Just when you thought desktop PCs couldn't get slower. Along comes AMD...

[cbn]

anyway, AMD's biggest problem is inability to get a fab partner that isn't stuck on the level intel was at 5 years ago. next biggest problem is how bad bulldozer was not just compared to intel's offerings but also AMD's own prior offerings. that leads people to think that bulldozer's offspring have been worse than stars. they're not. but the stink is there. kaveri isn't a bad part at all at its ideal TDP (45-65 watt). if AMD can keep performance level and get the power down to the 25-35 watt range with carrizo they'll have a compelling notebook chip.

Regarding kaveri at 65 watts, even the A8-7600 is prone to lots of cpu throttling under iGPU load. In this test, the cpu clocks dropped to 2.4 GHz during Furmark.

Couple that to the relative lack of bandwidth for the large iGPU and it just seems AMD should decrease iGPU size. (either that or use the very high iGPU die size investment to run extra low clocks in order to get performance per watt up, and decrease cpu throttling.)

 

[Idontcare]

iirc, moore's law not only predicts transistor density but also an optimum die size for any process tech. between that and the sheer number of interconnects necessary on a big iron processor like core, the silicon is going to be there. additional processing cores aren't really useful for typical desktop apps (and if you want/need more cores intel likes its market segregation just like airlines), nor are more SoC type features (and chipsets had typically been manufactured on the previous gen equipment anyway to get some additional use out of it), so the graphics is kinda free.

anyway, AMD's biggest problem is inability to get a fab partner that isn't stuck on the level intel was at 5 years ago. next biggest problem is how bad bulldozer was not just compared to intel's offerings but also AMD's own prior offerings. that leads people to think that bulldozer's offspring have been worse than stars. they're not. but the stink is there. kaveri isn't a bad part at all at its ideal TDP (45-65 watt). if AMD can keep performance level and get the power down to the 25-35 watt range with carrizo they'll have a compelling notebook chip.

and that's all on an ancient fab process.

It's true.

Something I found even more interesting was that if you use Moore's Law on Intel's processors and extrapolate the graph back to when Moore's Law would predict the transistor was invented you actually get the correct answer to within 3 months.

^extrapolation is the dashed line, the yellow square is the actual invention date (not included in the Moore's Law best fit)

Quite the coincidence.

 

[cbn]

It's true.

Something I found even more interesting was that if you use Moore's Law on Intel's processors and extrapolate the graph back to when Moore's Law would predict the transistor was invented you actually get the correct answer to within 3 months.

^extrapolation is the dashed line, the yellow square is the actual invention date (not included in the Moore's Law best fit)

Quite the coincidence.

Those are awesome graphs and I remember you bringing this topic up before in the past.

But these days, It seems the extra xtor budget (in the typical cases we discuss here involving processors) is mostly just buying us extra iGPU (at least on the mainstream consumer socket).

And I feel this gain is not as valuable as it was in the past (think late 90's early 2000s) when the xtor doubling every two years almost went entirely into making the single core processors on a die faster than they used to be.

 

[ElFenix]

Regarding kaveri at 65 watts, even the A8-7600 is prone to lots of cpu throttling under iGPU load. In this test, the cpu clocks dropped to 2.4 GHz during Furmark.

do you have evidence of thermal throttling under real world loads?


carrizo is also supposed to bring a 28 nm implementation of GCN2.0, is it not?

I hear you about "optimum die size". Make the die too large and yields decrease. Make die too small and the increase in yield is not offset by the loss in functionality.

However, at this point I would argue that the ideal die size is smaller than it was in the past and the increase in iGPU really isn't worth it (at least for desktop).

ivy and haswell GT2 have similar die sizes, i wouldn't be the least bit shocked if that is close to the ideal figure.

 

[AtenRa]

Larry. I myself own a very nice looking desktop powered by an A4-5000. Its got USB3 at the front, optical drive, PS/2 outputs, VGA and HDMI (nice for legacy devices) and an extra DDR3L slot, I have 16 gigs of ram in this baby. Problem is, it does rather feel slow, even for regular browsing (albeit it has 4 cores). So I use it just for storage, movie playback, and downloads. It consumes in the range of 15-20w with a large mechanical 7200RPM drive. Other than super low power use and size, I don't see any other advantages (it certainly wasn't cheap). I see it as more of an "accessory", rather than a proper computer.

Disable every power saving settings from BIOS and let the CPU run at 1.5GHz. Power consumption will not be that much higher, a couple of watts more but performance will go up and it will feel much much faster and browsing will be snappier.

 

[cbn]

ivy and haswell GT2 have similar die sizes, i wouldn't be the least bit shocked if that is close to the ideal figure.

Ivy Bridge dual core GT2 (16 EUs) was 118.2 mm2 and Haswell dual GT2 (20 EUs) should only be a little larger.

And Broadell dual core GT2 (24 EUs) is something like ~80mm2 (on 14nm).

