AMD Wants To Stop Being Known As The “Cheaper Solution”

[Dave2150]

Why would Intel give away sales like that?

They make much more profit from the much smaller quadcore dies, than they do from the much larger xeon dies.

For example (random numbers) a single die could make either 250 quadcore's, or 100 8, 12 core xeon dies. The xeon dies are then harvested into the 6, 8, 12 core dies etc.

It's much much cheaper for Intel to spew out the quadcore dies, I can't imagine they'll ever change that philosophy, unless AMD gained significant market share (which would take years and be unlikely).

 

[AtenRa]

If Zen isn't a game changer for AMD then AMD's days as a CPU vendor will virtually end. They will be stuck in the low price commodity market with no exit (except shutting down their CPU business and becoming a different kind of company). 'Good enough' apparently isn't, as far as OEMs are concerned.

Havent you noticed already ??? AMD is not a PC CPU only company anymore.

After 2016 they could be profitable and make more Revenue than today even without selling a single CPU in the PC Desktop market.

 

[VirtualLarry]

Havent you noticed already ??? AMD is not a PC CPU only company anymore.

After 2016 they could be profitable and make more Revenue than today even without selling a single CPU in the PC Desktop market.

They should start selling Jukeboxes for bars, with AMD inside. Use their APUs with TrueAudio inside, and allow the user to save and load their favorite mixer settings for songs.

If they got a percentage of revenue from the jukeboxes, as well as the component costs... hey, there's a new revenue stream! Can't be worse than re-branding RAM and SSDs, and would get the AMD name out in front of the "working man".

 

[cbn]

I think Intel will still be stubborn, and keep their 6 and 8 cores reserved for the HEDT (enthusiast) platform, even if Zen 8 core mainstream CPU's completely outperform the mainstream quadcore I7's.

I think at 14nm Intel's HEDT platform will move on to 12 cores or more for the smallest die (Not sure whether we will see this at the level of Broadwell or Skylake though).

That means a hexcore will have half the cores disabled.

That is not terribly efficient.

Instead I think Intel will use a smaller multicore die to service the needs of hexcore and octocore enthusiasts. (eg, making a consumer version of Xeon-D for Skylake)

 

[rtsurfer]

I think at 14nm Intel's HEDT platform will move on to 12 cores or more for the smallest die (Not sure whether we will see this at the level of Broadwell or Skylake though).

That means a hexcore will have half the cores disabled.

That is not terribly efficient.

Instead I think Intel will use a smaller multicore die to service the needs of hexcore and octocore enthusiasts. (eg, making a consumer version of Xeon-D for Skylake)

Considering SandyBridge-E based 6 core 3960X had 2 cores disabled & it took like ~ 3-4 years till we got a 8 Core chip with Haswell-E.

Dream on.. Not gonna happen in this decade.

 

[VirtualLarry]

Considering SandyBridge-E based 6 core 3960X had 2 cores disabled & it took like ~ 3-4 years till we got a 8 Core chip with Haswell-E.

Dream on.. Not gonna happen in this decade.

It's a shame, but Intel really is THAT greedy.

 

[cbn]

Considering SandyBridge-E based 6 core 3960X had 2 cores disabled & it took like ~ 3-4 years till we got a 8 Core chip with Haswell-E.

Dream on.. Not gonna happen in this decade.

Sandy Bridge E5 Xeon came in two die sizes: 4C and 8C
Ivy Bridge E5 Xeon came in three die sizes: 6C, 8C, 12C
Haswell E5 Xeon comes in three die sizes: 8C, 12C, 18C

That is a very rapid progression in cores count amounting to a 50% increase in cores at the top end and a two core increase at the smallest die size for each generation.

So why do you think 12 cores for the smallest die size (for either Broadwell or Skylake) is dreaming?

Any why do you think Intel would not base a hexcore on Xeon-D type die?

 

[rtsurfer]

Sandy Bridge E5 Xeon came in two die sizes: 4C and 8C
Ivy Bridge E5 Xeon came in three die sizes: 6C, 8C, 12C
Haswell E5 Xeon comes in three die sizes: 8C, 12C, 18C

That is a very rapid progression in cores count amounting to a 50% increase in cores at the top end and a two core increase at the smallest die size for each generation.

So why do you think 12 cores for the smallest die size (for either Broadwell or Skylake) is dreaming?

I don't doubt Intel's ability to increase core count on the server side.

What I am saying is that we enthusiasts will not see the benefits of those increases even if we invest in the more expensive Extreme platform.

They'll continue to sell us defective server dies with cores disabled.

Edit:- Also if you look at the leaked Skylake slides, nothing points to us getting a 6 core on the mainstream platform, let alone 8 cores. Maybe the gen after it we get a 6 core.

Sorry I partially misunderstood your initial post. I though you were saying we would get 12 Core Xtreme editions, which isn't gonna happen for a while.

 

[cbn]

What I am saying is that we enthusiasts will not see the benefits of those increases even if we invest in the more expensive Extreme platform.

