AMD CPU speculation... and expert conjecture

[gamerk316]

AMD held APU14 today in China. According to this talk, AMD plans to not move away from desktop. Rather, talks even about growth. Who here was saying that AMD plans to abandon desktop?

I have it on VERY good authority that a new HEDT platform is in the works...my latest information says it will be a little way out, but will offer dCPU and APU options...

Unfortunately, I cannot share more than that...be PATIENT.

(Just FYI for those who would want to know: ARM is not expected to take a significant portion of desktop marketshare...)

Quick question, open to all. Isn't a normal feature of an NDA that you can't even reveal that you are under an NDA, never mind give some not-so-subtle hints on what the NDA is about?

Depends on the NDA, but when it has to deal with new products, typically yes, it is the norm for the NDA to cover revealing the product in question. In extreme cases, even the fact you are under an NDA is covered by the NDA.

 

[gamerk316]

i wanted to discuss about arm and x86 cores being swappable on the same platform. from the s/a articles, it looks like the biggest obstacle is memory access i.e. how an x86 core and arm v8 core would access memory. do they mean cache access or system memory? i assumed both arm and x86 cores in project skybridge will come with their own L2 caches and they will need some kind of hardware virtualization to make the socket compatibility real and software virtualization to make o.s. compatibility real. anyone know how that is possible? is it possible without virtualization? i mean, an always-on, real time hardware and software virtualization seem expensive in terms of power use.

Hence the problem; you can either do everything to support both platforms in HW, and have a REALLY power hungry beast of a CPU, or you can do it in SW, which is going to suck performance.

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

 

[de5_Roy]

Spoiler

Hence the problem; you can either do everything to support both platforms in HW, and have a REALLY power hungry beast of a CPU, or you can do it in SW, which is going to suck performance.

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

i think amd is trying to build a "universal uncore" from the looks of things. they're doing it before moving on to a universal core, possibly or inevitably. amd is taking a "we'll solve it when we get there" approach.
arm's (32-bit) memory access seems to be less complex than x86 (64 bit being less than 32 bit). all the arm 64 bit OoO cores have L3 cache (e.g. cyclone), so skybridge's uncore might have L3 cache as well. but beema has no L3 cache, has igpu. then there's the entirely different MMU. i wonder how can amd build a core-agnostic uncore.

 

[gamerk316]

Hey look, we're back!

And also: http://www.fudzilla.com/home/item/34769-amd-plans-new-architecture-for-2016

Kinda proves 8350's point of AMD working on a new HEDT core.

 

[juanrga]

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

No. AMD is working in universal UNcore.

AMD is working in two high-performance coreS: K12 and its x86 "sister".

I expect K12 core performance to outperform Piledriver/Steamroller by a large margin and be close to Haswell IPC.

 

[juanrga]

Unsurprising Q1 results: Intel gains GPU market share. AMD and Nvidia don't

http://vr-zone.com/articles/jpr-nvidia-amd-big-losers-q1-2014/77587.htmlQ1

This chart is very interesting

Spoiler

The sales of desktop dGPU drops, but don't correspond to a similar drop in the sales of desktop PCs. This implies that many new desktops PCs are using integrated graphics: aka APUs. However, there is a correlation in the notebook case, which implies that notebooks are being replaced by tablets and the like.

I mentioned before some technical reasons why dGPU will be killed in future. But long-term unsuitable cost of development is an economic reason why dGPUs have no future. No company will spend millions on developing new discrete cards when volume sales will be too low to recover inversion made.

 

[8350rocks]

AMD held APU14 today in China. According to this talk, AMD plans to not move away from desktop. Rather, talks even about growth. Who here was saying that AMD plans to abandon desktop?

I have it on VERY good authority that a new HEDT platform is in the works...my latest information says it will be a little way out, but will offer dCPU and APU options...

Unfortunately, I cannot share more than that...be PATIENT.

(Just FYI for those who would want to know: ARM is not expected to take a significant portion of desktop marketshare...)

