Desktop : Higher performance or higher efficiency ?

[R0H1T]

Again, by having more performance at the same power usage will raise the performance/watt and it will still have a higher efficiency than before. I may have not made it clear in the poll but the first chose is all about having more performance at the higher power levels we had before, that is at 130W TDP CPUs.

Actually the first option is ambiguous as in the performance/power increase haven't been quantified so what I assumed was that it was higher performance at the same efficiency levels, lets say that of IVB &/or Haswell, which is plenty enough for most of us that don't like burning power through the desktop aka winter room heaters

When i say "I want" it refers to what the majority of Power users and High-End gamers care about. With graphics cards we get 50%+ performance at almost the same power usage with every new node. That’s also raises the performance/watt isn’t it ?? Well, we could have the same CPU performance gains with every new node.

I've said this earlier but here I go again ~ around 90% of people on desktops don't care & Intel doesn't want to limit themselves to enthusiasts hence you will not get what you want, not to mention less than 5% of applications/games benefit from this charade at quad core(or above) levels !

Core 2 Quad 9xxx was a 95W TDP, Core i7 920 was 130W TDP at 45nm. The performance gains were substantial and nobody complained about the higher power usage.

TDP wasn't the only thing that changed back then & I suspect you already know this, so why carry this baggage of unreal expectations ?

Then SB came at 32nm and raised the performance lowering the power usage at the same time and everyone was happy. But that have changed at 22nm, we barely got a performance increase, instead we only got lower power usage and higher iGPU performance that power users and High-End gamers doesn’t care about.

Again did you consider that the core microarch is reaching its limits & unless Intel jumps the ship again, like from Netburst back in the day, you will get even less of an increase with every subsequent tock ! I suspect AMD hit the same wall with K10 & hence went with bulldozer, the same reason you saw great(er) gains with piledriver, so that path is fairly open to major improvements over the next few revisions well at least till excavator I hope.

That’s the point of this Poll, but if you or others feel that it is biased or it have a hidden agent that’s something I can’t do nothing about, we are a democratic forum after all and everyone’s opinion is free to be heard

The poll is missing a fair few options but anyways it isn't necessarily biased & don't think anyone is suggesting that, however there needs to be some clarification as to what is being said here !

 

[Idontcare]

Again, by having more performance at the same power usage will raise the performance/watt and it will still have a higher efficiency than before. I may have not made it clear in the poll but the first chose is all about having more performance at the higher power levels we had before, that is at 130W TDP CPUs.

When i say "I want" it refers to what the majority of Power users and High-End gamers care about. With graphics cards we get 50%+ performance at almost the same power usage with every new node. That’s also raises the performance/watt isn’t it ?? Well, we could have the same CPU performance gains with every new node.

Core 2 Quad 9xxx was a 95W TDP, Core i7 920 was 130W TDP at 45nm. The performance gains were substantial and nobody complained about the higher power usage.
Then SB came at 32nm and raised the performance lowering the power usage at the same time and everyone was happy. But that have changed at 22nm, we barely got a performance increase, instead we only got lower power usage and higher iGPU performance that power users and High-End gamers doesn’t care about.

That’s the point of this Poll, but if you or others feel that it is biased or it have a hidden agent that’s something I can’t do nothing about, we are a democratic forum after all and everyone’s opinion is free to be heard

I think at a minimum Intel should have released desktop IB SKUs that were the same power footprint as SB SKUs (which would obviously mean the I SKUs would have been higher clocked than the SB SKUs).

Fine if they want to release lower TDP SKUs (95W -> 77W) while keeping clockspeeds the same as the SB ones, but why not also release 95W SKUs with higher stock clocks?

Intel has swung the pendulum too far to the power-miser side with the creation of their 22nm node IMO.

If they don't correct for it with their 14nm then they will leave the door open for AMD in the high-performance arena. It won't take too much more in terms of raw GHz for Piledriver to rival a 2700K (both 32nm chips), so when AMD finally gets around to releasing a 28nm or 20nm FX chip (be it steamroller or EX) with IPC improvements it will only need another 0.5-1GHz and it should be on top of Haswell at this rate.

 

[Sleepingforest]

I'm surprised Intel doesn't offer a higher power ~4GHz CPU series. They love to segment, after all.

Maybe they would have to do more quality checks and include a better cooler and decided the cost wasn't worth it?

 

[IntelUser2000]

I'd like an option #3 = Better performance at same TDP

I haven't upgraded my desktop from the 2600K because I don't feel the need to. Heck, I didn't even notice the gains in performance going from i5 661 to 2600K, even though I did going from Core 2 Duo E6600 to the i5 661.

