Are AMD processors worth considering for mid to high end?

[ThatBuzzkiller]

Crappy yield.

Yields only factor in how many chips will succeed in working, not when it comes to performance ...

 

[Raftina]

I'm not even sure how Intel managed to make the 6700K have lower turbo and slightly higher TDP than the 4790K when their 14nm transistors have improved capacitance ...

That is quite simple: Intel designed the chip for the 65W i7-6700 to offer improved performance over the 84W i7-4790.

That Intel is treating your market segment as an afterthought does not mean it is not making progress in CPU design.

 

[LTC8K6]

I'm not even sure how Intel managed to make the 6700K have lower turbo and slightly higher TDP than the 4790K when their 14nm transistors have improved capacitance ...

HD530 graphics might account for extra watts in the TDP spec?

 

[LTC8K6]

Comparing the 6700 to the 4790, the wattage jump for the 6700K is odd.

The 4790 to the K was 4 watts TDP. 400mhz base increase / 400mhz turbo increase.
the 6700 to the K is 26 watts TDP. 600mhz base increase / 200mhz turbo increase.

 

[LTC8K6]

The 6700K doesn't really do turbo at all compared to the Haswell chips.

It runs at the same speed, except on 1 core, when it can hit 4.2ghz.

 

[NTMBK]

Comparing the 6700 to the 4790, the wattage jump for the 6700K is odd.

The 4790 to the K was 4 watts TDP. 400mhz base increase / 400mhz turbo increase.
the 6700 to the K is 26 watts TDP. 600mhz base increase / 200mhz turbo increase.

I'd love to see a clock speed/voltage curve for 14nm. Can we start a fund to buy IDontCare a Skylake system?

 

[Erenhardt]

Keep up the great thread crap an you will be paid fistful.

 

[ShintaiDK]

Yields only factor in how many chips will succeed in working, not when it comes to performance ...

Yield is also used as a factor to describe how many chips are within the useable boundary. A 1Ghz A9 Apple chip for example is a yield loss to make it obvious.

 

[Arachnotronic]

Yields only factor in how many chips will succeed in working, not when it comes to performance ...

If only a handful of them can actually hit the higher frequencies at acceptable voltage/power, then it's not viable to sell SKUs clocked higher. Parametric yield, baby...

 

[MiddleOfTheRoad]

It's simple when you are a monopoly. You just pump out a new one whenever you get about a 5% gain. Ohh and just go ahead and create a new socket too. MB sales have been rather low as of late.

Yeah, that's the thing that people on this forum don't seem to grasp. Intel has stepped on the brakes on performance because they can. Might as well slow it down and maximize profits for each generation when there is no performance competition. There are limits to how small they can shrink the dies with today's technology -- and it's actually smart on Intel's part stretch out before they hit the physical brick wall. Many scientists don't think we can shrink below 5nm -- and Intel's already working on getting to 10nm.

http://www.geek.com/chips/theoretical-physicist-explains-why-moores-law-will-collapse-1486677/

 

[myocardia]

Many scientists don't think we can shrink below 5nm -- and Intel's already working on getting to 10nm.

Whether or not we will ever be able to go smaller than 5nm, it will still be a very, very long time before any of us will be able to purchase our first 5nm CPU, I'm willing to bet.

 

[Magic Carpet]

It seems it can use a few

I see. Well, godspeed, can't wait for your review

 

[MrTeal]

Yeah, that's the thing that people on this forum don't seem to grasp. Intel has stepped on the brakes on performance because they can. Might as well slow it down and maximize profits for each generation when there is no performance competition. There are limits to how small they can shrink the dies with today's technology -- and it's actually smart on Intel's part stretch out before they hit the physical brick wall. Many scientists don't think we can shrink below 5nm -- and Intel's already working on getting to 10nm.

http://www.geek.com/chips/theoretical-physicist-explains-why-moores-law-will-collapse-1486677/

Undoubtedly Intel could stagnate their increases since anyone choosing to buy a computer today is pushed to them by AMD's lack of competitiveness. Intel's problem isn't capturing sales, it's driving sales. A 5% increase per year means it will take 14 years to double performance. At work my company has already moved from a 3 year replacement cycle to a 5 year cycle. We're already getting close to the point where people replace machines for reliability reasons rather than performance reasons.

If Intel could maintain their margins and put out a Skylake chip that was 20% faster than Haswell clock for clock, clock higher and consume less power, you've got blinders on if you think they'd hold it back to 5% just because there's no competition. The desktop market isn't really growing, and they're not competing with AMD for new sales. They're competing with Nehalem, SB and Bulldozer to create enough incentive to replace existing systems with new Intel product.

 

[VirtualLarry]

At work my company has already moved from a 3 year replacement cycle to a 5 year cycle. We're already getting close to the point where people replace machines for reliability reasons rather than performance reasons.

I'm not in corp. IT, but I was under the impression that this was already the case, that PCs were replaced when the warranty or lease was up, because of downtime concerns due to potential reliability issues.

