SPCR

since most of what i do on a pc is typing forums ie web browsing chatting, listening to music watching videos, and only occasionally make movies or play games, should Intel emulate Arm and offer a 2-core that are haswell-sklylake level performance coupled to 5 watt 2 atom baytrail cpu when doing light work? baytrail should be powerful enough to watch youtube vids

What about the IGP?

quest_for_silence wrote:What about the IGP?

?

dan wrote:

quest_for_silence wrote:What about the IGP?

?

Should Intel use the same littlecore.bigcore approach for the IGP, and in case how? The IGP in BayTrail is quite different from current GT2 (let alone the various GT3s), which account itself for twice the TDP of a whole Silvermont (for up to five times the graphics GFlops).

Wouldn't make sense. The idea behind big.LITTLE is that smartphones are used predominantly with the screen off. Idling in your pocket, your phone is listening for calls, checking e-mails, scanning for wifi, etc... For those idle tasks, it's potentially benefitial to have a dedictated low power/performance CPU and only switch on your high power/performance CPU for the 30 minutes of the day the user is turning on the screen.

In a "traditional" computing device using means having the screen on and actually doing stuff. I can't imagine the power savings to be anything to write home about. A Haswell CPU can basically idle while you browse the web and such. (an Atom may not be able to do so, but I don't know much about the performance of the new Atoms)

quest_for_silence wrote:

dan wrote:

quest_for_silence wrote:What about the IGP?

?

Should Intel use the same littlecore.bigcore approach for the IGP, and in case how? The IGP in BayTrail is quite different from current GT2 (let alone the various GT3s), which account itself for twice the TDP of a whole Silvermont (for up to five times the graphics GFlops).

does arm use that for gpu?

tim851 wrote:Wouldn't make sense. The idea behind big.LITTLE is that smartphones are used predominantly with the screen off. Idling in your pocket, your phone is listening for calls, checking e-mails, scanning for wifi, etc... For those idle tasks, it's potentially benefitial to have a dedictated low power/performance CPU and only switch on your high power/performance CPU for the 30 minutes of the day the user is turning on the screen.

In a "traditional" computing device using means having the screen on and actually doing stuff. I can't imagine the power savings to be anything to write home about. A Haswell CPU can basically idle while you browse the web and such. (an Atom may not be able to do so, but I don't know much about the performance of the new Atoms)

what about mobile laptop notebook ultrabook? use the little atom like core for light work, so in effect you have a netbook, use the big core for playing 3d games.

in a desktop, the atom like core for doing light work, with the big core completely shut down.

dan wrote: what about mobile laptop notebook ultrabook? use the little atom like core for light work, so in effect you have a netbook, use the big core for playing 3d games.

in a desktop, the atom like core for doing light work, with the big core completely shut down.

It is still not the same thing. Interactive response vs non-interactive. You don't care if your email notification on a phone is delayed by a few tenths of a second, but you really would notice if your email program on your desktop took a few tenths of a second longer to respond. When you are actively using a device you want it to respond as fast as possible or it feels slow.

Also, in many cases it has been shown that a faster but higher power CPU doing the same work in less time than a slower but lower power CPU actually saves power overall. The faster CPU spends more time idle so the overall power consumption is lower. This is called "race to idle".

washu wrote:

dan wrote: what about mobile laptop notebook ultrabook? use the little atom like core for light work, so in effect you have a netbook, use the big core for playing 3d games.

in a desktop, the atom like core for doing light work, with the big core completely shut down.

It is still not the same thing. Interactive response vs non-interactive. You don't care if your email notification on a phone is delayed by a few tenths of a second, but you really would notice if your email program on your desktop took a few tenths of a second longer to respond. When you are actively using a device you want it to respond as fast as possible or it feels slow.

Also, in many cases it has been shown that a faster but higher power CPU doing the same work in less time than a slower but lower power CPU actually saves power overall. The faster CPU spends more time idle so the overall power consumption is lower. This is called "race to idle".

what you say sounds reasonable i guess. still netbooks are sold and they have atom or bobcat cpu's. oh well.

dan wrote: what you say sounds reasonable i guess. still netbooks are sold and they have atom or bobcat cpu's. oh well.

The Atom/Bobcat/etc machines are sold primarily for once reason: They are cheap! Sure a few are sold based on battery life or size, but low price is the primary sales factor.