With Haswell I'll see it based on platform features. Definitely getting an Ultrabook based on it though.

I think at a minimum Intel should have released desktop IB SKUs that were the same power footprint as SB SKUs (which would obviously mean the I SKUs would have been higher clocked than the SB SKUs).

That may have meant it would run hotter on stock, when it already runs at equal temperatures to Sandy Bridge. The balancing they have to do is whether to satisfy the people who want little better performance at stock versus those that don't want higher temperatures.

The interesting thing is, despite people talking about how they want better air-overclocking and attributing lot of 3770K's disappointments due to that, there's significant voice for faster stock CPUs. So which is which?

 

[ClockHound]

The interesting thing is, despite people talking about how they want better air-overclocking and attributing lot of 3770K's disappointments due to that, there's significant voice for faster stock CPUs. So which is which?

It's both! Is that asking so much?

Higher clocks with lower temps? Just rip out that silly little IGP wafer waster and make make the core layout symmetrical and add 2 or moar cores. Then hire IDC and his crack robotic team of IHS modders to perform final assembly and volcanic thermal tests. How hard can that be?

It's the 2nd decade of the 21st century, Intel, let's make the breaking of Moore's Law have a happy ending.

 

[Sable]

It's both! Is that asking so much?

Higher clocks with lower temps? Just rip out that silly little IGP wafer waster and make make the core layout symmetrical and add 2 or moar cores. Then hire IDC and his crack robotic team of IHS modders to perform final assembly and volcanic thermal tests. How hard can that be?

It's the 2nd decade of the 21st century, Intel, let's make the breaking of Moore's Law have a happy ending.

Exactly this. ^^^

 

[pandemonium]

Option 3:

Just plan your purchases around whichever side of the clock you prefer.

 

[Charles Kozierok]

If they don't correct for it with their 14nm then they will leave the door open for AMD in the high-performance arena. It won't take too much more in terms of raw GHz for Piledriver to rival a 2700K (both 32nm chips), so when AMD finally gets around to releasing a 28nm or 20nm FX chip (be it steamroller or EX) with IPC improvements it will only need another 0.5-1GHz and it should be on top of Haswell at this rate.

You could be right, but I think if it gets to that point, Intel will have to change its approach, as it has done in the past.

I wonder if Intel's now-miserly nature with respect to performance chips is a secondary effect. The server/WS market is Intel's cash cow. With AMD being all but driven out of the professional side, Intel's bean-counters may view the only threat to the low end of the Xeon line as coming from consumer-grade chips that are too close to them in power and capability. So, they keep the enthusiast side "under wraps", basically knowing that we're going to just have to live with whatever crumbs they throw at us anyway.

It's unfortunate, but this is what happens when one company dominates a field, I suppose.

 

[Idontcare]

You could be right, but I think if it gets to that point, Intel will have to change its approach, as it has done in the past.

I wonder if Intel's now-miserly nature with respect to performance chips is a secondary effect. The server/WS market is Intel's cash cow. With AMD being all but driven out of the professional side, Intel's bean-counters may view the only threat to the low end of the Xeon line as coming from consumer-grade chips that are too close to them in power and capability. So, they keep the enthusiast side "under wraps", basically knowing that we're going to just have to live with whatever crumbs they throw at us anyway.

It's unfortunate, but this is what happens when one company dominates a field, I suppose.

My view on Intel's node development priorities are admittedly "tainted" with the personal experience gathered at TI, so take my view with a note that it may be "too much tree, can't see the forest"...but at TI, same as any foundry, we tried to do "all of the above" when building our nodes and that meant pushing hard on low-power (low leakage) R&D at the expense of building higher clockspeed capable process tech.

If 22nm is the result of Intel's higher-level (management-wise) aspirations of getting into mobile phones and tablets then it stands to reason, to me anyways, that the node itself finds itself to be essentially treading water performance-wise with respect to its 32nm predecessor.

In other words it could very well be intentional, albeit not desired, that 22nm turned out to be so miserly relative to 32nm. Intel wanted (needed) it to be a power-miser node, and prioritized it so while applying the constraint that it must be capable of at least the same clockspeeds as 32nm (so it at least wouldn't take a step back in that regard).

Now that is my view of things through the goggles of that being how things routinely went down at TI where we constantly traded off high-performance technologies (for SUN, for which we were their foundry) to get low-power technologies into manufacturing (ironically enough it was also for the pursuit of the mobile phone market).

