Overclock.net banner

[Barrons] Intel: Morgan Stanley Encouraged by Renduchintala’s Pragmatic Influence

10K views 143 replies 63 participants last post by  Fuell 
#1 ·
Article

Quote:
"While Intel still hasn't released many details yet on 10 nm, Murthy did say that 10 nm would substantially enhance performance along multiple axes. There will be better power efficiency for thin and light notebooks, which we think was the primary benefit of a 14 nm process defined during a period where the company was under siege from other form factors. But there will be a significant focus also on higher performance microprocessors for desktops and servers, both from higher instructions per clock but also in other key metrics. Our faith in 10 nm raising the bar for enthusiast PCs is why we see the threat presented by AMD's Zen as being fairly manageable, with only short term disruption in 2017."
 
#3 ·
Intel expects it's current mainstream parts to lose in overall performance to Zen (as should we all), but probably expects future mainstream hex cores to regain the lead.
 
#6 ·
Quote:
Originally Posted by Kommanche View Post

I'm worried that AMD will get the performance right, but the pricing wrong.
I'm worried that AMD will achieve parity, and then fall behind again the moment Intel is able to move to a new node.
 
#7 ·
I still don't care until we see performance numbers.
 
#8 ·
Quote:
Originally Posted by Blameless View Post

I'm worried that AMD will achieve parity, and then fall behind again the moment Intel is able to move to a new node.
most likely Intel will continue to milk their mainstream market and let AMD gain some market share, so it has some money to survive in the future. As soon as AMD is taking what Intel feel it is more than it should, Intel will release whatever they have been sandbagging for so many years.
 
  • Rep+
Reactions: ForNever
#9 ·
Quote:
Originally Posted by Nestala View Post

Article

From Intel president Dr. Venkata Murthy Renduchintala:
That's more or less admitting that Zen should be pretty good.
I would believe that intel doesn't feel threatened if they had managed to increase the CPU clocks since Sandy Bridge. As of yet, nothing they have produced has per-clock exceeded the i7 3960X's performance.

They have released newer nm scaled chips on a bi-yearly basis but still no increase in actual performance... so, knowing the way AMD improves hardware - Zen is going to tear up intel's house of cards.

That dog is gonna get wagged.
 
#11 ·
Quote:
Originally Posted by prjindigo View Post

Intel has had 3 new nodes since Sandy Bridge and 0 increase in performance per core.
Plenty of tasks see quite a large performance jump from Sandy to Skylake, but performance per core isn't the issue here; how many cores Intel feels comfortable sticking on mainstream parts and the sort of margins they are able to maintain is.

2017 is the first time AMD and Intel will be on similar nodes in quite a long time and this will be AMD's only real hope to return to profitability.

Intel hitting 10nm significantly sooner would let them sell CPU that would likely have better performance in all areas, for more money, while spending less to build them.
 
  • Rep+
Reactions: GamerusMaximus
#12 ·
Quote:
Originally Posted by Blameless View Post

Plenty of tasks see quite a large performance jump from Sandy to Skylake, but performance per core isn't the issue here; how many cores Intel feels comfortable sticking on mainstream parts and the sort of margins they are able to maintain is.

2017 is the first time AMD and Intel will be on similar nodes in quite a long time and this will be AMD's only real hope to return to profitability.

Intel hitting 10nm significantly sooner would let them sell CPU that would likely have better performance in all areas, for more money, while spending less to build them.
Most of the performance jumps we see are due to the newer instruction sets. There have only been 2 actual improvements in ipc, Haswell and Skylake.
 
#13 ·
The one thing that everyone seems to forget about when it comes to AMD vs intel is that with AMD the pcie lanes are all still on the chipset. So in theory, you should be able to get 40 lanes with a relatively cheap board.

That means you can run a couple gpus, an m.2 SSD, and have a couple of thunderbolt 3 connectors without much of an issue.
 
