With the current rate of Intel CPU performance increases, could AMD be catching up?

[Fjodor2001]

So the last few mainstream desktop CPU generations from Intel (Haswell, IB, and perhaps to some degree SB) have not been that focused on increasing CPU performance. We're seeing about 5-10% IPC improvements, perhaps 100 MHz frequency increase, and no increase in the number of cores going from one generation to the next.

Does this mean that AMD now has a chance to catch up performance wise? Has the performance of the mainstream desktop AMD CPUs increased more than the corresponding Intel CPUs in the last few years? And what can we expect looking forward 1-3 years?

 

[mikeymikec]

Does this mean that AMD now has a chance to catch up performance wise?

Oh yes. All they need is significantly more R&D funding than Intel has, the right people, and the facilities. All of that should be a cinch.

Has the performance of the mainstream desktop AMD CPUs increased more than the corresponding Intel CPUs in the last few years?

No.

And what can we expect looking forward 1-3 years?

Depending on the PS4 deal, hopefully their accounts department will need less red ink and can start using black again.

 

[pantsaregood]

AMD caught up with Intel around 1999 and then ran over them from 2003-2006. This happened largely because Intel made quite a few dumb decisions.

Intel had issues pushing clocks up on Coppermine, so they decided to push NetBurst because they realized clock speeds were marketable. Pentium 4 came, and it wasn't able to compete with K7 (or P3 Tualatins, which were released later) until it reached very high clock speeds. K7 and K8 were both great designs for performance; NetBurst was designed to be marketable.

 

[jaedaliu]

AMD caught up with Intel around 1999 and then ran over them from 2003-2006. This happened largely because Intel made quite a few dumb decisions.

Intel had issues pushing clocks up on Coppermine, so they decided to push NetBurst because they realized clock speeds were marketable. Pentium 4 came, and it wasn't able to compete with K7 (or P3 Tualatins, which were released later) until it reached very high clock speeds. K7 and K8 were both great designs for performance; NetBurst was designed to be marketable.

This story is an example of how AMD may never pass Intel long term. AMD has always competed with Intel because the 2 companies set their chip prices to be competitive.

However, with the K7, AMD had the dominant CPU for quite a while. Intel recognized it, and pushed R&D hard, and pretty quickly passed AMD (quick for chip design. I believe it was within 1 generation) and hasn't looked back.

So, yes, I think AMD might be catching up, but Intel's size and resources allow it to get back in front

 

[BrightCandle]

During AMD's performance lead Intel actively did deals with many companies to exclude AMD CPU's from their line up, which is how they continued to outsell a superior product at the time. There was a massive amount of money that exchanged hands for the settlement.

AMD is not catching up, its gradually falling behind despite the limited performance increases we are seeing.

 

[ShintaiDK]

Remember chip design cost goes up rapidly. So while a company might have been able to compete x years ago doesnt mean it can even afford it today.

Its expected that 14nm kills of half the current semiconductor companies. Simply due to design cost.

 

[Idontcare]

During AMD's performance lead Intel actively did deals with many companies to exclude AMD CPU's from their line up, which is how they continued to outsell a superior product at the time. There was a massive amount of money that exchanged hands for the settlement.

AMD is not catching up, its gradually falling behind despite the limited performance increases we are seeing.

AMD was capacity constrained, they sold every cpu they made. Intel wasn't keeping them from selling more chips, their own capex decisions were keeping them from having more chips to sell.

And when they found a way to line up more financing which could have been used to expand production and enable them to sell more chips and take more marketshare they instead opted to use the financing to pay a ridiculously outsized price for ATI. A $5B investment that has never netted more than about $100m profit per year.

Intel's exclusivity deals certainly diminished AMD's brand recognition, but they did not prevent AMD from selling every chip they could make.

But Intel's ability to outsell a superior product had nothing to do with brand recognition of that superior product. Intel sold their chips because they were the only ones you could buy because AMD's chips were severly supply limited.

Irate Partners Blame AMD, Dell For Athlon Chip Shortage

System builders are irked about a severe shortage of Advanced Micro Devices Athlon 64 X2 processors and blame the chip maker's partnership with Dell.