 

[frozentundra123456]

My view on Intel's node development priorities are admittedly "tainted" with the personal experience gathered at TI, so take my view with a note that it may be "too much tree, can't see the forest"...but at TI, same as any foundry, we tried to do "all of the above" when building our nodes and that meant pushing hard on low-power (low leakage) R&D at the expense of building higher clockspeed capable process tech.

If 22nm is the result of Intel's higher-level (management-wise) aspirations of getting into mobile phones and tablets then it stands to reason, to me anyways, that the node itself finds itself to be essentially treading water performance-wise with respect to its 32nm predecessor.

In other words it could very well be intentional, albeit not desired, that 22nm turned out to be so miserly relative to 32nm. Intel wanted (needed) it to be a power-miser node, and prioritized it so while applying the constraint that it must be capable of at least the same clockspeeds as 32nm (so it at least wouldn't take a step back in that regard).

Now that is my view of things through the goggles of that being how things routinely went down at TI where we constantly traded off high-performance technologies (for SUN, for which we were their foundry) to get low-power technologies into manufacturing (ironically enough it was also for the pursuit of the mobile phone market).

This may be true, but I tend to agree more with your previous post. Especially, couldnt they at least increase clockspeed however much is possible to stay within say a 100 watt tdp? I know I am always criticizing AMDs power consumption, but that is because it doesnt come with increased performance for the uses I am interested in. If Intel could increase clockspeed even 7 or 8 percent for an increase in TDP to 100 watts, I would gladly accept that.

I also think if they are really interested in high end consumer performance (which I actually question), they will eventually have to bring out a reasonably priced hex core on the mainstream platform. IMO they are becoming too reliant on high IPC and hyperthreading at the expense of brute force more cores.

 

[Idontcare]

This may be true, but I tend to agree more with your previous post. Especially, couldnt they at least increase clockspeed however much is possible to stay within say a 100 watt tdp?

Yeah, while I don't now how narrow the intra-bin distribution is with respect to power consumption and Vmin, but my 3770k can run IBT at 4.4GHz and consumes only 101W (peak max) while hitting only 62C (peak max).

This is about 10% higher clockspeed than what my 2600K can do with the same power footprint (4GHz @ 58C and 97W power usage).

The difference is Intel would have to address their rather craptastic IHS thermal barrier problem before that would be possible without requiring customers to delid their CPUs.

So it is possible, they have the technology to make it happen.

 

[Ajay]

I think at a minimum Intel should have released desktop IB SKUs that were the same power footprint as SB SKUs (which would obviously mean the I SKUs would have been higher clocked than the SB SKUs).

Fine if they want to release lower TDP SKUs (95W -> 77W) while keeping clockspeeds the same as the SB ones, but why not also release 95W SKUs with higher stock clocks?

Intel has swung the pendulum too far to the power-miser side with the creation of their 22nm node IMO.

If they don't correct for it with their 14nm then they will leave the door open for AMD in the high-performance arena. It won't take too much more in terms of raw GHz for Piledriver to rival a 2700K (both 32nm chips), so when AMD finally gets around to releasing a 28nm or 20nm FX chip (be it steamroller or EX) with IPC improvements it will only need another 0.5-1GHz and it should be on top of Haswell at this rate.

Hmm, some interesting points; especially the last one. AMD/GFL can tune their transistors to have a performance optimized xtor switch speed vs vGate curve than Intel and possibly take back the enthusiast market (assuming they stay solvent long enough).

If EX on 20nm FD-SOI is real (can't really tell WTH AMD/GFL is doing lately) they could make up considerable ground or even beat Intel (depending mainly on GFL delivering, I think). Won't save the company, but could be a good market niche for them.

It would be really ironic if I changed out my plans for a Haswell-E build to an AMD FX build. Man, may you live in interesting times for sure.

 

[Concillian]

There's nothing I do that needs more CPU performance. ATs latest multi-GPU article shows how little CPU is needed for good gaming performance on high end single GPU systems... which probably covers a good 95+% of "hardcore" PC gamers. A very slow march on CPU performance coupled to aggressively pursuing energy efficiency is acceptable to me.

For those who need raw CPU performance, the vast majority of them are able to gain speed through extra threads. Pushing IPC like in the past just doesn't make business sense. Times have changed, paradigm has already shifted.

 

[bleucharm28]

Mini ITX is a TON of fun to play with... no joke.

I have built one for almost all my family members who wanted a HTPC.

YEP!! I love mini itx. Small yet powerful little bastards!

 

[Zucker2k]

I think at a minimum Intel should have released desktop IB SKUs that were the same power footprint as SB SKUs (which would obviously mean the I SKUs would have been higher clocked than the SB SKUs).