Quick question, open to all. Isn't a normal feature of an NDA that you can't even reveal that you are under an NDA, never mind give some not-so-subtle hints on what the NDA is about?

I answered your question above...

 

[Cazalan]

i think amd is trying to build a "universal uncore" from the looks of things. they're doing it before moving on to a universal core, possibly or inevitably. amd is taking a "we'll solve it when we get there" approach.
arm's (32-bit) memory access seems to be less complex than x86 (64 bit being less than 32 bit). all the arm 64 bit OoO cores have L3 cache (e.g. cyclone), so skybridge's uncore might have L3 cache as well. but beema has no L3 cache, has igpu. then there's the entirely different MMU. i wonder how can amd build a core-agnostic uncore.

For this to be out next year the universal uncore has already been designed. They're both 64bit so that made it easier.

 

[8350rocks]

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

No. AMD is working in universal UNcore.

AMD is working in two high-performance coreS: K12 and its x86 "sister".

I expect K12 core performance to outperform Piledriver/Steamroller by a large margin and be close to Haswell IPC.

LOL...whatever...performance will improve, and it may be on the same level as haswell...but...you are just not accurate. Uncores are not even existent.

 

[juanrga]

No. AMD is working in universal UNcore.

AMD is working in two high-performance coreS: K12 and its x86 "sister".

I expect K12 core performance to outperform Piledriver/Steamroller by a large margin and be close to Haswell IPC.

LOL...whatever...performance will improve, and it may be on the same level as haswell...but...you are just not accurate. Uncores are not even existent.

I will be happy being only semiaccurate

P.S.: I like your remark about uncores being inexistent.

 

[juanrga]

Hey look, we're back!

And also: http://www.fudzilla.com/home/item/34769-amd-plans-new-architecture-for-2016

Kinda proves 8350's point of AMD working on a new HEDT core.

The link just repeat what has been known from more than one week and posted here. E.g. I already posted here that the new big core abandon the CMT architecture.

Fudzilla get some points completely wrong. More accurate info was given by Anand a week ago

http://www.anandtech.com/show/7991/amd-is-also-working-on-a-new-64bit-x86-core

 

[blackkstar]

ARM will win in HPC and Servers. It won't unseat x86 desktop and laptops.

More popular laptop at Amazon is powered by ARM.

You will notice that some people have fallen for the logical fallacy of seeing AMD talk about ARM and x86, AMD saying ARM will win, and then assuming that AMD is only talking about the fight between x86 and ARM and that ARM will unseat x86.

As you can see from the AMD slide referencing projected market shares, ARM is planning on taking a massive chunk away from the "other" category and not the x86 one.

x86 --> [ x86 + ARM ] --> ARM

AMD is now in ^^^ second stage of the transition.

If you draw trend lines and extend the very graph you posted from AMD's very own marketing slides done by their own projections, you'll see that ARM and x86 will find parity in growth and contraction. In fact, x86 has basically contracted as far as AMD thinks it will.

If you don't believe me, look at the growth rates of ARM between 2005, 2010, and 2015.

2005 - 2010: almost doubled
2010 - 2015: almost doubled again
2015 - 2018: flat, nothing growing

Look at the graph you posted again. If AMD agreed with you, ARM and x86 wouldn't be finding parity in growth while maintaining what AMD projects to be damn near equal market share.

ARM is going to stop growing so rapidly like you think. We've already reached the point with mobile devices where existing products are "fast enough" and people no longer need to spend $700 (with provider eating most of the cost) when they can do something and buy Moto E for $129 and get a good experience still.

As I've been saying, once this drop off hits, ARM will slow down massively. The only new market that remains for them is mobile and servers/HPC. I see them doing well in servers and HPC but you're over-exaggerating mobile laptop wins.