"It's a fiasco. There's no product in the channel. It's all going to Dell," said Glen Coffield, president of CheapGuys, a system builder in Orlando, Fla. "AMD is divorcing the channel."
Coffield said AMD is hurting its loyal system builders, who feel they have been used and dumped now that the chip maker has a partnership with Dell and other top-tier OEMs.

"I'm being treated like the red-headed stepchild. Intel was always a company for the big guys, and AMD was always a company for the small guys like us," Coffield added. "The channel made AMD. Now they're at the dance with Dell, and they've thrown us to the wolves."

http://www.informationweek.com/irate-partners-blame-amd-dell-for-athlon/193500909

^ does that sound like a company that was being held back in terms of having places to sell their chips? Didn't matter how good AMD's chips were, they didn't have enough of them to sell to meet demand. That left 70% of the market having little choice but to buy second-best because second-best was at least in stock and available for shipment.

 

[cytg111]

it is pretty simple calc, given the same cadence, if amd improves more, it is catching up.
It is worth noting that when we talk about % IPC increase, it is relative to it's predecessors performance, so say piledriver is index 100 and next revision yields a 10% ipc increase == index 110 .. Now an IVB, relative to piledriver, may be index 200, a 10% ipc increase for next revision is this 220. 10 vs 20, same percentage ipc increase. We may be seeing AMD picking up the fruits that intels allready gathered.

 

[Idontcare]

We may be seeing AMD picking up the fruits that intels allready gathered.

Doesn't AMD have access to all of Intel's patented innovations?

Sure they can't introduce them in real-time as Intel files the patents but they can incorporate them with about a 3yr lag time.

So at worst AMD should never be more than about 3 years behind Intel's microarchitecture IPC, unless they choose to not implement Intel IP into their CPU designs. And process node is another matter too.

Intel is going to be on 14nm while AMD will still be trying to get to 28nm. That kind of delta can't be easily overcome with CPU design alone.

 

[inf64]

We are already hitting the wall when it comes to "IPC" ,intel is already near the maximum of what Core design can achieve with Haswell derivative. AMD has some room left since their uarchitecture(BD/PD) is not IPC oriented but they will soon too hit the ceiling . Then we are left with heterogeneous computing and multicore programming. These two areas are the only ones that can provide tangible performance increases in TDP limited scenarios.

 

[Shimmishim]

Is Intel's lack of performance increase partially due to non-competitiveness of AMD? Are we seeing what we saw back in early 2000 where Intel got comfortable with where they were at while AMD pushed ahead? I really loved AMD and their K7/K8 line of chips.

 

[Ventanni]

I think AMD can catch up in performance of the "Core" designs using the Bulldozer architecture, but they will only be able to do so with a larger transistor budget.

 

[Arachnotronic]

If you guys think Intel is asleep at the wheel as it pushes performance AND power efficiency up dramatically (Haswell will go into TABLETS), then you don't really "get it".

AMD is long gone as a competitor. Intel can build faster chips, that consume less power, and cost less to make.

 

[MisterMac]

I think we should rather debate wether we will ever see more huge IPC based increases at all - unless we're talking a drastic change in x86\x64 ARCH in itself.

Software will always have to play catch up to new massively improved designs.

a HSA alike layer needs to be implemented in software everywhere.
And everyone needs to agree to a standard.

HSA isn't even remotely close to a standard.
If intel keeps it''s current position - and steals more\more in Mobile we're going to see real stagnation in performance absolutes.

I would argue both companies are stalling in favour of more secure revenue.
It would be risky as fuck to undertake and actually force something like HSA thru.

Intel has the sheer force - but why risk it when theres endless revenue ready as mobile and cloud(read Datacenter sales) take over and each square MM2 becomes worth more to intel then in desktop?

 

[cytg111]

So we are in relative agreement that singlethreaded performance improvements is levelling off. Sad times. As some code constructs are inherently seriel, impossible to parallelize, we can begin to paint a picture of absolute performance for some workloads. Sad times indeed.

 

[Homeles]

With the current rate of Intel CPU performance increases, could AMD be catching up?

Sure.

