads when both chips are at maximum overclock.

If SR can hit 5ghz, Broadwell can hit 4.3ghz. that would give SR a 16% clockspeed advantage, coupled with a 15% IPC increase. SR has the potential to be right behind Intel in single thread and probably win very aggressively in multi-thread.

Now if AMD skips SR and goes to Excavator, and Excavator is that crazy leaked die shot we saw earlier and AMD could get 15% out of just changing the front end of PD, then I would be surprised if AMD did not have a superior chip to Intel's quads with HT.

The only problem is that if they do not shrink nodes and do not use HDL that they will still have a 300mm^2 die on their hands. IB-E hex core is about 215mm^2. With a die that big, Intel could just lower the price of hex core and then walk away from 4m/8c SR.

I can see why AMD is not making a move on HEDT. Even if they beat Intel quad with HT in raw performance, at 32nm they will not be able to compete with die size to profit ratio compared to Intel. IMHO AMD HEDT is dead until they can get die size down. Even if they have an architecture that can compete with Haswell quad with HT right this very second in raw performance, AMD would be putting a 300mm^2+ chip against 181mm^2 Haswell.

AMD would be in Nvidia's position in regards to GK110 and Hawaii. Slower chip that is also bigger. However AMD is not like Nvidia and they will just take smaller profits.
 


I can agree with a lot of that thought process...though with a half node shrink to 28nm, they should be able to get die size down. Still not a situation like Intel has, though the full node shrink from SB to IB went from 216mm^2 to 160mm^2. From that, we can extrapolate that a half node shrink would put the die size somewhere around ~285-290mm^2.

That would put the 8 core FX based on SR at a die size roughly 20nm larger than the Intel hex-core IB-E platform (265mm^2), which isn't terrible...especially considering the fact that IB-E is an entire half node shrink smaller than the 28nm process for steamroller.
 
So, I just thought I'd say BF4 likes fast RAM. You can gain 10-15 FPS over 1600Mhz Memory by using 2400Mhz Memory. Looks like AMD is going to start using optimization with games. Well, not that they haven't already been doing so... I think We're going to see some huge performance gains out of MANTLE. I look forward to buying a r9 280X.
 
Looks like we will have a review of the 290x tonight and the 290 on the 31st.

I predict the 290 will be competitive with the 780(but cheaper) and the 290x will be a titan beater while still being cheaper. Man i love Nvidia vs Amd actually competitive
 


I love seeing AMD make NVidia and their *WAY* overpriced cards look stupid.
 
Before the review even comes out i'll say i predict a hot running video card however look at the power consumption of a 280x and then compare it to a 770. 770 is around 5% faster while consuming 30% less power on load.(Techpowerup)

A 780 is 20% faster than a 280x on average while using 10% less power(techpowerup) And the titan is 10% faster then a 780 while using less power then one so Amd might be faster but i think its going to come at a cost of higher power consumption and higher temps or a louder card
 


What your seeing is a combination of global inflation and free market forces. They are priced that high because people will pay that much, there is a demand for it. Those cards also have insane profit margins yet have low volume. They are the Lamborghini's and Porsche's of the GPU world.

I just bought 2 GTX 780 Hydros along with a 900D while swapping out quite a bit of WC loop components (fittings / hose / pump / ect..).
 


Then you'd never be buying anything as there is always something "coming out soon". For a long time my dual GTX 580 Hydro's served they're purpose. Recently they stopped being sufficient for what I do so I've decided to replace them. And since I'll be draining and redoing my loop now is a good time to upgrade the entire kit. These cards should last me another three or so years before being replaced. Sometime in the next year or so I will be swapping out my CPU, might even go Intel if AMD doesn't have a product to replace my FX8350.
 
When i know something is actually coming out soon i never buy it for example i knew the 280x was coming out soon so why buy a 7970ghz edition for 400$. Soon to some people may mean different things but to me when it comes to video cards it means 1-2 months or so, unless i need something right now.

But its not like Amd is really going to come out with a superior product in terms of performance its just you could of saved money for some new games or something else. 2 780's will last quite some time you should have no issues for 3+ years.
 
Its like i have magic and that i use that magic to see into the future. Lol

290x beats the titan more than it loses but the issue with heat and noise keep it from being as attractive if if was priced the same. Does anyone here ever buy video cards with that type of cooling i also get a Sapphire with a proper heat-sink and fan.

 
95C is max it will ever be, the longevity probably isn't a problem. This is the first card that runs at the same temperature in furmark as in games. Its just if you have bad cooling, it will throttle your performance down to sub 780 levels.
 
I still recommend anyone to wait until the better heat-sink and fans are out i have no idea why Amd persist on staying with this type of cooling. For just 50$ more they could of had a better card in all fronts except in performance per watt and even then its not to bad.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom