AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[lilcinw]

But teh HYPEZ!!! We have to assume double IPC [strike]until proven otherwise[/strike].

Seriously though I am hoping that 40% is accurate because anything less and Zen will be universally panned as an also ran. I am not sure that gaming is the best place to look for that difference either.

 

[jimmysmitty]

I agree that all out raw performance is more important to a specific crowd...I fall into that crowd. However, Since AMD has had higher TDP, people have constantly complained about power consumption or heat generation while overclocking everything under the sun.

now a days for most people (except for the ignorants) it's less about power consumption and more about power management - that's what really governs efficiency in consumer parts. pd is old - fact. amd didn't implement a lot of optimization in that area when pd came out, intel didn't much either. right now, both intel and amd have extensive power management as well as power saving tech. combo that with the power savings node shrinks are gonna bring in (as long as glo doesn't fail again), TDP arguments will be moot. they're more or less useless right now - TDP really is a guideline for the oems and parts manufacturers. i like referring to TDP because it gives a general idea about power use. at least it used to before haswell and carrizo came out. it's not hopeless though, review sites like toms have ways to measure processor power consumption (still improving) - so we'll still have good sources for info.

your claims about tdp and overclocking contradict each other, as overclocking is pushing a chip beyond it's factory specs, TDP is one of those specs. heat generation and power consumption depend on the elertrical and physical characteristics of the chip itself.

TDP is technically thermal design. Essentially it dictates what the chip should be able to consume/dissipate in terms of thermal limits.

So...sure...you could somewhat equate that to power...though it is not really entirely accurate.

I agree that the overclocking comments are oxymorons...though you see people with 4.7 GHz intel chips talking about how terrible AMD power consumption is...then they show a power consumption benchmark with a stock, non-k series, i5 and talk about the massive gap in power consumption between their PC way overclocked and an AMD PC.

The main reason is the performance difference. AT stock an i7 will easily take on a overclocked 8350 using less power. The overclocked i7 is just butter on the biscuit.

The truth is that BD is a bad design both performance and thermal/power wise. It is not a horrible CPU but the easiest way to put it is that FX (current) is AMDs Netburst.

Doesn't mean they wont have a decent chip again. I just don;t see them making such a major jump especially when their R&D keeps getting cut and considering the Fury series I am not as hopeful anymore. For the first time in 12 years I have considered buying an nVidia GPU because I feel that AMD is dropping the ball in that department. I didn't feel that way even with the R9 290X although it was a hot, power hungry chip.

Maybe the next few months will prove it wrong, I can only hope. But things look to be the status quot to me. AMD will continue to only challenge Intel at the mid to mid-high mainstream level while Intel continues to reap in the top end.

And again I am hoping AMD is putting more focus towards the server market because I would. Intel certainly does.

 

[cdrkf]

I agree that all out raw performance is more important to a specific crowd...I fall into that crowd. However, Since AMD has had higher TDP, people have constantly complained about power consumption or heat generation while overclocking everything under the sun.

now a days for most people (except for the ignorants) it's less about power consumption and more about power management - that's what really governs efficiency in consumer parts. pd is old - fact. amd didn't implement a lot of optimization in that area when pd came out, intel didn't much either. right now, both intel and amd have extensive power management as well as power saving tech. combo that with the power savings node shrinks are gonna bring in (as long as glo doesn't fail again), TDP arguments will be moot. they're more or less useless right now - TDP really is a guideline for the oems and parts manufacturers. i like referring to TDP because it gives a general idea about power use. at least it used to before haswell and carrizo came out. it's not hopeless though, review sites like toms have ways to measure processor power consumption (still improving) - so we'll still have good sources for info.

your claims about tdp and overclocking contradict each other, as overclocking is pushing a chip beyond it's factory specs, TDP is one of those specs. heat generation and power consumption depend on the elertrical and physical characteristics of the chip itself.

TDP is technically thermal design. Essentially it dictates what the chip should be able to consume/dissipate in terms of thermal limits.