See, even with that smaller die size on Broadwell I would argue that Intel should make a GT1 version of that (with 6EUs) even if it meant dropping die size much smaller than ~80mm2.

So even though the extra xtors for that GT2 may not be that costly, my argument is that on desktop the average person will never notice the difference between 6 EUs and 24 EUs. Therefore Intel might as well save the money and get the extra chips on the wafer.

 

[AtenRa]

The larger iGPUs help with higher wafer volumes that leads to higher fab utilization that leads to faster depreciation. Also larger iGPUs have high yields due to the larger part of the iGPU consisting of the same logic (EUs or GPU Cores etc).

 

[Idontcare]

Those are awesome graphs and I remember you bringing this topic up before in the past.

But these days, It seems the extra xtor budget (in the typical cases we discuss here involving processors) is mostly just buying us extra iGPU (at least on the mainstream consumer socket).

And I feel this gain is not as valuable as it was in the past (think late 90's early 2000s) when the xtor doubling every two years almost went entirely into making the single core processors on a die faster than they used to be.

Thanks! And yes, I agree with you. For all the extra hundreds of millions of transistors our CPU's have today over their older siblings from 8-10 years ago, it is rather sad how little that has given us performance-wise versus where things went from the 80's to the 90's and the 90's to mid-decade of the 2000's.

If you put a billion or two transistors towards the creation of a single core CPU then I have to imagine you get some killer performance. Trouble is it would cost so much money to design (architecture complexity) that the chip would probably cost a ridiculous amount, and most people don't like paying ridiculous amounts of money for their chips like they use to in the 80's and 90's.

 

[ShintaiDK]

carrizo is also supposed to bring a 28 nm implementation of GCN2.0, is it not?

1.2 like Tonga it seems.

 

[Magic Carpet]

Disable every power saving settings from BIOS and let the CPU run at 1.5GHz. Power consumption will not be that much higher, a couple of watts more but performance will go up and it will feel much much faster and browsing will be snappier.

When I said "slow" it wasn't slow due to power saving features kicking in. Slow in comparison to my other computers. Hell, even my smartphone is opening pages faster

4 slow cores @ 1.5 Ghz was just a fail, imo. It's still somewhat useable but for a demanding user such as myself, it's giving me waaay too much agony. But for watching/downloading stuff it is decent (where it can take advantage of 2+ threads).

 

[Eug]

I'm curious about those power settings in the BIOS. What do they do? I have an AMD Athlon II X3 with nVidia GPU integrated on the mobo. Yeah, go figure.

I was having stuttering in Netflix HD and it was driving me nuts. I even tried installing a discrete ATI video card but it made no difference. It turns out the problem was having AMD's Cool n Quiet on. Perfectly fine for casual surfing and running MS Office but a disaster for some video playback.

The weird part though was Blu-ray worked fine.

 

[Essence_of_War]

I was having stuttering in Netflix HD
...
The weird part though was Blu-ray worked fine.

That sounds like an h.264 decode problem then, right? It sure seems like a good guess since blu-ray IS NOT h.264 encoded, and netflix certainly IS.

 

[Eug]

That sounds like an h.264 decode problem then, right? It sure seems like a good guess since blu-ray IS NOT h.264 encoded, and netflix certainly IS.

Actually, Blu-ray is H.264 mostly, although a lot is VC1 and some is MPEG2.

Anyway my nVidia integrated GPU had hardware h.264 decoding.

 

[ninaholic37]

I was having stuttering in Netflix HD and it was driving me nuts. I even tried installing a discrete ATI video card but it made no difference. It turns out the problem was having AMD's Cool n Quiet on. Perfectly fine for casual surfing and running MS Office but a disaster for some video playback.

Were you running Netflix in your browser? The Netflix app for Windows 8 should be faster.

Intel's Enhanced Speedstep can stutter videos too if you set the max freq to start at 80%/95% CPU instead of 50%.

 

[cbn]

The larger iGPUs help with higher wafer volumes that leads to higher fab utilization that leads to faster depreciation. Also larger iGPUs have high yields due to the larger part of the iGPU consisting of the same logic (EUs or GPU Cores etc).

For desktop, I would argue that smaller iGPUs could help with higher wafer volumes if it meant the price of the chip was better. (ie, sell more processors because the lack of large iGPU allows for a lower retail price)

However, in the case of Intel they are very strong in cpu at this time and therefore they do have some ability to effectively subsidize this extra iGPU. With that mentioned, I still don't really see the point of doing this.

P.S. I will say I do believe having a larger iGPU is not a bad idea for gaming notebooks (re: 1.) notebooks have limited interior volume and better integration may help with this and 2.) swapping in a dGPU after the laptop is purchased is not possible like it is in desktop)

 

[Zodiark1593]

For desktop, I would argue that smaller iGPUs could help with higher wafer volumes if it meant the price of the chip was better. (ie, sell more processors because the lack of large iGPU allows for a lower retail price)

However, in the case of Intel they are very strong in cpu at this time and therefore they do have some ability to effectively subsidize this extra iGPU. With that mentioned, I still don't really see the point of doing this.