Fine if they want to release lower TDP SKUs (95W -> 77W) while keeping clockspeeds the same as the SB ones, but why not also release 95W SKUs with higher stock clocks?

Intel has swung the pendulum too far to the power-miser side with the creation of their 22nm node IMO.

If they don't correct for it with their 14nm then they will leave the door open for AMD in the high-performance arena. It won't take too much more in terms of raw GHz for Piledriver to rival a 2700K (both 32nm chips), so when AMD finally gets around to releasing a 28nm or 20nm FX chip (be it steamroller or EX) with IPC improvements it will only need another 0.5-1GHz and it should be on top of Haswell at this rate.

A 4.0-4.5Ghz SR or EX chip on 28/20nm silicon with a stock AMD cooler and fan could be a potential fire hazard. Even though Intel's 22nm stock cooler seems rather anorexic compared to the LGA 775 variants, IMO heat could very well have been a factor in the lowering of TDP. On the contrary, IF heat is not an issue for Intel, then they could simply release higher clocked parts to counter any threat from AMD.

NVM. Looks like you touched on my last point above.

 

[Bubbleawsome]

I think a lot of people just want so say that they got "dat GHz" and "dat coarz" despite that fact that they don't use even use it, and those are the really fanboyish people on these forums.

THB, I am saving for a x2 i7 3(or 4 or 5) 930k + x4 titan or = by then. I try and judge all CPUs and other components equally. I want this for a 4 screen 1600p each setup. (and the bragging rights) :whiste:

But as I don't pay for my electricity at the moment I like MOAR performance.

 

[Exophase]

It won't take too much more in terms of raw GHz for Piledriver to rival a 2700K (both 32nm chips), so when AMD finally gets around to releasing a 28nm or 20nm FX chip (be it steamroller or EX) with IPC improvements it will only need another 0.5-1GHz and it should be on top of Haswell at this rate.

I assume we're talking about max turbo speeds and single threaded performance here. 500MHz is only a 12.8% increase over Haswell's max turbo. That's similar to the IPC improvement from Ivy Bridge to Haswell, so we're talking about AMD matching Ivy Bridge's IPC. Based on what we know about Steamroller I'd be floored if they could get anywhere remotely close to this. Since I have no idea what Excavator changes I can't comment on that, but it'd still be impressive and unexpected even by then.

I don't think they'll be able to push clocks that much higher than that on stock air. 4.9GHz seems too high. But I guess we'll see.

If they drive up platform power consumption too much they're going to be shooting themselves in the foot by increasing the cost of their cooling and increasing demands on motherboard manufacturers and OEMs. Having several different coolers and sets of guidelines doesn't sound too attractive either. I'm not sure a halo product will impress people that much anyway, if a lot of the more interested parties will be overclockers to begin with and will be using unlocked Intel CPUs.

 

[Ajay]

I assume we're talking about max turbo speeds and single threaded performance here. 500MHz is only a 12.8% increase over Haswell's max turbo. That's similar to the IPC improvement from Ivy Bridge to Haswell, so we're talking about AMD matching Ivy Bridge's IPC. Based on what we know about Steamroller I'd be floored if they could get anywhere remotely close to this. Since I have no idea what Excavator changes I can't comment on that, but it'd still be impressive and unexpected even by then.

I don't think they'll be able to push clocks that much higher than that on stock air. 4.9GHz seems too high. But I guess we'll see.

If they drive up platform power consumption too much they're going to be shooting themselves in the foot by increasing the cost of their cooling and increasing demands on motherboard manufacturers and OEMs. Having several different coolers and sets of guidelines doesn't sound too attractive either. I'm not sure a halo product will impress people that much anyway, if a lot of the more interested parties will be overclockers to begin with and will be using unlocked Intel CPUs.

Even if the "Force" is with them, AMD will not be able match Intel with SR. Maybe 20nm and EX if Haswell and Broadwell deliver the next the single digit gains we've been seeing lately on Intel performance wise (as opposed to power/performance). Actually, if GFL follows the current SOI consortium mantra - I'd expect that we'd see a 28nm FD-SOI 'SR II' before we'd see Excavator or 20nm.

 

[Vesku]

A 4.0-4.5Ghz SR or EX chip on 28/20nm silicon with a stock AMD cooler and fan could be a potential fire hazard. Even though Intel's 22nm stock cooler seems rather anorexic compared to the LGA 775 variants, IMO heat could very well have been a factor in the lowering of TDP. On the contrary, IF heat is not an issue for Intel, then they could simply release higher clocked parts to counter any threat from AMD.

