AMD CPU speculation... and expert conjecture

[jdwii]

Yeah and see how well 8 worked out

 

[juanrga]

2) To compete with 14nm Broadwell on efficiency they have to reduce clocks. Whereas performance increases with frequency is about linear, power increases as cube frequency: P ~ f^3

Where do you come up with this pseudo math?

P = C*F*V^2

http://en.wikipedia.org/wiki/Frequency_scaling

Power remains linear with frequency. The reason the power starts going crazy is they have to bump the core voltage to remain stable.

Voltage is a function of frequency. As explained in my article about power consumption of CPUs, substituting the (very good) approximation V ~ f on the expression for the dynamic power gives the well-known cubic dependence that is quoted in many sites

https://software.intel.com/en-us/articles/the-new-era-of-tera-scale-computing

A concrete case. First the almost linear dependence of voltage with frequency

http://i272.photobucket.com/albums/jj163/idontcare_photo_bucket/Intel%20Core%20i7-2600K/VoltageCurve.png

and then the almost cubic dependence of (dynamic) power with frequency

http://i272.photobucket.com/albums/jj163/idontcare_photo_bucket/Intel%20Core%20i7-2600K/PtotalVccTGHzDataGraph.png

 

[juanrga]

^^^ Yes and as stated its over them caring about tablets and smartphones. I'm not sure what your point is

reading comprehension, not just tablets the whole mobile division. intel has been subsidizing laptops as well. intel is not giving up on tablets and smartphones, only that it can no longer pay to do so.

Intel is the underdog on mobile. They have been barely competing thanks to the incentive program (which I only know for tablets/phones) and the foundry advantage (22nm FinFET against 28nm planar). But they cannot sustain the incentives forever (Intel loses in a quarter what AMD spends on R&D per year) and with ARM migrating to a more efficient 64bit ISA, and with the rest of foundries caughting (Samsung 14FF is already producing chips), Intel has no other option than abandon its pretensions and focus on new markets.

 

[juanrga]

Carrizo seems to support UVD 6, H.265, and inter frame power gating

 

[szatkus]

This is my explanation of AMD recent strategy for Carrizo.

1) AMD is stuck with 28nm node.
2) To compete with 14nm Broadwell on efficiency they have to reduce clocks. Whereas performance increases with frequency is about linear, power increases as cube frequency: P ~ f^3
3) Architectural changes (probably at cache level) will increase the IPC and will about compensate the clocks reduction.
4) Now Excavator doesn't scale well above 3GHz and anything above 35W is canceled because a 65W Carrizo @3GHz would be slower than 95W Kaveri @3.7GHz.

This is more or less the history. Integer cores, FPU, all is about the same than Steamroller.

Apparently all the efforts are put in the iGPU side of Carrrizo. It seems will be more advanced than Tonga or any other discrete GPU.

Looks like they used High Density Library in Carrizo, so clocks should be higher. Looking at Kaveri it will be at least 2.7GHz/3.6GHz at 35W and 3.5GHz/3.9GHz at 65W (if they would release anything like that).

 

[esrever]

HDL should reduce clocks if anything, they do get area scaling from it.

 

[jdwii]

^^^ Yes and as stated its over them caring about tablets and smartphones. I'm not sure what your point is

reading comprehension, not just tablets the whole mobile division. intel has been subsidizing laptops as well. intel is not giving up on tablets and smartphones, only that it can no longer pay to do so.

http://www.maximumpc.com/intel_reports_record_year_despite_massive_drop_mobile_revenue_2015

Oh how will Intel handle record sales. I know by giving their research department more money!!!

 

[esrever]

No, they throw the money into their ultrabook fund and their atom fund to bribe for market share in spaces where they can't easily compete.

 

[jdwii]

^^^ Otherwise their research department as juan showed. Anyways this is an Amd thread but lets not pretend Intel is in some trouble. Even more so lets not pretend Intel is in more trouble then Amd. If i had to be in charge of one of them i would pick Amd over the challenge. I mean even Nvidia puts more money in their research department and they have a easier time atm.