So...sure...you could somewhat equate that to power...though it is not really entirely accurate.

I agree that the overclocking comments are oxymorons...though you see people with 4.7 GHz intel chips talking about how terrible AMD power consumption is...then they show a power consumption benchmark with a stock, non-k series, i5 and talk about the massive gap in power consumption between their PC way overclocked and an AMD PC.

The main reason is the performance difference. AT stock an i7 will easily take on a overclocked 8350 using less power. The overclocked i7 is just butter on the biscuit.

The truth is that BD is a bad design both performance and thermal/power wise. It is not a horrible CPU but the easiest way to put it is that FX (current) is AMDs Netburst.

Doesn't mean they wont have a decent chip again. I just don;t see them making such a major jump especially when their R&D keeps getting cut and considering the Fury series I am not as hopeful anymore. For the first time in 12 years I have considered buying an nVidia GPU because I feel that AMD is dropping the ball in that department. I didn't feel that way even with the R9 290X although it was a hot, power hungry chip.

Maybe the next few months will prove it wrong, I can only hope. But things look to be the status quot to me. AMD will continue to only challenge Intel at the mid to mid-high mainstream level while Intel continues to reap in the top end.

And again I am hoping AMD is putting more focus towards the server market because I would. Intel certainly does.

I agree with you on the FX, on things though, I find it odd you use the Fury series as an example of AMD 'failing' ?!

It didn't *outright surpass* the nVidias biggest and most efficient ever GPU, but it pretty much matches it.

The facts are:

- Fiji is a similar die size to GM200
- Its pretty much equivalent in terms of transistors
- It offers comparable performance to fully enabled GM200 in 1440p and higher.

It's 'failing's are:
- Only 4gb vram (although recent analysis show this is fairly academic as the larger vram pool on the 980ti doesn't give it enough of an edge to achieve playable frame rates at 4k either). It's something that suggests the 980ti will age better, not so much an issue now but also confirmed fixed for next gen.
- The fact that HBM didn't make it outstrip the 980ti (but then I think neither Fury or the 980ti are really bandwidth limited, it will matter more next gen with large more complex gpu's that need the extra throughput).
- Power consumption, whilst much improved on Fiji compared to previous gen products doesn't quite match the efficiency of GM200, but it's actually fairly close*

*note I'll be interested to see if Fury nano improves this or not.

Trust me, if AMD can get as close to Intels latest and greatest with Zen, as they got to Maxwell with fiji in a single generation, I'll be *very impressed*. I think that is 90% of AMD's problem, one way or another things get way over hyped.

@gamerk I agree that 40% ipc improvement may not be across the board. I mean in a few specific instances (like when using specific FMA instructions) BD / PD are significantly faster clock for clock than Phenom II and probably as fast as most of Intels kit. Sadly that isn't the case for most legacy code.

I'd personally hope for an *average* of 40% ipc increase over a range of applications, although that's probably a bit optimistic.

 

[gamerk316]

It didn't *outright surpass* the nVidias biggest and most efficient ever GPU, but it pretty much matches it.

At a higher cost after hyping it half to death, without adequate supply in channel to support those who might otherwise have impulse brought. AMD did EVERYTHING wrong with the launch, and that's why they're continuing the shed share.

Again, AMD is in a position where "good enough" doesn't cut it anymore. They have to actively convince consumers to purchase their products instead of Intel/NVIDIA. And guess what? 95% of the performance at 110% the price isn't going to cut it.

 

[juanrga]

*claimed* higher IPC. I don't think I need to remind anyone how BD got trumpeted here before it's release. Until I see real benchmarks that specifically measure IPC (Gaming versus Intel and FX would be a good place to start), I'm assuming real improvements in the 20% range, not the claimed 40%.

BD family is a speed-demon microarchitecture. Zen is a brainiac architecture. BD was explicity designed to have low-IPC, as any other speed-demon design.