A quick search in the computer section for laptops @ amazon and sorting by popularity shows the most popular laptops to be:

1. Toshiba Satelite with Celeron
2. Chromebook
3. HP with AMD E2 APU
4. Chromebook
5. HP with AMD A4 APU
6. Apple iBook with G4
7. Dell with Pentium
8. Dell with Celeron
9. Acer with Celeron
10. Dell with Core 2 Duo

But my point is that you are doing the same thing the apple kids did when they were losing the market share war. They compare a few iOS models to a few Android models and then go "lol Apple is fine it still has more market share than Galaxy S2!" without realizing that iOS is already a minority (at that point) compared to every single Android phone.

My point is that even with two chrome books doing well on Amazon (and an iBook G4 and Core 2 Duo laptop are doing well too!), that's still a little drip into the ocean of x86 laptops, because Chromebooks are not as diverse so the sales are concentrated in a few models. So when you compare and say that Chromebook is doing just as well as some x86 models, you're forgetting that there's a handful of Chromebooks and there's an endless army of x86 CPU laptops.

EDIT: Juan, this is it: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=http%3A%2F%2Fnews.mydrivers.com%2F1%2F305%2F305092.htm

Lienhard: Last year we introduced a Piledriver (hammers) architecture, and achieved good market performance, there are about 30% growth, Kaveri also used Steamroller updates (excavator) architecture. FX Series of high-performance market positioning in the future will, within two years you will definitely see an update.

HEDT is coming back. Either jumbo APUs or dCPUs with dGPUs with HSA butter.

 

[jdwii]

I think it's impossible for certain users on this site not even just talking about juan to understand the marketing terms and how it works or what "emerging markets" is.
"An emerging market is a nation with social or business activity in the process of rapid growth and industrialization."
Now that's the industry standard for the definition, to state its replacing X86 would be a lie if their is still just as many X86 devices if not more this year. So again Arm will replace X86 in low-end servers and possibly mid-range markets or more importantly emerging markets

 

[jimmysmitty]

After Anand analysis showed that Cyclone architecture is "desktop class" and can compete with big Intel cores, an Intel fan did run a comparison between cyclone core and sandy bridge core and got that both are comparable on IPC terms.

http://forums.anandtech.com/showpost.php?p=35511711&postcount=1

The results are unsurprising because Cyclone is already ahead of Piledriver:

Cyclone IPC > A57 IPC
A57 IPC > Jaguar IPC
Jaguar IPC > Piledriver IPC

That is another useless benchmark. What OS? What CPU did he use to compare it to? Why did he not clock it to the same clock speed as that can be done? What other platform parts were used?

I can pull and post a bunch of numbers but without the specifics it is pointless. Why does THG have a page showing the test setup for every benchmark including each component used, which OS and what software they will use? Because it shows it is a fair test on an even platform level.

Sure they will sometimes throw in a $1K CPU tested against a lower priced CPU but that is more for reference. Still we know all their current tests run on Windows 8.1 with whatever software/games they are currently using. That allows us to judge just the results instead of wondering, like with Java Script benchmarks, if something else is changing the numbers.

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

No. AMD is working in universal UNcore.

AMD is working in two high-performance coreS: K12 and its x86 "sister".

I expect K12 core performance to outperform Piledriver/Steamroller by a large margin and be close to Haswell IPC.

So by 2016 AMD plans to have Haswell level performance in their CPUs when Intel will be moving to Sky Lake which is uArch change and as well supposed to be a massive iGPU boost. As well it will have all kinds of fun features such as support for Thunderbolt v3 (up to 40Gbps), DDR4, SATA Express and PCIe 4.0. Now this is all rumored but I would not be surprised.

AMD is not making the smart choice in my eyes as Intel will easily keep ahead of them. AMD is making the choice that allows them to make money faster instead. Smart in some ways but not in others. They are not going to grab as much market share as they want.

Again, one other issue with the move or want to move to ARM is software. Software will need to be rewritten and that costs money. Now why hasn't Linux taken over in this recent economic down turn when it is free vs the super expensive Windows? Windows works with pretty much everything you can buy off the shelf. The rest needs to be rewritten for Linux. Same applies for ARM.

Until the major software devs decide to rewrite their software for ARM, it will never take off on desktops or servers.