 

[szatkus]

HDL should reduce clocks if anything, they do get area scaling from it.

The tradeoff is peak frequency.

According to AMD with HDL reduce power consumption by 15-30% (probably because wires are shorter and it reduces leakages), so of course they can bump clocks a little. The problem is heat dissipation (because of smaller area), probably that's why we've never seen Excavator@95W on any leaked roadmap.

 

[juanrga]

This is my explanation of AMD recent strategy for Carrizo.

1) AMD is stuck with 28nm node.
2) To compete with 14nm Broadwell on efficiency they have to reduce clocks. Whereas performance increases with frequency is about linear, power increases as cube frequency: P ~ f^3
3) Architectural changes (probably at cache level) will increase the IPC and will about compensate the clocks reduction.
4) Now Excavator doesn't scale well above 3GHz and anything above 35W is canceled because a 65W Carrizo @3GHz would be slower than 95W Kaveri @3.7GHz.

This is more or less the history. Integer cores, FPU, all is about the same than Steamroller.

Apparently all the efforts are put in the iGPU side of Carrrizo. It seems will be more advanced than Tonga or any other discrete GPU.

Looks like they used High Density Library in Carrizo, so clocks should be higher. Looking at Kaveri it will be at least 2.7GHz/3.6GHz at 35W and 3.5GHz/3.9GHz at 65W (if they would release anything like that).

High Performance Libraries are optimized for speed, whereas High Density and Ultra High Density Libraries are optimized for area. And since power is proportional to area, HDL also saves power.

The Carrizo equivalent to the FX-7600P will be clocked at about 2.1/3.4GHz (I guess).

 

[juanrga]

HDL should reduce clocks if anything, they do get area scaling from it.

The tradeoff is peak frequency.

According to AMD with HDL reduce power consumption by 15-30% (probably because wires are shorter and it reduces leakages), so of course they can bump clocks a little. The problem is heat dissipation (because of smaller area), probably that's why we've never seen Excavator@95W on any leaked roadmap.

You forgot the rest of the quote:

The tradeoff is peak frequency. These heavily automated designs won’t be able to clock as high as the older hand drawn designs. AMD believes the sacrifice is worth it however because in power constrained environments (e.g. a notebook) you won’t hit max frequency regardless, and you’ll instead see a 15 - 30% energy reduction per operation. AMD equates this with the power savings you’d get from a full process node improvement.

We won’t see these new libraries and automated designs in Steamroller, but rather its successor in 2014: Excavator.

The reason why we didn't see any Excavator @95W is because HDL reduce clocks and the originally planned 65W Carrizo would be clocked at about 3.3GHz (base). My guess is that the 28SHP (Super High Performance) node is too expensive for AMD and changed it by the cheaper GF28A node which doesn't scale up well above 3GHz.

As a consequence any APU above 35W appearing in old roadmaps was canceled.

A cheaper node plus smaller dies plus socket compatibility (Carrizo/Carrizo-L) reduce costs. This gives me the belief that AMD whole strategy consists on competing against Broadwell on price.

 

[logainofhades]

intel already makes cpu's running android ? But it is not just tablets, laptops and every other ulv cpu has been bleeding money. Not to hard to receive design wins when you just about give away your product for free. forget about android it will stay in mobile until Microsoft releases a arm version of windows, if ever. But i see no need going forward both intel and amd have driven tdp down to a acceptable point to offer full windows comparability on mobile devices. Given a choice most users will take windows over android.

I would take Android over Metro.

 

[cdrkf]

HDL should reduce clocks if anything, they do get area scaling from it.

The tradeoff is peak frequency.

According to AMD with HDL reduce power consumption by 15-30% (probably because wires are shorter and it reduces leakages), so of course they can bump clocks a little. The problem is heat dissipation (because of smaller area), probably that's why we've never seen Excavator@95W on any leaked roadmap.