P.S. I will say I do believe having a larger iGPU is not a bad idea for gaming notebooks (re: 1.) notebooks have limited interior volume and better integration may help with this and 2.) swapping in a dGPU after the laptop is purchased is not possible like it is in desktop)

The Iris Pro wouldn't be too bad, if it weren't for the fact the only notable laptop that ever uses them is the MacBook line.

 

[cbn]

The Iris Pro wouldn't be too bad, if it weren't for the fact the only notable laptop that ever uses them is the MacBook line.

I've read that the Broadwell-K models will be using GT3 dies.

I wonder if this is partly because Intel had leftovers from lack of laptop adoption?

 

[Enigmoid]

I've read that the Broadwell-K models will be using GT3 dies.

I wonder if this is partly because Intel had leftovers from lack of laptop adoption?

No.
Haswell K and Broadwell K are not the same chip. BW-K appeared on roadmaps well before volume manufacturing of 14 nm began.

 

[VirtualLarry]

Necro, just to note that I dug out my two C-60 (C-70?) Foxconn NanoPCs recently, and I put a trial install of Win10 64-bit on them. Using the April 2016 WIM build, they installed fairly quickly.

Once everything got installed, and settled down a bit, I found that these little boxes (1.0Ghz dual-core, with brief turbo excursions to 1.33Ghz), weren't actually as unusable as I had remembered.

I think that there is some truth to the rumors that Win10 is a more lightweight OS than Win7, in some respects.

It does help to disable as many of the "background apps" as possible, though.

In fact, the performance reminded me very much of my two N3050 Cherry Trail netbooks. Which, although slow, are usable.

I've found that in Win10, even when CPU usage is 99%, it's still usable, it's not like it's locked up like it was back in the XP days. Not sure if that's just because they are doing process time accounting different or the scheduler, or what, but it's pretty slick.

Of course, the SSD helps a bunch too, along with AHCI mode.

Edit: And in keeping with the spirit of the discussion started by the OP, it's interesting that AMD is phasing out their "small core" CPUs.

 

[NTMBK]

Necro, just to note that I dug out my two C-60 (C-70?) Foxconn NanoPCs recently, and I put a trial install of Win10 64-bit on them. Using the April 2016 WIM build, they installed fairly quickly.

Once everything got installed, and settled down a bit, I found that these little boxes (1.0Ghz dual-core, with brief turbo excursions to 1.33Ghz), weren't actually as unusable as I had remembered.

I think that there is some truth to the rumors that Win10 is a more lightweight OS than Win7, in some respects.

It does help to disable as many of the "background apps" as possible, though.

In fact, the performance reminded me very much of my two N3050 Cherry Trail netbooks. Which, although slow, are usable.

I've found that in Win10, even when CPU usage is 99%, it's still usable, it's not like it's locked up like it was back in the XP days. Not sure if that's just because they are doing process time accounting different or the scheduler, or what, but it's pretty slick.

Of course, the SSD helps a bunch too, along with AHCI mode.

Edit: And in keeping with the spirit of the discussion started by the OP, it's interesting that AMD is phasing out their "small core" CPUs.

Windows 7 has a notoriously broken Update service... that thing hogs up vast amounts of CPU time every time you boot up. I know it helped make my C60 netbook unusable.

If Windows 10 fixed that, that old netbook would probably be useable.

 

[coffeemonster]

Necro, just to note that I dug out my two C-60 (C-70?) Foxconn NanoPCs recently, and I put a trial install of Win10 64-bit on them. Using the April 2016 WIM build, they installed fairly quickly.

Once everything got installed, and settled down a bit, I found that these little boxes (1.0Ghz dual-core, with brief turbo excursions to 1.33Ghz), weren't actually as unusable as I had remembered.

I think that there is some truth to the rumors that Win10 is a more lightweight OS than Win7, in some respects.

It does help to disable as many of the "background apps" as possible, though.

In fact, the performance reminded me very much of my two N3050 Cherry Trail netbooks. Which, although slow, are usable.

I've found that in Win10, even when CPU usage is 99%, it's still usable, it's not like it's locked up like it was back in the XP days. Not sure if that's just because they are doing process time accounting different or the scheduler, or what, but it's pretty slick.

Of course, the SSD helps a bunch too, along with AHCI mode.

Edit: And in keeping with the spirit of the discussion started by the OP, it's interesting that AMD is phasing out their "small core" CPUs.

I had a netbook with a 1.7Ghz Brazos that was once used as a desktop stand-in when I had a MOBO failure. It was surprisingly competent. I even installed and ran GW2 on it once to test the igpu. managed 16fps in crowded cities.

 

[DrMrLordX]

edit: necro, post redacted cuz . . . necro.