NVM. Looks like you touched on my last point above.

AMD has actually been using decent coolers for their 125W chips for quite some time. Only real gripe is they use smaller fans than enthusiast cooling.

As for the foundry node talk, I would think GF is also looking at making power efficiency gains with each node. Seems like 28nm GF might compare favorably to Intel 32/22nm in terms of performance chips, might. But after 28nm should be on a similar path of the other foundries: efficiency. Their PR releases regarding 20nm progress tend to focus on ARM chips.

http://globalfoundries.com/newsroom/2012/20120813.aspx

And this:

http://www.3dcenter.org/news/neue-fertigungs-roadmap-von-globalfoundries-zeigt-28nm-shp-prozess

 

[Ajay]

As for the foundry node talk, I would think GF is also looking at making power efficiency gains with each node. Seems like 28nm GF might compare favorably to Intel 32/22nm in terms of performance chips, might. But after 28nm should be on a similar path of the other foundries: efficiency. Their PR releases regarding 20nm progress tend to focus on ARM chips.

http://globalfoundries.com/newsroom/2012/20120813.aspx

And this:

http://www.3dcenter.org/news/neue-fertigungs-roadmap-von-globalfoundries-zeigt-28nm-shp-prozess

Sadly, GFL foundry roadmaps are pointless. We are just getting 28nm in 2013. GFL has run some test wafers @ 20nm, but they are no where near ready to ramp up to full production. We are probably talking 2015 for 20nm

 

[ShintaiDK]

Sadly, GFL foundry roadmaps are pointless. We are just getting 28nm in 2013. GFL has run some test wafers @ 20nm, but they are no where near ready to ramp up to full production. We are probably talking 2015 for 20nm

Hey we all know GLoFo shipped 28nm since 2010 :awe:

 

[Idontcare]

And this:

http://www.3dcenter.org/news/neue-fertigungs-roadmap-von-globalfoundries-zeigt-28nm-shp-prozess

The key thing to note in this graphic is that GloFo has no intention of developing or fielding a process node that is capable of competing with TSMC in the HPC space post 28nm.

Yeah sure they draw the arrows up to the top of their chart, but the placement of the box is exactly where reality is showing up in all their design rules and product development kits (PDKs).

If AMD is going to compete with Intel in non-mobile business SKUs post 28nm then those chips are absolutely not going to be coming from GloFo, simple as that.

(and no, I'm not basing that statement on this PR graphic, I am pointing out that even the speaking between the lines that we must do publicly at times is not needed because the PR slides just so happen to already tell the full story provided people are willing to see it for what it really is )

 

[Ajay]

The key thing to note in this graphic is that GloFo has no intention of developing or fielding a process node that is capable of competing with TSMC in the HPC space post 28nm.

Yeah sure they draw the arrows up to the top of their chart, but the placement of the box is exactly where reality is showing up in all their design rules and product development kits (PDKs).

If AMD is going to compete with Intel in non-mobile business SKUs post 28nm then those chips are absolutely not going to be coming from GloFo, simple as that.

(and no, I'm not basing that statement on this PR graphic, I am pointing out that even the speaking between the lines that we must do publicly at times is not needed because the PR slides just so happen to already tell the full story provided people are willing to see it for what it really is )

Supposedly, GFL is moving towards FD-SOI - I guess if true they are doing it for the power efficiency. Would it be a big deal to add the buried oxide gate and make a high performance part for AMD? I realize that GFL may then asses an R&D fee since they can just leave that off their standardized node list. Not that they'd do that for their biggest customer (what a screwed up arrangement, as we've noted before)...

 

[Ajay]

Hey we all know GLoFo shipped 28nm since 2010 :awe:

Hmm, makes me wonder if 20nm will be out in 2015. I'm expecting that only because some sources said ATIC (the new ATIC) was pushing GFL hard to speed up it's node delivery and supposedly more $$s and effort have been put into that. Still, with their track record ^_^

 

[Idontcare]

Supposedly, GFL is moving towards FD-SOI - I guess if true they are doing it for the power efficiency. Would it be a big deal to add the buried oxide gate and make a high performance part for AMD? I realize that GFL may then asses an R&D fee since they can just leave that off their standardized node list. Not that they'd do that for their biggest customer (what a screwed up arrangement, as we've noted before)...

FDSOI would be relevant for AMD's needs if it were on an availability timeline that was also relevant for AMD's needs.

FDSOI is being brought up in production on a timeline that is relevant to VIA's needs.