You forgot the rest of the quote:

The tradeoff is peak frequency. These heavily automated designs won’t be able to clock as high as the older hand drawn designs. AMD believes the sacrifice is worth it however because in power constrained environments (e.g. a notebook) you won’t hit max frequency regardless, and you’ll instead see a 15 - 30% energy reduction per operation. AMD equates this with the power savings you’d get from a full process node improvement.

We won’t see these new libraries and automated designs in Steamroller, but rather its successor in 2014: Excavator.

The reason why we didn't see any Excavator @95W is because HDL reduce clocks and the originally planned 65W Carrizo would be clocked at about 3.3GHz (base). My guess is that the 28SHP (Super High Performance) node is too expensive for AMD and changed it by the cheaper GF28A node which doesn't scale up well above 3GHz.

As a consequence any APU above 35W appearing in old roadmaps was canceled.

A cheaper node plus smaller dies plus socket compatibility (Carrizo/Carrizo-L) reduce costs. This gives me the belief that AMD whole strategy consists on competing against Broadwell on price.

Well I'd actually argue they might be shooting at competing with Broadwell on performance as well *but only up to the 35w power envelope*.

The thing is, when you look at the performance of Intel's 'U' parts (which are turning up in loads of form factors, not just ultrabooks- presumably because they're cheaper), they are 99% of the time dual core and not really that fast (much slower than the previous generation 'M' components for example, albeit with much lower power usage).

The point is if AMD can maintain or slightly exceed the performance of Kaveri quad core implementations, whilst simultaneously reducing power consumption significantly, all of a sudden they could potentially be very competitive with Intel for lower wattage parts. You'll be comparing an Intel dual core + ht vs an AMD quad core part, but if the power consumption is comparable then fair enough (that was the point of their architecture in the first place after all).

The thing is the decisions taken to make the chip competitive at the 15 - 35w power range also mean it won't scale up to higher power designs. So they'd either have to make a second die on a different process node, or not release one (and it looks like the latter is the case). Whilst that's disappointing for desktop users, I'd take a decent AMD quad over a dual core i3 with HT any day if the performance is a bit better *and it has better GPU thrown into the deal*. I actually think that is possible. Kaveri looked much better at low clocks than Trinity / Richland for example (but doesn't higher up).

At the end of the day this looks like they're playing to the strengths of what they have. Carrizo isn't going to win any outright performance awards however I think it has the possibility to be really good if they've done their sums right. Obviously I could be wrong I guess, though I remember low power Kaveri was actually quite impressive, it just didn't scale up enough on the Desktop (something like 90% of the performance of the top part at 45w).

 

[esrever]

I just hope AMD gets more design wins with carrizo than kaveri. Currently most AMD laptops I find at bestbuy are Kabini or Richland. And they are all cheap laptops with terrible design. If AMD doesn't do something about that, they won't be able to really do anything no matter how good their new chips are. Its so hard to find an AMD laptop thats not crappy screen and a decent battery.

 

[logainofhades]

intel already makes cpu's running android ? But it is not just tablets, laptops and every other ulv cpu has been bleeding money. Not to hard to receive design wins when you just about give away your product for free. forget about android it will stay in mobile until Microsoft releases a arm version of windows, if ever. But i see no need going forward both intel and amd have driven tdp down to a acceptable point to offer full windows comparability on mobile devices. Given a choice most users will take windows over android.

I would take Android over Metro.

what do you hate about metro, the tiles, home screen? i assume you talking about phones or tablets

I do not like Metro in phones, tablets, or windows 8. There isn't enough beer in the world to make me like the look of Metro.

 

[logainofhades]

That doesn't fix the crap on phones or tablets. I think a 5yr old could make a better looking UI than those stupid tiles. I am considering trying Win 10, only because support for Win 7 is dying off.

 

[esrever]

tiles seem to look fine to me, maybe you just have things against tiles.

 

[logainofhades]

I find the tiles of metro about as visually appealing as wallpaper from the 1970's.