I own two laptops bought within a week of each other from the same manufacturer. Once has an AMD E-350 and the other an i5-2410M. Similar size, same screen res, same RAM, similar battery life (despite the i5 having 10 WH less battery) but not even comparable performance. The i5 night and day faster. Even with both having good SSDs the E-350 is painful to use on the modern web, which is basically what you are proposing (only using the powerful CPU for 3d games). The kicker: maxed out (prime95+furmark) the i5 uses about 4W less power from the wall and about 6W less at idle. Of course, the i5 was nearly 2X the cost of the E-350.

dan wrote:does arm use that for gpu?

As far as I know, no, the graphics is used just with the Big cores. One point is that a basic GT2 IGP should have a 9W TDP, twice a whole 5W Bay-Trail, let alone if you use a discrete graphics, so possible power gain over the overall power draw could be negligible (with contemporary technologies).

washu wrote:

dan wrote: what you say sounds reasonable i guess. still netbooks are sold and they have atom or bobcat cpu's. oh well.

The Atom/Bobcat/etc machines are sold primarily for once reason: They are cheap! Sure a few are sold based on battery life or size, but low price is the primary sales factor.

I own two laptops bought within a week of each other from the same manufacturer. Once has an AMD E-350 and the other an i5-2410M. Similar size, same screen res, same RAM, similar battery life (despite the i5 having 10 WH less battery) but not even comparable performance. The i5 night and day faster. Even with both having good SSDs the E-350 is painful to use on the modern web, which is basically what you are proposing (only using the powerful CPU for 3d games). The kicker: maxed out (prime95+furmark) the i5 uses about 4W less power from the wall and about 6W less at idle. Of course, the i5 was nearly 2X the cost of the E-350.

thanks for the warning! i guess AMD purchase of ATI was a mistake? i guess AMD biggest problem is manufacturing

dan wrote:thanks for the warning!

Not sure how it was a warning. At the time there wasn't anything else that performed better that wasn't either much more expensive (the i5) or much bulkier (my two laptops are 12.5"). My statement was more of a "you get what you pay for" and that Atoms/Bobcats are slow. In no way did I mean that one should not purchase them, they have their use cases.

dan wrote: i guess AMD purchase of ATI was a mistake?

I would not say it was a mistake, just that AMD is putting too much emphasis on the GPU/APU over CPU. GPUs are great when you can use them, but there are still many tasks that need CPU power and that is not likely to change.

i guess AMD biggest problem is manufacturing

While they are behind Intel in process technology, so is pretty much everyone else. I don't think they have a manufacturing problem, more of a design problem. AMD designs and manufactures great GPUs despite the process technology deficiency. AMD designed a bad CPU. Putting a Bulldozer derivative on Intel's latest and greatest process technology would not magically make it competitive, it's simply a poor design.

washu wrote:

dan wrote:thanks for the warning!

Not sure how it was a warning. At the time there wasn't anything else that performed better that wasn't either much more expensive (the i5) or much bulkier (my two laptops are 12.5"). My statement was more of a "you get what you pay for" and that Atoms/Bobcats are slow. In no way did I mean that one should not purchase them, they have their use cases.

dan wrote: i guess AMD purchase of ATI was a mistake?

I would not say it was a mistake, just that AMD is putting too much emphasis on the GPU/APU over CPU. GPUs are great when you can use them, but there are still many tasks that need CPU power and that is not likely to change.

i guess AMD biggest problem is manufacturing

While they are behind Intel in process technology, so is pretty much everyone else. I don't think they have a manufacturing problem, more of a design problem. AMD designs and manufactures great GPUs despite the process technology deficiency. AMD designed a bad CPU. Putting a Bulldozer derivative on Intel's latest and greatest process technology would not magically make it competitive, it's simply a poor design.

by warning i mean intel is a better choice for me than amd.

how does amd gpu compare with nvidia?

but being at .28 vs .14 finfet sounds like a lot to overcome. if they had the transistor budget that .14 affords, they could add a lot of performance enhancing features like trace cache, better ooo deeper pipeline etc

dan wrote: by warning i mean intel is a better choice for me than amd.

While in most cases today that is a true statement, that wasn't the point I was trying to make.

how does amd gpu compare with nvidia?