If Intel could maintain their margins and put out a Skylake chip that was 20% faster than Haswell clock for clock, clock higher and consume less power, you've got blinders on if you think they'd hold it back to 5% just because there's no competition. The desktop market isn't really growing, and they're not competing with AMD for new sales. They're competing with Nehalem, SB and Bulldozer to create enough incentive to replace existing systems with new Intel product.

But then, if they blow their 20% improvement in one go, what are they going to do for their NEXT chip?
You had better believe that they're 'pacing' their improvements, in steps.

 

[ShintaiDK]

They are not holding anything back. That tinfoil stuff needs to go.

The game is performance/watt.

 

[VirtualLarry]

They are not holding anything back. That tinfoil stuff needs to go.

So... the fact that every generation since Core2 has shown roughly a 5-8% improvement, is... coincidental?

The game is performance/watt.

So, why haven't they dropped Core entirely, and gone with an improved Atom?

 

[Arachnotronic]

Yeah, that's the thing that people on this forum don't seem to grasp. Intel has stepped on the brakes on performance because they can. Might as well slow it down and maximize profits for each generation when there is no performance competition. There are limits to how small they can shrink the dies with today's technology -- and it's actually smart on Intel's part stretch out before they hit the physical brick wall. Many scientists don't think we can shrink below 5nm -- and Intel's already working on getting to 10nm.

http://www.geek.com/chips/theoretical-physicist-explains-why-moores-law-will-collapse-1486677/

Your understanding of the semiconductor business really leaves much to be desired, MiddleOfTheRoad.

 

[VirtualLarry]

Your understanding of the semiconductor business really leaves much to be desired, MiddleOfTheRoad.

I'm not so sure he's wrong. After all, Intel has been selling smaller and smaller dice, for the same price, on the market. If they really cared about pushing performance, they would have put (equivalent) silicon area to use, for performance sake, and not shrunk their dice, to make more profit.

 

[ShintaiDK]

I'm not so sure he's wrong. After all, Intel has been selling smaller and smaller dice, for the same price, on the market. If they really cared about pushing performance, they would have put (equivalent) silicon area to use, for performance sake, and not shrunk their dice, to make more profit.

Is this turning into another infamous 6-8 core demand? While ignoring all downsides? He is wrong.

So... the fact that every generation since Core2 has shown roughly a 5-8% improvement, is... coincidental?

Try chart the IPC graph from Banias and up.

So, why haven't they dropped Core entirely, and gone with an improved Atom?

Because Core offers better performance/watt.

 

[VirtualLarry]

Is this turning into another infamous 6-8 core demand? While ignoring all downsides? He is wrong.

I said nothing about additional cores, don't put words in my mouth. I was suggesting to use the additional silicon area to improve performance, rather than conserve silicon area for increased profits. It used to be, CPU companies would use shrinks, to increase the number of transistors per silicon area, and the silicon area of most CPUs, per the prices charged for them, were relatively constant. But in Intel's case, their CPUs became barely better performing, for the same price, but with drastically reduced silicon area (MOAR PROFITS!!!).

Edit: In other words, Intel has been primarily interested in profit-taking, rather than actually improving performance.

 

[ShintaiDK]

They should just run at a loss right? I mean business is pro bono after all. Smaller dies is there for a reason that I doubt you will be willing to understand. Perhaps you should look at transistor budget instead.

Also you still forget the performance/watt part. Unless you add IGP its not going to change much if anything.

 

[Arachnotronic]

I'm not so sure he's wrong. After all, Intel has been selling smaller and smaller dice, for the same price, on the market. If they really cared about pushing performance, they would have put (equivalent) silicon area to use, for performance sake, and not shrunk their dice, to make more profit.

Silicon area gets more expensive each generation as capital intensity goes up, so what you are saying is that Intel should basically sell product at the same price while eating an increased cost structure? You realize this is not economically viable, right?

 

[VirtualLarry]

They should just run at a loss right? I mean business is pro bono after all.

Also you still forget the performance/watt part. Unless you add IGP its not going to change much if anything.

Again, with the putting words in my mouth. Where did I say, that selling equivalent silicon area for equivalent prices over time, would result in Intel selling for a loss? It's a fact, that Intel's CPUs are getting smaller, over time, yet, their prices are increasing.

 

[ShintaiDK]

Again, with the putting words in my mouth. Where did I say, that selling equivalent silicon area for equivalent prices over time, would result in Intel selling for a loss? It's a fact, that Intel's CPUs are getting smaller, over time, yet, their prices are increasing.

Price is pretty much inflation only. So forget that part. And if we are to use your Core 2 argument. Then prices haven't been lower!

Dies are smaller due to increased cost and lower volume. Welcome to business 101. If you dont think all other companies are going this route, then you are naive at best.

 

[VirtualLarry]

Silicon area gets more expensive each generation as capital intensity goes up, so what you are saying is that Intel should basically sell product at the same price while eating an increased cost structure? You realize this is not economically viable, right?

Wasn't it true, that the last few quarters, Intel's desktop CPU volume was down, but ASP's were up?