#14 ·
Quote:
Originally Posted by Blameless View Post

I'm worried that AMD will achieve parity, and then fall behind again the moment Intel is able to move to a new node.
That's possible, but it's not like AMD is going to stay on 14nm for more than ~two years. It's perfectly fine if AMD gets to the same node within a year of Intel. And Glofo 7nm is actually closer to Intel 10nm than Glofo 14nm is to Intel 14nm.
 
#15 ·
Quote:
Originally Posted by Marios145 View Post

Most of the performance jumps we see are due to the newer instruction sets. There have only been 2 actual improvements in ipc, Haswell and Skylake.
Even independent of instruction sets it's not hard to find performance improvements, and performance improvements are performance improvements, no matter where they come from.

Haswell and Skylake are the only significant architectural changes since Sandy, but even what's between them aren't totally devoid of IPC increases.
 
#16 ·
Quote:
Originally Posted by Marios145 View Post

Most of the performance jumps we see are due to the newer instruction sets. There have only been 2 actual improvements in ipc, Haswell and Skylake.
As should be the norm. Architectural improvements are made during the tock, or the actual new generation. The die shrinks generally don't receive architecture improvements, just lower power consumption and possibly higher clocks as a result. People forget that Ivy Bridge and Broadwell are not full generation parts, they are the half generation parts.
Quote:
Originally Posted by Tjj226 Angel View Post

The one thing that everyone seems to forget about when it comes to AMD vs intel is that with AMD the pcie lanes are all still on the chipset. So in theory, you should be able to get 40 lanes with a relatively cheap board.

That means you can run a couple gpus, an m.2 SSD, and have a couple of thunderbolt 3 connectors without much of an issue.
The HT link is a bottleneck, and the chipset basically acts like a PLX chip. AM4 should do away with the northbridge, and it makes no sense for AMD to keep the northbridge, considering that FM2 processors don't have them. It also adds additional cost and power consumption, a huge con for mobile. AM4's chipsets are no different from the PCH found on modern Intel platforms in terms of function.
 
#17 ·
Well i didn't express it right, while newer instruction sets are good, they tend to take some time to get implemented widely. That's the main reason bd never caught up with intel, zero improvements for legacy code, only support for newer instruction sets and most improvements were highly software-dependent.
Now i'm not saying that intel didn't improve IPC, but it only improved it by 5-10% each gen essentially by refining sandy-bridge conservatively.(if it ain't broke, don't fix it)
They could easily spend R&D money on a new arch which would provide 30-40% jump in ipc, maybe even higher, but it takes 4-5 years to make a new design and there was no competition to create this need.

I just hope that Zen will ignite the x86 cpu race once again.
 
#18 ·
Quote:
Originally Posted by Clocknut View Post

most likely Intel will continue to milk their mainstream market and let AMD gain some market share, so it has some money to survive in the future. As soon as AMD is taking what Intel feel it is more than it should, Intel will release whatever they have been sandbagging for so many years.
Which is what exactly?
Intel's mainstream offerings are not that impressive. Most i5's are locked and with lower clocks than their K versions. AMD has a real chance of successfully attacking these parts with unlocked 4c/8t and 6c/12t 14nm CPUs even if the single core performance would be 15% lower. Obviously this is what worries Intel.

Their FX CPUs apart from being very old had a lot of disadvantages which made them almost irelevant in these last 2 years but Zen sure will change that.
 
#19 ·
Quote:
Originally Posted by Tjj226 Angel View Post

The one thing that everyone seems to forget about when it comes to AMD vs intel is that with AMD the pcie lanes are all still on the chipset. So in theory, you should be able to get 40 lanes with a relatively cheap board.

That means you can run a couple gpus, an m.2 SSD, and have a couple of thunderbolt 3 connectors without much of an issue.
AMD already has done away with the northbridge, it was previously stated that on all am4 boards, including zen. That the cpus would have them all integrated just like on intel chips.
 
#20 ·
Quote:
Originally Posted by Marios145 View Post

They could easily spend R&D money on a new arch which would provide 30-40% jump in ipc, maybe even higher, but it takes 4-5 years to make a new design and there was no competition to create this need.
More IPC requires more ILP and most of the low-hanging fruit has been picked already in this regard.

