AMD Market Share Gains Accelerate in Desktop PCs, Servers and Mobile

[evald28]

Just one question. By share in market do you mean share in current sales or share in active/working PC units?

 

[JamesSneed]

It is hard to believe that OEM are still dragging their feet in adopting AMD CPUs. What are they going to do when AMD is on 7nm and Intel on 14nm+++?

The nm process size is not comparable between different companies. TSMC's 7nm process yields at 83 million transistors per mm^2. Intel's 14nm process yields 37.5 million transistors per mm^2, but their 10nm process yields 101 million transistors per mm^2. So Intel's 10nm process (if they can ever get it working reliably) is actually smaller than TSMC's 7nm process.

https://en.wikipedia.org/wiki/7_nanometer#7_nm_process_offerings
https://en.wikipedia.org/wiki/10_nanometer#10_nm_process_nodes

One Caveat ... Considering how tight Intel's 10nm was initially planned, how long and hard they struggled with failure, and the wiggle room they have to still make a 10nm claim ... I am not 100% certain that what they set out to achieve for 10nm will be what they produce in the end.

There 10nm+ won't be as dense to allow higher clocks. Intel overshot on how dense they could go using quad patterning.

Think of a stencil of the letter B. You set it down on some paper then spray paint it, pick it up, set it back down spray paint it, repeating this for a total of 4 times. You get slightly fuzzy edges on the letter B because you are not perfect every time you place the stencil. Similar thing here and those little defects cause heat as you ramp up frequencies which is why you only see laptop CPU's at low frequencies on Intel's 10nm. You really need to switch over to EUV to get the nice clean lines at these nm wavelengths. Nobody was sure how far you could go until Intel hit a wall. So Intels 10nm+ will be a little less dense to be able to make high frequency parts and they will be pushing hard to get EUV working for the future. Intels 7nm on EUV is likely going to be worlds best process when it comes out so I totally expect them to catch up or slightly lead going in another 2 years.

 

[robapacl]

When I was designing computer circuits back in the 80's, AMD had the best data and the best CS. Any parameter not spec'd that was measurable was available by phone in as short a time as feasible. I thought of them as the ones who kept Intel honest, unlike the software. The disaster of a single supplier, viz MS, doesn't need duplication in the hardware sector. And in a non-corrupt society, MS would be "one of the many".

 

[pjeirefretter]

Intels' trash is now "trash" and not a downgraded crippled high end processor, it will affect their overall margin much more then what they would like you to believe.

 

[joeblowsmynose]

It is hard to believe that OEM are still dragging their feet in adopting AMD CPUs. What are they going to do when AMD is on 7nm and Intel on 14nm+++?

The nm process size is not comparable between different companies. TSMC's 7nm process yields at 83 million transistors per mm^2. Intel's 14nm process yields 37.5 million transistors per mm^2, but their 10nm process yields 101 million transistors per mm^2. So Intel's 10nm process (if they can ever get it working reliably) is actually smaller than TSMC's 7nm process.

https://en.wikipedia.org/wiki/7_nanometer#7_nm_process_offerings
https://en.wikipedia.org/wiki/10_nanometer#10_nm_process_nodes

One Caveat ... Considering how tight Intel's 10nm was initially planned, how long and hard they struggled with failure, and the wiggle room they have to still make a 10nm claim ... I am not 100% certain that what they set out to achieve for 10nm will be what they produce in the end.

There 10nm+ won't be as dense to allow higher clocks. Intel overshot on how dense they could go using quad patterning.

Think of a stencil of the letter B. You set it down on some paper then spray paint it, pick it up, set it back down spray paint it, repeating this for a total of 4 times. You get slightly fuzzy edges on the letter B because you are not perfect every time you place the stencil. Similar thing here and those little defects cause heat as you ramp up frequencies which is why you only see laptop CPU's at low frequencies on Intel's 10nm. You really need to switch over to EUV to get the nice clean lines at these nm wavelengths. Nobody was sure how far you could go until Intel hit a wall. So Intels 10nm+ will be a little less dense to be able to make high frequency parts and they will be pushing hard to get EUV working for the future. Intels 7nm on EUV is likely going to be worlds best process when it comes out so I totally expect them to catch up or slightly lead going in another 2 years.

The 10nm chips that they have made also seem to want to draw more power than they should - no less or even more than 14 nm equivs. I suspect that is the result of having to pump in more voltage just to get reasonably usable frequencies.

 

[PaulAlcorn]

Just one question. By share in market do you mean share in current sales or share in active/working PC units?

The share of unit sales during the quarters listed.

 

[daddywalter]

It used to be said, "Nobody ever got fired for recommending IBM". Same holds true today for Intel; AMD is making inroads with OEMs, but it's slow going. Intel's own production problems are helping AMD, of course, but many decision-makers are probably expecting Intel to fix those and possibly come out with something that leapfrogs Ryzen; they don't want to risk favorable relationships with Intel just yet.

The real test may come with successors to the Core and Ryzen platforms. If AMD continues to provide more bang for the buck, it could supplant Intel as the processor-maker of choice. (And it doesn't hurt that motherboards for AMD *also* usually cost less than motherboard for Intel processors.)