And no, I do not count Android as that is a horrible OS.

 

[juanrga]

If you draw trend lines and extend the very graph you posted from AMD's very own marketing slides done by their own projections, you'll see that ARM and x86 will find parity in growth and contraction. In fact, x86 has basically contracted as far as AMD thinks it will.

If you don't believe me, look at the growth rates of ARM between 2005, 2010, and 2015.

2005 - 2010: almost doubled
2010 - 2015: almost doubled again
2015 - 2018: flat, nothing growing

Look at the graph you posted again. If AMD agreed with you, ARM and x86 wouldn't be finding parity in growth while maintaining what AMD projects to be damn near equal market share.

ARM is going to stop growing so rapidly like you think. We've already reached the point with mobile devices where existing products are "fast enough" and people no longer need to spend $700 (with provider eating most of the cost) when they can do something and buy Moto E for $129 and get a good experience still.

As I've been saying, once this drop off hits, ARM will slow down massively. The only new market that remains for them is mobile and servers/HPC. I see them doing well in servers and HPC but you're over-exaggerating mobile laptop wins.

A quick search in the computer section for laptops @ amazon and sorting by popularity shows the most popular laptops to be:

1. Toshiba Satelite with Celeron
2. Chromebook
3. HP with AMD E2 APU
4. Chromebook
5. HP with AMD A4 APU
6. Apple iBook with G4
7. Dell with Pentium
8. Dell with Celeron
9. Acer with Celeron
10. Dell with Core 2 Duo

But my point is that you are doing the same thing the apple kids did when they were losing the market share war. They compare a few iOS models to a few Android models and then go "lol Apple is fine it still has more market share than Galaxy S2!" without realizing that iOS is already a minority (at that point) compared to every single Android phone.

My point is that even with two chrome books doing well on Amazon (and an iBook G4 and Core 2 Duo laptop are doing well too!), that's still a little drip into the ocean of x86 laptops, because Chromebooks are not as diverse so the sales are concentrated in a few models. So when you compare and say that Chromebook is doing just as well as some x86 models, you're forgetting that there's a handful of Chromebooks and there's an endless army of x86 CPU laptops.

EDIT: Juan, this is it: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=http%3A%2F%2Fnews.mydrivers.com%2F1%2F305%2F305092.htm

Lienhard: Last year we introduced a Piledriver (hammers) architecture, and achieved good market performance, there are about 30% growth, Kaveri also used Steamroller updates (excavator) architecture. FX Series of high-performance market positioning in the future will, within two years you will definitely see an update.

HEDT is coming back. Either jumbo APUs or dCPUs with dGPUs with HSA butter.

Pay close attention to the chart. From 2015 to 2018 ARM continues growing whereas x86 continue falling. Of course, the rate is very different than in previous years: market evolution is not linear.

And yes, AMD agrees with me about what arch. will win in the long run

check also the part that says OEM, ODMs, and large customers want ARM to win. I already explained why the entire industry wan to move away from x86.

If you check AMD last conference, Keller predicted an wide industry move away from x86 to ARM.


The ARM laptop has been in top100 best seller list during 571 days. The Toshiba Satelite with Celeron only during 21 days. The former has 4706 reviews and 76% like, whereas the Celeron has only 7 reviews and 33% like.

My point was that ARM has been already doing well in laptops.

I am unsure why the same news about FX comeback is mentioned again. Someone care to explain?

In reality it is not a comeback, because AMD extended the current Piledriver FX series towards 2015, whereas was preparing a replacement (one not based in modules) for 2016

I already mentioned here aspects of the new FX, such as the number of cores. In fact I did it twice:

http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-263#13305301

http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-265#13318543

 

[juanrga]

That is another useless benchmark. What OS? What CPU did he use to compare it to? Why did he not clock it to the same clock speed as that can be done? What other platform parts were used?

I can pull and post a bunch of numbers but without the specifics it is pointless. Why does THG have a page showing the test setup for every benchmark including each component used, which OS and what software they will use? Because it shows it is a fair test on an even platform level.