They compare quite well. At this very moment Nvidia has a slight edge, but ATI/AMD has been on top many times in the past and likely will again. Nvidia/AMD are much closer in the GPU space than Intel/AMD are in the CPU space.

but being at .28 vs .14 finfet sounds like a lot to overcome. if they had the transistor budget that .14 affords, they could add a lot of performance enhancing features like trace cache, better ooo deeper pipeline etc

All those extra features would have to be designed in, they don't just magically appear because of a process shrink. And having an overly deep pipeline is a bad thing, not a good one. While it is not the only reason for Bulldozer's performance issues, its long pipeline is a contributing factor. AMD watched Intel make the long pipeline mistake with the P4 and then did it anyway.

washu wrote:

dan wrote: by warning i mean intel is a better choice for me than amd.

While in most cases today that is a true statement, that wasn't the point I was trying to make.

how does amd gpu compare with nvidia?

They compare quite well. At this very moment Nvidia has a slight edge, but ATI/AMD has been on top many times in the past and likely will again. Nvidia/AMD are much closer in the GPU space than Intel/AMD are in the CPU space.

but being at .28 vs .14 finfet sounds like a lot to overcome. if they had the transistor budget that .14 affords, they could add a lot of performance enhancing features like trace cache, better ooo deeper pipeline etc

All those extra features would have to be designed in, they don't just magically appear because of a process shrink. And having an overly deep pipeline is a bad thing, not a good one. While it is not the only reason for Bulldozer's performance issues, its long pipeline is a contributing factor. AMD watched Intel make the long pipeline mistake with the P4 and then did it anyway.

yeah i wonder about that with amd. is their idea of unified pointers a good idea and should intel adopt it?

so if one person has a 6 core haswell 95 tdp, and another person has say an atom netbook, and both are reading a firefox page or an adobe pdf, or listening to the same mp3 or watching same utube vids, are the cpu both using the same energy?

dan wrote:i guess AMD purchase of ATI was a mistake?

That depends on whether or not it was the financial and logistical burden of merging the two companies that made Bulldozer be the failure it was.

Right now, AMD is only alive because of ATI. If they had no GPUs to sell, they'd be finished.

dan wrote:so if one person has a 6 core haswell 95 tdp, and another person has say an atom netbook, and both are reading a firefox page or an adobe pdf, or listening to the same mp3 or watching same utube vids, are the cpu both using the same energy?

It's not just the CPU, it's the platform. Nobody is putting 6-core Haswell-E's in power-sensitive netbooks.

When everything else (desktop class mainboard, magnetic hard drives, gaming GPUs, 24" monitors) is drawing 50w already, it doesn't really matter if the CPU idles at 2w or at 5w.

dan wrote:yeah i wonder about that with amd. is their idea of unified pointers a good idea and should intel adopt it?

I will be completely honest and say I have no idea what you are talking about when you say "unified pointers".

so if one person has a 6 core haswell 95 tdp, and another person has say an atom netbook, and both are reading a firefox page or an adobe pdf, or listening to the same mp3 or watching same utube vids, are the cpu both using the same energy?

As tim851 said, it really depends on the rest of the system. It's not fair to compare a haswell-E to a netbook. One is a desktop system with desktop parts and one is a netbook with at least some power saving in mind. A haswell-e is not the same as the rest of the CPU line, it is meant for performance above all, power consumption is secondary. It's a worthy trade off if you need it.

When comparing similar systems like my two laptops you can see faster CPUs using less power than slower ones. As I said above, my i5 laptop uses less power than my e-350 despite being much faster. I also have an atom N270 netbook which uses less power than the i5, but is even slower than the e-350. It does use less power, but it is so slow that you simply do less on it out of necessity. For example there is simply no way you would be watching youtube at anything but minimal res while doing anything else, the machine is just to slow. It's using less power because it simply cannot do as much work.

tim851 wrote:

dan wrote:i guess AMD purchase of ATI was a mistake?

That depends on whether or not it was the financial and logistical burden of merging the two companies that made Bulldozer be the failure it was.

Right now, AMD is only alive because of ATI. If they had no GPUs to sell, they'd be finished.

if they only focused on cpu's maybe they'd be equal to intel, if samsung also made them.

washu wrote:

dan wrote:yeah i wonder about that with amd. is their idea of unified pointers a good idea and should intel adopt it?

I will be completely honest and say I have no idea what you are talking about when you say "unified pointers".

so if one person has a 6 core haswell 95 tdp, and another person has say an atom netbook, and both are reading a firefox page or an adobe pdf, or listening to the same mp3 or watching same utube vids, are the cpu both using the same energy?

