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

[Vesku]

AMD went wide while Intel has focused on squeezing out performance. It's only a matter of time before Intel needs to consider "wide", we are already seeing that with things like TSX. AMD changed first because "squeezing out performance" clearly gets more and more expensive over time.

I don't see AMD catching up but I do think they have a good chance of getting "close enough" as the gains from focusing on IPC per core continue to peter out. Assuming they are still around.

 

[Sheep221]

It's possible that AMD could catch up or overdo intel in performance, but they don't have it easier but harder, since intel is doing small steps, they also gonna do smaller and smaller steps, it's not like it was in the past for both companies.

 

[Cookie Monster]

It's a similar point to what I said about desktops. You can't really draw conclusions about the idle power of a chip by looking at platform idle power. My point was that if he's deluded enough to believe the absolutely asinine assertion that Trinity has better idle power than Ivy Bridge on the desktop (I mean come on. Really? Are you really that ignorant?), then by his ridiculous logic, AMD loses out just as badly in the laptop sector.

You and I both agree that this isn't the case. There are far too many variables to draw a foregone conclusion about idle power draw of a CPU when the platforms are completely different.

And like I said, power gating isn't all about idle power. Workloads are dynamic. Individual parts of a core are constantly being powered on and off. Intel has a performance per watt lead for good reason, and it's not simply because of fab tech.

And he can live in his own little world all he likes about AMD having a lead, but apparently he's not heard of Haswell and the changes that are coming with it that pertain specifically to power gating. What was the number that Intel's throwing around? 10x better idle power? Yeah. Way to miss that one, grimpr.

Yet you are here drawing conclusions that AMD has some "hacked" version of power gating without any scientific evidence to suggest so. If you think otherwise, it'd be very interesting to know why its "hacked".. whatever that means.

 

[IntelUser2000]

I'm curious what percentage of "idle time" my 3770k's cores are truly power-gated. In task manager I see what appears to be a perpetual unending background activity around 1-3%.

From the datasheet:

C1E: 28W
C3: 22W
C6: 5.5W

 

[Idontcare]

From the datasheet:

C1E: 28W
C3: 22W
C6: 5.5W

That's idle power. I was wondering how much time my CPU's cores actually spend in those power states.

Looking at task manager and processor utilization I would guess my 3770k never puts a core into C6 because there is always some non-zero level of background activity on each core in the 1-3% range.

Is there any way of monitoring the individual core's power-states? (log file or some such?)

 

[grimpr]

I'm curious what percentage of "idle time" my 3770k's cores are truly power-gated. In task manager I see what appears to be a perpetual unending background activity around 1-3%.

You should try ThrottleStop, a little known program but usefull, it displays the idle states and keeps logs too.

http://www.techinferno.com/downloads/?did=41

About ThrottleStop

Performance Adjustment Tool for Core 2 / Core i CPUs

ThrottleStop is a small application designed to monitor for and correct the 3 main types of CPU throttling that are being used on many laptop computers.

The left side of ThrottleStop contains a variety of options which can be used to bypass CPU throttling and on the right side is a Monitoring panel that shows you the current state of each thread on your CPU.

 

[Makaveli]

You should try ThrottleStop, a little known program but usefull, it displays the idle states and keeps logs too.

http://www.techinferno.com/downloads/?did=41

Thank you I was trying to remember the name of this application.

 

[Idontcare]

You should try ThrottleStop, a little known program but usefull, it displays the idle states and keeps logs too.

http://www.techinferno.com/downloads/?did=41

Oh wow! Too cool

Interesting results. On my laptop (i5-2410M, a 2C/2T SB) which is idling and doing nothing beyond background thread servicing at Win7 desktop is seeing Core C7 ~90% and package C7 ~60% of the time according to that tool.

Will try it on my desktop too.

 

[Makaveli]

Oh wow! Too cool

Interesting results. On my laptop (i5-2410M, a 2C/2T SB) which is idling and doing nothing beyond background thread servicing at Win7 desktop is seeing Core C7 ~90% and package C7 ~60% of the time according to that tool.

Will try it on my desktop too.

This is what i'm seeing with my laptop idling.

The wifi monitor scans in 10 second intervals and skype is running so there is still some background activity.

 

[Idontcare]

This is what i'm seeing with my laptop idling.

The wifi monitor scans in 10 second intervals and skype is running so there is still some background activity.

This is what I see at idle on my laptop:

 

[Exophase]

Looking at task manager and processor utilization I would guess my 3770k never puts a core into C6 because there is always some non-zero level of background activity on each core in the 1-3% range.