The price of cores will drop.

i7 980X (Gulftown) hexcore was $999 and now the i7-5820K (which is much faster) is $389.

Hexcores will continue to get cheaper, but a point comes when Intel will use a different die.

 

[rtsurfer]

The price of cores will drop.

i7 980X (Gulftown) hexcore was $999 and now the i7-5820K (which is much faster) is $389.

Hexcores will continue to get cheaper, but a point comes when Intel will use a different die.

Look at my edit.
6 core on mainstream after Skylake probably.

 

[cbn]

Edit:- Also if you look at the leaked Skylake slides, nothing points to us getting a 6 core on the mainstream platform, let alone 8 cores. Maybe the gen after it we get a 6 core.

If we ever saw such a platform it would certainly be separate from LGA 1151. (re: the current Xeon-D is SoC, so a Skylake-D would also be a SoC).

Then after Cannonlake the two LGA platforms would merge into one socket (with quad core + iGPU and 8C dies sharing the same platform much in the way as AMD's AM4).

 

[rtsurfer]

If we ever saw such a platform it would certainly be separate from LGA 1151. (re: the current Xeon-D is SoC, so a Skylake-D would also be a SoC).

Then after Cannonlake the two LGA platforms would merge into one socket (with quad core + iGPU and 8C dies sharing the same platform much in the way as AMD's AM4).

But where would they position it.?

Wouldnt that just be Skylake-E or Broadwell-E.?

They aren't gonna increase the core count on the Xtreme edition. If they price an 8 core below $1000 it will undercut their Xtreme edition. Above it, its meaningless, no one will buy it.

Makes no sense.

 

[cbn]

But where would they position it.?

Wouldnt that just be Skylake-E or Broadwell-E.?

It would be separate from that platform.

Skylake-D : 8C die
Skylake-E : 12C or 14C die

They aren't gonna increase the core count on the Xtreme edition. If they price an 8 core below $1000 it will undercut their Xtreme edition. Above it, its meaningless, no one will buy it.

Makes no sense.

You don't think Skylake-E (consumer extreme edition) will go beyond an octocore?

I think it definitely would.....however 12 cores as the minimum size die (for the E5 Xeons upon which it is based) is the tipping point IMO. If they go fourteen cores as the minimum die size for Skylake E5 then I definitely don't expect to see hexcores based on a 14 core die. Intel would logically use a smaller die instead (and the current Xeon-D 8C die for Broadwell is pretty small, so Skylake Xeon-D should also be small).

 

[rtsurfer]

It would be separate from that platform. Call it Skylake-D (because it would likely be based on a very small Xeon-D SoC type die).



You don't think Skylake-E (consumer extreme edition) will go beyond an octocore?

I think it definitely would.....however 12 cores as the minimum size die (for the E5 Xeons upon which it is based) is the tipping point IMO. If they go fourteen cores as the minimum die size for Skylake E5 then I definitely don't expect to see hexcores based on a 14 core die. Intel would logically use a smaller die instead (and the current Xeon-D 8C die for Broadwell is pretty small, so Skylake Xeon-D should also be small).

Gulftown to Haswell. That's how long it took for us to go from 6 cores to 8.

Increasing the core count in Skylake-E would be equivalent to halving the previous transition time. Skylake-E wouldn't be here till sometime in 2017.

Considering we are long overdue for a core count increase on mainstream platform, I will put it as a strong maybe. A 10 core extreme edition, I don't know.

 

[cbn]

Considering we are long overdue for a core count increase on mainstream platform.

Looking at the specs for the current Xeon-D, IMO it looks like mainstream platform from a high level:

24 PCIe lanes (Xeon-D) vs. 40 PCIe lanes (Xeon E5/HEDT)
6 SATA (Xeon-D) vs. 10 SATA (Xeon E5/HEDT)
dual channel memory (Xeon-D) vs. quad channel memory (Xeon E5/HEDT)

So a Skylake-D seems like a believable first step for a "mainstream level" Intel chip beyond four cores.

P.S. If Intel ever launched a Skylake consumer version of that chip I would imagine motherboard cost would be lower than X99 since Xeon-D is a SoC.

 

[CHADBOGA]

Havent you noticed already ??? AMD is not a PC CPU only company anymore.

After 2016 they could be profitable and make more Revenue than today even without selling a single CPU in the PC Desktop market.

What magical items will AMD be selling that could bring about this fanciful scenario?

 

[R0H1T]

What magical items will AMD be selling that could bring about this fanciful scenario?

Zen for servers, I thought that'd be pretty obvious right? It depends on a lot of predictions to come true of course, but if it doesn't then AMD is screwed anyway cause the desktop market & GPU's will not save them & that's a fact

I'm ignoring their semi-custom & Console endeavors for now because the desktop segment is bleeding more money than what the other small(er) segment are making, excluding the GPU's.

 

[SAAA]

Destkops aren't servers, you need a tighter balance between ST and MT performance. If AMD go gung-ho on core numbers then this will be reflected in lower clocks, with correspondingly lower ST performance.