 

[anxiousinfusion]

The thing is the decisions taken to make the chip competitive at the 15 - 35w power range also mean it won't scale up to higher power designs. So they'd either have to make a second die on a different process node, or not release one (and it looks like the latter is the case). Whilst that's disappointing for desktop users, I'd take a decent AMD quad over a dual core i3 with HT any day if the performance is a bit better *and it has better GPU thrown into the deal*. I actually think that is possible. Kaveri looked much better at low clocks than Trinity / Richland for example (but doesn't higher up).

I'm probably a bit late to the party on this but this write up takes a fascinating look at adaptive clock scaling. Really has a lot to do with why Kaveri shows diminishing returns as frequency goes above 3Ghz. I wouldn't be surprised if Carrizo only increases the aggressiveness of this feature.

 

[jdwii]

I find the tiles of metro about as visually appealing as wallpaper from the 1970's.

Your in the majority with this feeling so don't feel bad, i'm a person who dislikes IOS however i'd use it way more then W8 phones any day.

I HATE the look of metro i KNEW people would to since windows phones never took off.

I was just looking at it the other day and looking at metro also made me feel like the old days were android looked newer. Not only that but the tiles are just an inefficient design even when you shrink them they are bigger then a icon.

 

[jdwii]

Yeah and see how well 8 worked out

^^^ Otherwise their research department as juan showed. Anyways this is an Amd thread but lets not pretend Intel is in some trouble. Even more so lets not pretend Intel is in more trouble then Amd. If i had to be in charge of one of them i would pick Amd over the challenge. I mean even Nvidia puts more money in their research department and they have a easier time atm.

pretending is for children, i knew posting that tidbit about intel would get you and juans panties in a bind. To be honest that was my intention, i was tired of all the amd is dead BS based on juans assumption. intel is not in trouble and as you pointed out they have record profits. But conjecture comparing r&d budgets equals amd is history is short sighted. look at it this way, intel is 95x bigger than amd but there r&d budget is only 10x more, amd allocates more money as a percentage of size towards r&d than intel. is the glass half empty or half full ?

The reality is Intel is spending 10 times the amount as Amd regardless of why means nothing. Back in the day when Amd did win or make a competitive design they were spending more money and they were closer to Intel's budget. Not only that but you can't base things on history and try and claim it can repeat itself without evidence even more so with technology its illogical to do so.

CPU's have gotten more complicated since 1999 and Intel will NOT allow it to happen again. They even came right out and said it. To sum of the athlon destroying pentium 4 thing- Amd was putting WAY more money towards their research and Intel was really doing bad in terms of CPU engineering.

As for the Amd offers more money for research since there company is so small. Well duh if anything i wouldn't really brag about that since Amd is in more markets then Intel atm.
Trying to stay competitive in CPU's and GPU's isn't easy Intel is just pushing for a one design for all approach for their X86 architecture while improving their GPU.

Anyways juan has nothing to do with reality or budgets even if he wished he did. No one here does its facts and spreadsheets anything beyond that is opinions.

 

[jdwii]

To sum up the athlon*^^^ Hate not being able to edit wish we could just create another thread and take this one down

 

[esrever]

Intel can spend as much R&D as it wants but the gains you get aren't as good as they used be able to get just by throwing money at it. Intel spent 5x anyone did on process tech and is still doing it while they are still losing their lead. Its just not going to be easy for intel to keep its lead through R&D alone in the long run. They are slowing down big time and sooner or later, the silicon will just become too volatile that even with intel's spending, they won't get more than a couple percent out of it than the competition who spend a small fraction of what intel spends.

 

[jdwii]

. Back in the day when Amd did win or make a competitive design they were spending more money and they were closer to Intel's budget.

.

wait your like juan pulling numbers out of thin air to fit your narrative, p4 was released in 2008 intels R&D at the time was 7 billion dollars amds was only 1.7bil
amd-http://static.cdn-seekingalpha.com/uploads/2014/8/4/saupload_95687689f29cdcb215eee2a237661f7c.png

intel-http://static.cdn-seekingalpha.com/uploads/2014/8/4/saupload_3802b574dde8184878e8641e7c1765da.png
[/quotemsg]

The Pentium 4 WTF. It as released in 2000.