I have zero doubts that Zen will improve IPC. It is very easy to do. Just cut the superpipeline (eliminating extra stages) to hit 5GHz max and automatically you gain IPC, even if the rest of the microarchitecture is unchanged.

The doubt is how much Zen will increase IPC. My early prediction was 40% over Piledriver, which means about 20--30% over Excavator.

 

[juanrga]

I agree that all out raw performance is more important to a specific crowd...I fall into that crowd. However, Since AMD has had higher TDP, people have constantly complained about power consumption or heat generation while overclocking everything under the sun.

now a days for most people (except for the ignorants) it's less about power consumption and more about power management - that's what really governs efficiency in consumer parts. pd is old - fact. amd didn't implement a lot of optimization in that area when pd came out, intel didn't much either. right now, both intel and amd have extensive power management as well as power saving tech. combo that with the power savings node shrinks are gonna bring in (as long as glo doesn't fail again), TDP arguments will be moot. they're more or less useless right now - TDP really is a guideline for the oems and parts manufacturers. i like referring to TDP because it gives a general idea about power use. at least it used to before haswell and carrizo came out. it's not hopeless though, review sites like toms have ways to measure processor power consumption (still improving) - so we'll still have good sources for info.

your claims about tdp and overclocking contradict each other, as overclocking is pushing a chip beyond it's factory specs, TDP is one of those specs. heat generation and power consumption depend on the elertrical and physical characteristics of the chip itself.

TDP is technically thermal design. Essentially it dictates what the chip should be able to consume/dissipate in terms of thermal limits.

So...sure...you could somewhat equate that to power...though it is not really entirely accurate.

I agree that the overclocking comments are oxymorons...though you see people with 4.7 GHz intel chips talking about how terrible AMD power consumption is...then they show a power consumption benchmark with a stock, non-k series, i5 and talk about the massive gap in power consumption between their PC way overclocked and an AMD PC.

The main reason is the performance difference. AT stock an i7 will easily take on a overclocked 8350 using less power. The overclocked i7 is just butter on the biscuit.

The truth is that BD is a bad design both performance and thermal/power wise. It is not a horrible CPU but the easiest way to put it is that FX (current) is AMDs Netburst.

Doesn't mean they wont have a decent chip again. I just don;t see them making such a major jump especially when their R&D keeps getting cut and considering the Fury series I am not as hopeful anymore. For the first time in 12 years I have considered buying an nVidia GPU because I feel that AMD is dropping the ball in that department. I didn't feel that way even with the R9 290X although it was a hot, power hungry chip.

Maybe the next few months will prove it wrong, I can only hope. But things look to be the status quot to me. AMD will continue to only challenge Intel at the mid to mid-high mainstream level while Intel continues to reap in the top end.

And again I am hoping AMD is putting more focus towards the server market because I would. Intel certainly does.

I agree with you on the FX, on things though, I find it odd you use the Fury series as an example of AMD 'failing' ?!

It didn't *outright surpass* the nVidias biggest and most efficient ever GPU, but it pretty much matches it.

The facts are:

- Fiji is a similar die size to GM200
- Its pretty much equivalent in terms of transistors
- It offers comparable performance to fully enabled GM200 in 1440p and higher.

It's 'failing's are:
- Only 4gb vram (although recent analysis show this is fairly academic as the larger vram pool on the 980ti doesn't give it enough of an edge to achieve playable frame rates at 4k either). It's something that suggests the 980ti will age better, not so much an issue now but also confirmed fixed for next gen.
- The fact that HBM didn't make it outstrip the 980ti (but then I think neither Fury or the 980ti are really bandwidth limited, it will matter more next gen with large more complex gpu's that need the extra throughput).
- Power consumption, whilst much improved on Fiji compared to previous gen products doesn't quite match the efficiency of GM200, but it's actually fairly close*

*note I'll be interested to see if Fury nano improves this or not.

Trust me, if AMD can get as close to Intels latest and greatest with Zen, as they got to Maxwell with fiji in a single generation, I'll be *very impressed*. I think that is 90% of AMD's problem, one way or another things get way over hyped.