As tim851 said, it really depends on the rest of the system. It's not fair to compare a haswell-E to a netbook. One is a desktop system with desktop parts and one is a netbook with at least some power saving in mind. A haswell-e is not the same as the rest of the CPU line, it is meant for performance above all, power consumption is secondary. It's a worthy trade off if you need it.

When comparing similar systems like my two laptops you can see faster CPUs using less power than slower ones. As I said above, my i5 laptop uses less power than my e-350 despite being much faster. I also have an atom N270 netbook which uses less power than the i5, but is even slower than the e-350. It does use less power, but it is so slow that you simply do less on it out of necessity. For example there is simply no way you would be watching youtube at anything but minimal res while doing anything else, the machine is just to slow. It's using less power because it simply cannot do as much work.

I was referring to AMD's “heterogeneous Uniform Memory Access”
I hope then SPCR offers updated recommends on cpu and cpu chipsets with silence as goal

dan wrote:I was referring to AMD's “heterogeneous Uniform Memory Access”

Ah ok. It's mostly marketing. Yes, in theory it's a great idea. Let CPU programs and GPU programs access the same memory without needless copying, basically making the "APU" concept that AMD has been pushing.

The reality:

1. There are still a massive number of problems that simply cannot be run well on a GPU. As time goes on more software will become GPU optimized, but there will always be problems which need to be solved serially.

2. Neither the GPU part and especially the CPU part of AMDs APUs are fast enough for this to matter. This will hopefully change in the future, but right now APUs are not nearly fast enough that the tiny amount of time saved in a memory copy offsets their slow speed otherwise.

3. For basic graphics (not GPU computing), Intel and every other integrated graphics vendor have already been doing this for years.

I hope then SPCR offers updated recommends on cpu and cpu chipsets with silence as goal

I'm confused, is this not what SPCR already does?

washu wrote:

dan wrote:I was referring to AMD's “heterogeneous Uniform Memory Access”

Ah ok. It's mostly marketing. Yes, in theory it's a great idea. Let CPU programs and GPU programs access the same memory without needless copying, basically making the "APU" concept that AMD has been pushing.

The reality:

1. There are still a massive number of problems that simply cannot be run well on a GPU. As time goes on more software will become GPU optimized, but there will always be problems which need to be solved serially.

2. Neither the GPU part and especially the CPU part of AMDs APUs are fast enough for this to matter. This will hopefully change in the future, but right now APUs are not nearly fast enough that the tiny amount of time saved in a memory copy offsets their slow speed otherwise.

3. For basic graphics (not GPU computing), Intel and every other integrated graphics vendor have already been doing this for years.

I hope then SPCR offers updated recommends on cpu and cpu chipsets with silence as goal

I'm confused, is this not what SPCR already does?

what about using it for ps4/xbox1? amd gives me the impression that only amd is doing it, not intel.

btw have u seen openrisc, RISC V papers?

dan wrote:

tim851 wrote:

dan wrote:i guess AMD purchase of ATI was a mistake?

That depends on whether or not it was the financial and logistical burden of merging the two companies that made Bulldozer be the failure it was.

Right now, AMD is only alive because of ATI. If they had no GPUs to sell, they'd be finished.

if they only focused on cpu's maybe they'd be equal to intel, if samsung also made them.

This assumes, like I implied, that the ATI acquisition cost AMD a lot of focus.

ATI already had plenty of great engineers, as evidenced by the success of their Radeon lineup. I don't think AMD shipped any of their engineers off to ATI. If anything, it was the other way around.

Then there's the question whether or not AMD spent too much money which they then lacked for funding R&D. That's certainly possible.

Lastly, and I think this is where AMD actually dropped the ball: they bet on APU. They thought it would be more or less trivial to mate a Radeon GPU and an Athlon CPU and greatness would ensue. Well, it wasn't trivial and no greatness ensued.

It got them design wins with console manufacturers, but reportedly that market is approaching a profit margin of 0. Let's hope it's not so and AMD can turn a decent profit on those 50 million consoles that shipped with their parts in 2014.

When you look how little power a current generation Ultrabook can sip at idle I wonder how much room there is for improvement?
Saving a few hundred mW in a phone Soc is worthwhile but in a laptop CPU probably not worth the effort.