What they'll never be able to do is catch up with Intel's performance per watt or power usage. Ever.

AMD doesn't have the fab tech that Intel does. That alone bars AMD from reaching Intel's power levels. Intel's the master of power gating, while AMD has some hacked copycat version. Intel was the first of the bunch to use turbo boost. They were the first to FinFETs. They were the first to move the voltage regulation on package. They're incredibly brilliant when it comes to power management, while AMD tends to leech off of their ideas.

Intel's not playing the performance game anymore; they're playing the power game. There's not a chance in hell that AMD can beat them at their own game. AMD could beat Intel's performance, but the power consumption would look really ugly.

Barring some sort of disastrous turn of events at Intel, there's no hope of AMD catching up in terms of the overall solution that they provide. The best they can hope for is beating Intel in performance per dollar.

 

[grimpr]

Sure.

What they'll never be able to do is catch up with Intel's performance per watt or power usage. Ever.

AMD doesn't have the fab tech that Intel does. That alone bars AMD from reaching Intel's power levels. Intel's the master of power gating, while AMD has some hacked copycat version. Intel was the first of the bunch to use turbo boost. They were the first to FinFETs. They were the first to move the voltage regulation on package. They're incredibly brilliant when it comes to power management, while AMD tends to leech off of their ideas.

Intel's not playing the performance game anymore; they're playing the power game. There's not a chance in hell that AMD can beat them at their own game. AMD could beat Intel's performance, but the power consumption would look really ugly.

Barring some sort of disastrous turn of events at Intel, there's no hope of AMD catching up in terms of the overall solution that they provide. The best they can hope for is beating Intel in performance per dollar.

I do agree with Intels big fab advantage but claiming AMDs powergating in its new chips that have better idle power usage than Ivy Bridge and its Turbo Core implementation which really works as hacked and lame copycats is ridiculous.

 

[mrmt]

And when they found a way to line up more financing which could have been used to expand production and enable them to sell more chips and take more marketshare they instead opted to use the financing to pay a ridiculously outsized price for ATI. A $5B investment that has never netted more than about $100m profit per year.

Wanna bet that Seamicro yields a comparable pie-in-the-sky ROI?

 

[Torn Mind]

Is Intel's lack of performance increase partially due to non-competitiveness of AMD? Are we seeing what we saw back in early 2000 where Intel got comfortable with where they were at while AMD pushed ahead? I really loved AMD and their K7/K8 line of chips.

Intel is pouring money and resources into mobile because that's where all the users and money are heading to and they are playing catchup in that marketspace(and they still haven't caught up). They are focused on making competitive mobile chips now because that's where the money is to be made.

 

[Homeles]

its new chips that have better idle power usage than Ivy Bridge

Proof?

and its Turbo Core implementation which really works as hacked and lame copycats is ridiculous.

And it only took them 3 years. Their solution still isn't as fine-grained, and will never be.

 

[mrmt]

I think we should rather debate wether we will ever see more huge IPC based increases at all - unless we're talking a drastic change in x86\x64 ARCH in itself.

When did we have huge IPC increases from uarch changes? The change from netburst to Core isn't a good comparison, as Netburst was really inefficient.

 

[Homeles]

When did we have pure IPC increases from uarch changes? The change from netburst to Core isn't a good comparison, as Netburst was really inefficient.

Nehalem, Sandy Bridge, Ivy Bridge. Have you not been paying attention?

 

[mrmt]

Nehalem, Sandy Bridge, Ivy Bridge. Have you not been paying attention?

sorry, it should have been huge, not pure. just edited.

 

[MisterMac]

When did we have huge IPC increases from uarch changes? The change from netburst to Core isn't a good comparison, as Netburst was really inefficient.

Yet Conroe devastated AMD as well.



You won't see these jumps anymore - particular because they're not the focus, and because i doubt it's possible even with a 50% increase in TDP to reach 50% raw IPC.

(Not being technical wonder here - that's a Mac Geusstimate).

 

[grimpr]

Proof?

And it only took them 3 years. Their solution still isn't as fine-grained, and will never be.

Here's your proof for claiming that AMDs power gating is a lame hack.