But AMD has used HBM to close the gap with Nvidia. Once Nvidia moves from GDDR5 to HBM2, the gap will increase again. AMD doesn't have a similar trick to caught Intel on CPUs.

 

[Cazalan]

If one assumes Greenland is just a die shrunk Tonga then you could make that assumption. Of course that would imply the GPU division has just been twiddling their thumbs the last 2 years.

 

[juanrga]

The lack of resources of AMD has halted the GPU division. AMD is putting all the money on Zen. Moreover, we have what Papermaster said at FAD 2015 about the GPUs of 2016.

 

[de5_Roy]

The lack of resources of AMD has halted the GPU division. AMD is putting all the money on Zen. Moreover, we have what Papermaster said at FAD 2015 about the GPUs of 2016.

i don't remember that one. what did he say at that time?

 

[juanrga]

The lack of resources of AMD has halted the GPU division. AMD is putting all the money on Zen. Moreover, we have what Papermaster said at FAD 2015 about the GPUs of 2016.

i don't remember that one. what did he say at that time?

He did split GCN evolution into three periods: 2012 ("launch"), 2013--2015 ("enhancements"), and 2016 ("optimizations"). And he detailed each period. For 2016, he only mentioned "2x energy efficiency" and "FinFET technology". The efficiency gain¹ claimed by AMD is just what one expects from 28nm-->16nm. Therefore, GCN will remain essentially unchanged next year.

¹ "Based on internal AMD estimates for 2016 Graphics Core Next GPU compared to previous generation GPU".

 

[de5_Roy]

The lack of resources of AMD has halted the GPU division. AMD is putting all the money on Zen. Moreover, we have what Papermaster said at FAD 2015 about the GPUs of 2016.

i don't remember that one. what did he say at that time?

He did split GCN evolution into three periods: 2012 ("launch"), 2013--2015 ("enhancements"), and 2016 ("optimizations"). And he detailed each period. For 2016, he only mentioned "2x energy efficiency" and "FinFET technology". The efficiency gain¹ claimed by AMD is just what one expects from 28nm-->16nm. Therefore, GCN will remain essentially unchanged next year.

¹ "Based on internal AMD estimates for 2016 Graphics Core Next GPU compared to previous generation GPU".

i don't see any link to an official statement.
i am still confused about what you said earlier, above. are you saying that papermaster had claimed that amd had put all it's money on zen, or, amd halted it's gpu division due to lack of resources, or papermaster had said that amd halted it's gpu division .... or whatever papermaster said in fad2015 actually directly relate to any of the aforementioned? your post is too unclear. seemingly a few unrelated bits pasted together. i'd like to see something official, and with a link to it. your statement about efficiency gain fails to relate to what you claimed papermaster had said earlier. need more, clearer info.

 

[Cazalan]

GCN has had untapped power for years. It is just finally coming to fruition with DX12 and asynchronous compute.

Maxwell is utterly incapable of performing asynchronous compute without heavy reliance on slow context switching.

 

[Cazalan]

i don't see any link to an official statement.
i am still confused about what you said earlier, above. are you saying that papermaster had claimed that amd had put all it's money on zen, or, amd halted it's gpu division due to lack of resources, or papermaster had said that amd halted it's gpu division .... or whatever papermaster said in fad2015 actually directly relate to any of the aforementioned? your post is too unclear. seemingly a few unrelated bits pasted together. i'd like to see something official, and with a link to it. your statement about efficiency gain fails to relate to what you claimed papermaster had said earlier. need more, clearer info.

Lisa Su, Q2-2015 earnings call.

We have several segments that we're investing in for long-term growth and the Computing and the Graphics IP are critical to make that happen.

The critical IP is being invested in. Plain and simple.

 

[juanrga]

GCN has had untapped power for years. It is just finally coming to fruition with DX12 and asynchronous compute.

Maxwell is utterly incapable of performing asynchronous compute without heavy reliance on slow context switching.