The results that the Intel fan found by himself are entirely compatible with other benchmarks and with what we know from the architectures (a "desktop class" chip scoring as a desktop chip). Thus, unsurprisingly many Intel fans in the thread liked the results and praised Apple by its good work.

In fact, if you read the Anand review link that I gave before, he mentions that neither the hardware surrounding the cyclone chip on A7 nor the current software allow the core to develop its full potential:

Cyclone is a bold move by Apple, but not one that is without its challenges. I still find that there are almost no applications on iOS that really take advantage of the CPU power underneath the hood. More than anything Apple needs first party software that really demonstrates what's possible. [...] The other problem I see is that although Cyclone is incredibly forward looking, it launched in devices with only 1GB of RAM. It's very likely that you'll run into memory limits before you hit CPU performance limits if you plan on keeping your device for a long time.

I don't know why the Intel fan did clock his Sandy Bridge i7 at 1.6GHz, but this is irrelevant for what is being discussed here, because if he had downclocked to the same 1.3GHz than the Apple chip then the benchmarks would look even better for ARM.

So by 2016 AMD plans to have Haswell level performance in their CPUs when Intel will be moving to Sky Lake which is uArch change and as well supposed to be a massive iGPU boost. As well it will have all kinds of fun features such as support for Thunderbolt v3 (up to 40Gbps), DDR4, SATA Express and PCIe 4.0. Now this is all rumored but I would not be surprised.

AMD is not making the smart choice in my eyes as Intel will easily keep ahead of them. AMD is making the choice that allows them to make money faster instead. Smart in some ways but not in others. They are not going to grab as much market share as they want.

Again, one other issue with the move or want to move to ARM is software. Software will need to be rewritten and that costs money. Now why hasn't Linux taken over in this recent economic down turn when it is free vs the super expensive Windows? Windows works with pretty much everything you can buy off the shelf. The rest needs to be rewritten for Linux. Same applies for ARM.

Until the major software devs decide to rewrite their software for ARM, it will never take off on desktops or servers.

And no, I do not count Android as that is a horrible OS.

I shared my expectations for the new cores. Other people claims that they will be faster than Skylake cores. AMD can do something completely different.

In any case Skylake CPU will be a small evolutionary improvement over Broadwell CPU and this again over Haswell CPU. Matching Haswell CPU performance is enough, because then Skylake CPU would be only 10-15% ahead.

Yes Intel will bring massive iGPU, but AMD will do as well. Why do you believe AMD will be offering HBM? I already mentioned that AMD could pack 2048 GCN cores in an 14nm APU, with dual-channel HBM 1.0 providing enough BW.

AMD will already bring DDR4 support the next, year thus I fail to see your point about mentioning DDR4 for skylake.

The desktop software is a chicken-egg problem software developers will port to ARM when there enough hardware and hardware makers will make ARM desktops when there is enough software. This has always happened. The software that I use and my friends use and my colleagues use and other people use has already been ported. Thus no problem here.

 

[colinp]

I answered your question above...

My apologies, been a bit busy today. My BS-Ometer has been gyrating so energetically that it took off, and I had to chase it around the office all day.

When you say words to the effect of, "I know HEDT is coming, but I can't tell you any more detail," what do you mean by "can't?" Do you mean:
1) You know more but you aren't telling?
2) You can't say more because you don't know more?
3) Something else?

 

[gamerk316]

I answered your question above...

My apologies, been a bit busy today. My BS-Ometer has been gyrating so energetically that it took off, and I had to chase it around the office all day.

When you say words to the effect of, "I know HEDT is coming, but I can't tell you any more detail," what do you mean by "can't?" Do you mean:
1) You know more but you aren't telling?
2) You can't say more because you don't know more?
3) Something else?

He's implied he's under an NDA.

 

[8350rocks]

I answered your question above...

My apologies, been a bit busy today. My BS-Ometer has been gyrating so energetically that it took off, and I had to chase it around the office all day.