I don't think the latency of going in and out of C6 is so high that you need to be able to see the core stay at 0% for it to be practical to enter it. That sampling period is over a sampling period of at least hundreds of ms if not multiple seconds while the latency going from C6 to C0 is only in the dozens to hundreds of microseconds. You can see here that even getting out of C6 is only around 162 microseconds on an old Core 2:

http://pm-blog.yarda.eu/2011/10/deeper-c-states-and-increased-latency.html

On an Ivy Bridge the latency may be even lower because there's less L2 cache (and because Intel has optimized power saving modes a lot), and because L2 can be flushed to L3 instead of main memory (since it's strictly inclusive). This is under the assumption that C6 doesn't power down L3 like the chart implies.

 

[Exophase]

Oops, double post :/

 

[Homeles]

Yet you are here drawing conclusions that AMD has some "hacked" version of power gating without any scientific evidence to suggest so. If you think otherwise, it'd be very interesting to know why its "hacked".. whatever that means.

Yeah, I probably shouldn't be making claims when I can't find the source for those claims. However, I think it's pretty much inarguable that AMD has and will likely be at least one step behind Intel when it comes to power management. Intel was first to turbo boost, first to power gating, first to HKMG, first to dynamic L2 (lesser known feature of Ivy Bridge), and will be the first to on package voltage regulation. I don't know why you all find it so remarkable that AMD's power gating solution isn't as good as Intel's.

 

[Exophase]

first to dynamic L2 (lesser known feature of Ivy Bridge)

Source? I don't remember reading anything about IB being able to power down parts of its L2..

 

[Homeles]

Source? I don't remember reading anything about IB being able to power down parts of its L2..

They didn't, AFAIK, until Hot Chips 2012, after it had released and the journalism crowd stopped caring.

They really go in intricate detail over all of the power management tweaks that they did in Ivy Bridge. The dynamic cache sizing in question is presented at 20:49 of this video. I was actually mistaken -- the dynamic cache sizing is only for the L3.

 

[Piroko]

Yeah, I probably shouldn't be making claims when I can't find the source for those claims. However, I think it's pretty much inarguable that AMD has and will likely be at least one step behind Intel when it comes to power management. Intel was first to turbo boost, first to power gating, first to HKMG, first to dynamic L2 (lesser known feature of Ivy Bridge), and will be the first to on package voltage regulation. I don't know why you all find it so remarkable that AMD's power gating solution isn't as good as Intel's.

It's not as black and white as you try to portrait. AMD generally led Intel in the A64 days with power features and got us dynamic clocks on desktop and even asynchronous dynamic clocks with the Phenom.
They were steamrolled though by Nehalem, SB and the constant troubles with GloFo.

I don't think their designs are generally uncompetitive, 40nm Bobcats held their own against 45nm Atoms. Also, depending on if the rumor is true that Broadwell will not see Desktop adoption there is an opening for AMD to at least partially catch up/get higher asps.

 

[CHADBOGA]

It's not as black and white as you try to portrait. AMD generally led Intel in the A64 days with power features and got us dynamic clocks on desktop and even asynchronous dynamic clocks with the Phenom.
They were steamrolled though by Nehalem, SB and the constant troubles with GloFo.

I don't think their designs are generally uncompetitive, 40nm Bobcats held their own against 45nm Atoms. Also, depending on if the rumor is true that Broadwell will not see Desktop adoption there is an opening for AMD to at least partially catch up/get higher asps.

If Broadwell isn't available for the Desktop, Intel will up the clocks on Haswell, so whilst it starts out with a max base rate of 3.5Ghz, Intel might end up releasing a 4.0Ghz base rate Haswell, before Desktop Skylake arrives.

 

[tweakboy]

NEVER! dont forget Ivy E 6 to 12 core chips are coming out Q3 2013.

These will be the fastest CPUs next to haswell quad cores. In 2014 they will bring out Haswell enthusiast chips, 6 core 8 core etc etc,,,

The fastest CPU of 2012 was 6 core 3960k. The fastest CPU of 2013 will be Ivy Bridge E 6 to 12 core monsters. Intel has hit a nice wall guys trust me. Or else haswell wouldnt be 3 percent faster in GPU and same similar results in CPU. Could also be 32GB max on that mobo you would be on.

 

[Fjodor2001]

Also, depending on if the rumor is true that Broadwell will not see Desktop adoption there is an opening for AMD to at least partially catch up/get higher asps.

Wasn't the rumor that there will only be BGA versions of Broadwell? That does not mean there will not be any desktop versions of Broadwell. You can fit a BGA CPU on a desktop motherboard.

My guess is that Intel intends to go in the direction of AIO (All-In-One) computers, like the iMac or Intel NUC.