IMHO all you need is a high enough turbo on these cpus, regardless of core count. If you have 100+W TDPs and aren't loading more than 4 cores why it shouldn't be possible?

Also remember their 220W TDP FX that works at 4.7GHz base, we know that Zen will be >40% better for clock (it's over excavator the statement) so at 3.36GHz maximum you already have its single thread performance.
Implement turbo up to 4GHz and it will be within 20% of Skylake quad core but with 4 more cores and better single than a 5.5GHz Piledriver.

 

[ShintaiDK]

I partially agree.
6C/12T can be a "waste" for casual mainstream workloads, but they can offer those cores on the mainstream platform for those who need it, instead of forcing people on the more expensive HEDT platforms.

Their profit margins also have something to do with it IMO.

I think the problem is the classic what you want contra what the 99% crowd wants. Products we can dream up wanting on Anandtech is one thing, products that will actually sell in desired volumes is entirely different.

Without servers the HEDT platform wouldnt even exist. Thats how small it is to begin with.

 

[ShintaiDK]

They make much more profit from the much smaller quadcore dies, than they do from the much larger xeon dies.

For example (random numbers) a single die could make either 250 quadcore's, or 100 8, 12 core xeon dies. The xeon dies are then harvested into the 6, 8, 12 core dies etc.

It's much much cheaper for Intel to spew out the quadcore dies, I can't imagine they'll ever change that philosophy, unless AMD gained significant market share (which would take years and be unlikely).

Not true at all. Manufactoring cost isnt the issue at all. Hell, they could even make one without IGP at all and be cheaper than the current ones to manufactor. The problem is what people want. It doesnt make sense to make a hexcore for LGA115x if its bandwidth starved and will only sell a limited amount. While overall perform worse than quadcores.

It seems some people mainly want 6-8 cores because 6-8 is a bigger number than 4. Not because their software or needs actually support it.

 

[rtsurfer]

I think the problem is the classic what you want contra what the 99% crowd wants. Products we can dream up wanting on Anandtech is one thing, products that will actually sell in desired volumes is entirely different.

Without servers the HEDT platform wouldnt even exist. Thats how small it is to begin with.

While it is true that HEDT is a minimal part of the market.

Let's not forget that they are make money out of defective dies that otherwise they would have to "possibly" throw away.

Its wrong to think of it as Intel doing us a favour, they do it because it makes economical sense for them.

 

[ShintaiDK]

While it is true that HEDT is a minimal part of the market.

Let's not forget that they are make money out of defective dies that otherwise they would have to "possibly" throw away.

Its wrong to think of it as Intel doing us a favour, they do it because it makes economical sense for them.

The 5960X isnt defective. And you forget most chips dont have cores disabled because they dont work. But because TDP or speed of the chip cant get high enough. And lets not fool ourselves. Lets look on Xeons frequencies and TDP:

http://ark.intel.com/products/family/78583/Intel-Xeon-Processor-E5-v3-Family#@Server

http://ark.intel.com/products/family/78585/Intel-Xeon-Processor-E7-v3-Family#@Server

Not so much fun anymore is it? Even Intel sells quadcore higher clocked Xeons where ST matters.

The fastest octocore you can buy for money is 3.2/3.6Ghz. While you can buy relatively cheap 4/4.4Ghz quadcores.

 

[cbn]

Let's not forget that they are make money out of defective dies that otherwise they would have to "possibly" throw away.

That is true and no doubt Intel will continue to do that.

I just wonder when it ends for the hexcores?

At the moment Intel has two hexcores: One "light version" (i7-5820K) based on 28 PCIe lanes and one "full version" (i7-5930K) with 40 PCIe lanes.

So with Intel seeing a market for two levels of hexcores do they decide to bifurcate the platform?

 

[ShintaiDK]

Those 2 hexcores are also clocked higher than any octocore. The elephant in the room.

 

[rtsurfer]

The 5960X isnt defective. And you forget most chips dont have cores disabled because they dont work. But because TDP or speed of the chip cant get high enough. And lets not fool ourselves. Lets look on Xeons frequencies and TDP:

http://ark.intel.com/products/family/78583/Intel-Xeon-Processor-E5-v3-Family#@Server

http://ark.intel.com/products/family/78585/Intel-Xeon-Processor-E7-v3-Family#@Server

Not so much fun anymore is it? Even Intel sells quadcore higher clocked Xeons where ST matters.

The fastest octocore you can buy for money is 3.2/3.6Ghz. While you can buy relatively cheap 4/4.4Ghz quadcores.

So..
This time it isn't a disabled part but a just a Server Xeon with Server stuff disabled. Doesn't wipe away what they did in the past.

Those 2 hexcores are also clocked higher than any octocore. The elephant in the room.

http://ark.intel.com/products/82765/Intel-Xeon-Processor-E5-1650-v3-15M-Cache-3_50-GHz

Matches this
http://ark.intel.com/products/81900/Intel-Xeon-Processor-E5-2643-v3-20M-Cache-3_40-GHz


Don't get me wrong, I don't have some anti-Intel agenda here. Just stating that HEDT is still a profitable market for them.