When you say words to the effect of, "I know HEDT is coming, but I can't tell you any more detail," what do you mean by "can't?" Do you mean:
1) You know more but you aren't telling?
2) You can't say more because you don't know more?
3) Something else?

I do know more, but am not permitted to say...

 

[8350rocks]

If you draw trend lines and extend the very graph you posted from AMD's very own marketing slides done by their own projections, you'll see that ARM and x86 will find parity in growth and contraction. In fact, x86 has basically contracted as far as AMD thinks it will.

If you don't believe me, look at the growth rates of ARM between 2005, 2010, and 2015.

2005 - 2010: almost doubled
2010 - 2015: almost doubled again
2015 - 2018: flat, nothing growing

Look at the graph you posted again. If AMD agreed with you, ARM and x86 wouldn't be finding parity in growth while maintaining what AMD projects to be damn near equal market share.

ARM is going to stop growing so rapidly like you think. We've already reached the point with mobile devices where existing products are "fast enough" and people no longer need to spend $700 (with provider eating most of the cost) when they can do something and buy Moto E for $129 and get a good experience still.

As I've been saying, once this drop off hits, ARM will slow down massively. The only new market that remains for them is mobile and servers/HPC. I see them doing well in servers and HPC but you're over-exaggerating mobile laptop wins.

A quick search in the computer section for laptops @ amazon and sorting by popularity shows the most popular laptops to be:

1. Toshiba Satelite with Celeron
2. Chromebook
3. HP with AMD E2 APU
4. Chromebook
5. HP with AMD A4 APU
6. Apple iBook with G4
7. Dell with Pentium
8. Dell with Celeron
9. Acer with Celeron
10. Dell with Core 2 Duo

But my point is that you are doing the same thing the apple kids did when they were losing the market share war. They compare a few iOS models to a few Android models and then go "lol Apple is fine it still has more market share than Galaxy S2!" without realizing that iOS is already a minority (at that point) compared to every single Android phone.

My point is that even with two chrome books doing well on Amazon (and an iBook G4 and Core 2 Duo laptop are doing well too!), that's still a little drip into the ocean of x86 laptops, because Chromebooks are not as diverse so the sales are concentrated in a few models. So when you compare and say that Chromebook is doing just as well as some x86 models, you're forgetting that there's a handful of Chromebooks and there's an endless army of x86 CPU laptops.

EDIT: Juan, this is it: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=http%3A%2F%2Fnews.mydrivers.com%2F1%2F305%2F305092.htm

Lienhard: Last year we introduced a Piledriver (hammers) architecture, and achieved good market performance, there are about 30% growth, Kaveri also used Steamroller updates (excavator) architecture. FX Series of high-performance market positioning in the future will, within two years you will definitely see an update.

HEDT is coming back. Either jumbo APUs or dCPUs with dGPUs with HSA butter.

Pay close attention to the chart. From 2015 to 2018 ARM continues growing whereas x86 continue falling. Of course, the rate is very different than in previous years: market evolution is not linear.

And yes, AMD agrees with me about what arch. will win in the long run

check also the part that says OEM, ODMs, and large customers want ARM to win. I already explained why the entire industry wan to move away from x86.

If you check AMD last conference, Keller predicted an wide industry move away from x86 to ARM.


The ARM laptop has been in top100 best seller list during 571 days. The Toshiba Satelite with Celeron only during 21 days. The former has 4706 reviews and 76% like, whereas the Celeron has only 7 reviews and 33% like.

My point was that ARM has been already doing well in laptops.

I am unsure why the same news about FX comeback is mentioned again. Someone care to explain?

In reality it is not a comeback, because AMD extended the current Piledriver FX series towards 2015, whereas was preparing a replacement (one not based in modules) for 2016

I already mentioned here aspects of the new FX, such as the number of cores. In fact I did it twice:

http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-263#13305301

http://www.tomshardware.co.uk/forum/352312-28-steamroller-speculation-expert-conjecture/page-265#13318543

Quick question, in 2035 when ARM still has not "won" are you still going to screen shot that slide?