I think the only way we are going to see dramatic increases in IPC are via more SMT, which of course won't help single threaded IPC very much.
 
#21 ·
Uhhh.

I mean, Bulldozer was disruptive. So was Bobcat, Llano and Trinity. They weren't any real threat, they just sort of...entered the market with a usable platform that allowed OEMs higher profit margins.

I don't think Intel sees AMD as any sort of threat. They are probably covering their bases to stockholders who are worried.
 
#22 ·
This guy is President of a Systems architecture group and nothing more, not the president of intel. And a threat in sales does not mean zen will out perform intels 6 or 8 core cpus.

And the title of this thread is click bait.

And this stuff does not have "" "" so its pretty clear the analyst is saying this stuff and not Renduchintala.

 
#23 ·
Quote:
Originally Posted by Nestala View Post

Article

From Intel president Dr. Venkata Murthy Renduchintala:
That's more or less admitting that Zen should be pretty good.
The quote that you are quoting appears to come from Moore (the analyst) rather than Renduchintala himself; the comments in the primary article which appear in italics are text excerpts from Moore.

As such, the analyst sees a threat from Zen. The president of "systems architecture" at Intel doesn't make any comment on Zen.
 
#24 ·
Quote:
Originally Posted by S.M. View Post

Uhhh.

I mean, Bulldozer was disruptive.
Your logic doesn't make sense, Bulldozer was like a present for Intel and actually managed to make them more successful.
If Zen would be the same kind of "disruptive" then there is no point in mentioning how will 10nm allow them to keep their lead.
 
#25 ·
Quote:
Originally Posted by Marios145 View Post

Well i didn't express it right, while newer instruction sets are good, they tend to take some time to get implemented widely. That's the main reason bd never caught up with intel, zero improvements for legacy code, only support for newer instruction sets and most improvements were highly software-dependent.
Now i'm not saying that intel didn't improve IPC, but it only improved it by 5-10% each gen essentially by refining sandy-bridge conservatively.(if it ain't broke, don't fix it)
They could easily spend R&D money on a new arch which would provide 30-40% jump in ipc, maybe even higher, but it takes 4-5 years to make a new design and there was no competition to create this need.
Agreed. AMD's implementation of top level SIMD ISA (AVX,XOP) was garbage and it was AMD's fault they horribly mutated those ISAs in their CPUs. I've read somewhere AVX on AMD BD/PD/EV was >2x slower than Intel's implementation in Sandy per clock. Despite the full compiler support from GCC, MSVC and others(except from ICC), AMD couldn't show any improvement at that time because of no-one but themselves.

I am extremely eager to know how Zen will perform now because they have declared that they are cutting many of their exclusive ISAs from it. Interesting thing is, what have they replaced them with ? Have they heavily invested the transistors for HSA ? Or they have reached near IPC's of latest Intel's AVX2(very unlikely) ? or are they providing more number of cores to compensate the stronger Intel cores ?

I've heard due to license agreement Intel has not allowed AMD to use AVX-512 and AVX-512 is already available in Server, yes it will take time to be utilized fully, but it is not new ISA, it is just more extended version of previous instructions. Anybody who knows to program for SSE/AVX should have almost no problem in using AVX-512 if there is problem domain available for it. The point is, in servers, Intel is already running on smooth path, without any obstacle and without giving any burden to programmers/compiler makers. How is AMD going to survive through all of this is interesting.
 
#26 ·
Quote:
Originally Posted by Dragonsyph View Post

This guy is President of a Systems architecture group and nothing more, not the president of intel. And a threat in sales does not mean zen will out perform intels 6 or 8 core cpus.

And the title of this thread is click bait.

Well it's not like Intel's most sold CPUs have 6 or 8 cores.
Also the guy is one of Intel's executives, certainly not just a simple guy from Intel that talks nonsense.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top