But you're right that in that case perhaps AMD can catch up in the performance desktop segment.

 

[Homeles]

It's not as black and white as you try to portrait. AMD generally led Intel in the A64 days with power features and got us dynamic clocks on desktop and even asynchronous dynamic clocks with the Phenom.
They were steamrolled though by Nehalem, SB and the constant troubles with GloFo.

A lot changes in 6 years. Especially in the tech sector. Yes, almost everyone knows the AMD glory days. There's no sign of those returning.

I don't think their designs are generally uncompetitive, 40nm Bobcats held their own against 45nm Atoms.

I think that's a pretty poor example. The underlying architecture was already 2 and a half years old.

Also, depending on if the rumor is true that Broadwell will not see Desktop adoption there is an opening for AMD to at least partially catch up/get higher asps.

The rumor was about BGA. There would still be desktop boards, but they'd be non-upgradeable.

I stand by what I said. There's really no way that AMD can beat Intel at the performance per watt or power game. The best they can do is be in Intel's shadow, undercutting them and offering consumers better performance per dollar.

It's not a bad place to be in. The Phenom II was a very popular alternative to Nehalem -- right now, they're not quite back to that "good enough" position that they were back then. Richland looks like it will only maintain the status quo, but Kaveri looks promising.

We all know that Intel has the larger R&D budget by a long shot. How could AMD surpass them? They're certainly turning things around, but that doesn't mean they'll advance at a faster rate than Intel.

 

[dave_the_nerd]

If Broadwell isn't available for the Desktop, Intel will up the clocks on Haswell, so whilst it starts out with a max base rate of 3.5Ghz, Intel might end up releasing a 4.0Ghz base rate Haswell, before Desktop Skylake arrives.

Given how idiot-proof overclocking a -K series i5 or i7 to 4+GHz is, I kinda have to believe they'd already be able to sell a 4GHz chip if they wanted to.

Which leads me to believe they haven't bothered because they don't have to. :whiste:

</paranoid>

 

[Fjodor2001]

We all know that Intel has the larger R&D budget by a long shot. How could AMD surpass them?

That was also true around 2003-2006, but yet AMD could surpass them then.

 

[Rvenger]

R&D money is very important but having good engineers is another part of the puzzle too. You can throw billions of R&D dollars at a bunch of engineers, but ultimately it takes the smarts to create the success.

I am not saying AMD will catch up, but they have some good engineers and I can honestly say they do pretty damn good with how 'trimmed' down they are right now. I don't think Intel could perform like AMD does with less.

 

[Exophase]

Given how idiot-proof overclocking a -K series i5 or i7 to 4+GHz is, I kinda have to believe they'd already be able to sell a 4GHz chip if they wanted to.

Which leads me to believe they haven't bothered because they don't have to. :whiste:

</paranoid>

Their TDP limits aren't arbitrary, they have real cost/design implications. Better to sell unlocked processors and let the users worry about accommodating the increased power consumption than put OEMs on the hook (and provide costlier HSFs). Intel has had enough headaches recalling P3s and Prescotts that pushed thermal requirements to levels probably not even as aggressive as what you're suggesting.

 

[Piroko]

If Broadwell isn't available for the Desktop, Intel will up the clocks on Haswell, so whilst it starts out with a max base rate of 3.5Ghz, Intel might end up releasing a 4.0Ghz base rate Haswell, before Desktop Skylake arrives.

Can't argue against that since Sandybridge is still looking good after almost 2 years compared to AMDs offerings. I just stated that there's a realistic possibility for AMD to get higher ASPs.

I think that's a pretty poor example. The underlying architecture was already 2 and a half years old.

I took it as a comparison point for uarches at similar process nodes, nothing more. And it was Intels fault that Bobcat was allowed to compete vs. such an old product.

The rumor was about BGA. There would still be desktop boards, but they'd be non-upgradeable.

Not a cheap solution and probably a dealbreaker for enthusiasts. That's a small group, of course, but a very loud one.

I stand by what I said. There's really no way that AMD can beat Intel at the performance per watt or power game. The best they can do is be in Intel's shadow, undercutting them and offering consumers better performance per dollar.
It's not a bad place to be in. The Phenom II was a very popular alternative to Nehalem -- right now, they're not quite back to that "good enough" position that they were back then. Richland looks like it will only maintain the status quo, but Kaveri looks promising.

Agreed, Richland is just a stopgap (better than price cuts though) and the real fight should be between Haswell i3+i5 vs. Kaveri. Kaveri will almost certainly not match Haswells performance per watt, but it doesn't have to to still be a viable product.