Is not that quote from the same AMD representative that said that Fury X memory cannot be overclocked?

 

[-Fran-]

Is not that quote from the same AMD representative that said that Fury X memory cannot be overclocked?

As long as you can mess with the feeding signal, any circuit can be "overclocked". If AMD doesn't want HBM to be overclocked, is not because you can't, but for another reason.

Most probably because they don't have much faith in this first massive tape out of HBM or something?

But why revisit the past again? Anything recent on what they're planning? It's not like AMD isn't bipolar on their speeches and roadmaps, haha.

Cheers!

 

[Cazalan]

Is not that quote from the same AMD representative that said that Fury X memory cannot be overclocked?

As long as you can mess with the feeding signal, any circuit can be "overclocked". If AMD doesn't want HBM to be overclocked, is not because you can't, but for another reason.

Most probably because they don't have much faith in this first massive tape out of HBM or something?

But why revisit the past again? Anything recent on what they're planning? It's not like AMD isn't bipolar on their speeches and roadmaps, haha.

Cheers!

There is always going to be some gray area between what the vendor wants to officially support and what the hardware can do. And reference cards have different requirements than custom cards. We're long past the days of fixed function hardware. Almost every aspect is configurable these days, that's what gives the enthusiast overclocker things to do. Like the guy that got HBM up to 1Ghz with a heavily modified card. And you got the people learning how to flash the bios and enable compute units that were factory disabled.

Now we're seeing some DX12 titles coming to life and it's quite promising.

 

[8350rocks]

http://www.kitguru.net/components/graphic-cards/anton-shilov/amd-we-are-actively-promoting-usage-of-hbm-and-do-not-collect-royalties/

Some interesting details in there regarding HBM2, licensing, etc.

 

[de5_Roy]

AMD Athlon X4 880K CPU spotted
http://www.cpu-world.com/news_2015/2015090601_AMD_Athlon_X4_880K_APU_spotted.html

 

[Cazalan]

Steppe Eagle gets some design wins with Fujitsu in secure thin clients.

http://finance.yahoo.com/news/amd-embedded-g-series-soc-120000120.html

 

[Cazalan]

AMD getting a cash injection?!?

http://semiaccurate.com/2015/09/08/behind-amds-new-funding/

 

[salgado18]

AMD getting a cash injection?!?

http://semiaccurate.com/2015/09/08/behind-amds-new-funding/

The article is for subscribers only.

What kind of cash injection is this? Enough for them to finish Zen and next GPUs? Enough to go beyond that? Enough to ensure it won't break in 2019?

 

[jimmysmitty]

AMD getting a cash injection?!?

http://semiaccurate.com/2015/09/08/behind-amds-new-funding/

The article is for subscribers only.

What kind of cash injection is this? Enough for them to finish Zen and next GPUs? Enough to go beyond that? Enough to ensure it won't break in 2019?

That is very annoying actually. They probably make plenty of money on the ads alone.

Either way the tag line below it is interesting and suggest someone who you might not expect but would make sense. Maybe a competitor of sorts? Intel possibly? That wouldn't be too far fetched to me since it was Microsoft who helped to save Apple.

Either way AMD really needs it....

 

[Cazalan]

I don't have a sub but if it's newsworthy it must be a substantial amount.

 

[jimmysmitty]

As I said I am thinking Intel because they have the extra cash. Or Samsung but I don't see why. I doubt Apple because, well they like to hoard their money.

It is hard to say but I am sure we will hear about it in a few weeks. That kind of news wont stay subscriber only, which again I think is stupid as they get plenty of ad money.

 

[Ranth]

As far as I know there are no adds on semiaccurate.
"Thank you to our Subscribers, past and present. You are appreciated. You are what keeps SemiAccurate going, what allows us to maintain our journalism, what keeps us ad-free, what allows us to tell it like it is, it is still just you. You, the reader and subscriber, we thank you.

If you want to know more about subscriptions, both free and paid, the information can be found here."