 

[Cazalan]

Death always wins in the end too, but we don't have to repeat it every day ad nauseam.

What's that about 20 times now the same slide has been linked? The forums really need an ignore button.

 

[de5_Roy]

AMD plans new architecture for 2016
http://www.fudzilla.com/home/item/34769-amd-plans-new-architecture-for-2016
fudzilla's back up, so reposted gamerk's link. hints at SMT design. but no confirmation... more like a logical speculation, which, considers bd's module design ineffective. here's another "logical conjecture" if this SMT hint holds water - 3 integer ALUs per core like sandy/ivy bridge.

AMD’s first 20nm APU is Nolan
http://www.fudzilla.com/home/item/34767-amd%E2%80%99s-first-20nm-apu-is-nolan
looks squarely aimed at consumer segment.

 

[wh3resmycar]

of those 8 billion ARM chips shipped a lot of those came from smart phones. ask any ordinary smart phone user and they don't give a frack about ARM and x86. and when people are done moving from feature phones to smart phones, that "shipped" number will dwindle.

 

[noob2222]

After Anand analysis showed that Cyclone architecture is "desktop class" and can compete with big Intel cores, an Intel fan did run a comparison between cyclone core and sandy bridge core and got that both are comparable on IPC terms.

http://forums.anandtech.com/showpost.php?p=35511711&postcount=1

The results are unsurprising because Cyclone is already ahead of Piledriver:

Cyclone IPC > A57 IPC
A57 IPC > Jaguar IPC
Jaguar IPC > Piledriver IPC

That is another useless benchmark. What OS? What CPU did he use to compare it to? Why did he not clock it to the same clock speed as that can be done? What other platform parts were used?

I can pull and post a bunch of numbers but without the specifics it is pointless. Why does THG have a page showing the test setup for every benchmark including each component used, which OS and what software they will use? Because it shows it is a fair test on an even platform level.

Sure they will sometimes throw in a $1K CPU tested against a lower priced CPU but that is more for reference. Still we know all their current tests run on Windows 8.1 with whatever software/games they are currently using. That allows us to judge just the results instead of wondering, like with Java Script benchmarks, if something else is changing the numbers.

Which is why I can say, without seeing a single diagram or piece of information on the core design, that I know AMD's universal core approach is going to suck in performance terms. The sacrifices in design needed to support both CPU architectures is going to sap the performance from both. I figure AMD is shooting for a low-performance core to undercut Atom in the marketplace.

No. AMD is working in universal UNcore.

AMD is working in two high-performance coreS: K12 and its x86 "sister".

I expect K12 core performance to outperform Piledriver/Steamroller by a large margin and be close to Haswell IPC.

So by 2016 AMD plans to have Haswell level performance in their CPUs when Intel will be moving to Sky Lake which is uArch change and as well supposed to be a massive iGPU boost. As well it will have all kinds of fun features such as support for Thunderbolt v3 (up to 40Gbps), DDR4, SATA Express and PCIe 4.0. Now this is all rumored but I would not be surprised.

AMD is not making the smart choice in my eyes as Intel will easily keep ahead of them. AMD is making the choice that allows them to make money faster instead. Smart in some ways but not in others. They are not going to grab as much market share as they want.

Again, one other issue with the move or want to move to ARM is software. Software will need to be rewritten and that costs money. Now why hasn't Linux taken over in this recent economic down turn when it is free vs the super expensive Windows? Windows works with pretty much everything you can buy off the shelf. The rest needs to be rewritten for Linux. Same applies for ARM.

Until the major software devs decide to rewrite their software for ARM, it will never take off on desktops or servers.

And no, I do not count Android as that is a horrible OS.

http://www.computingcompendium.com/p/arm-vs-intel-benchmarks.html

So which one is correct? the fake forum post that doesn't even try to describe how they came up with their numbers or this one showing the a7 being considerably slower IPC than haswell.

To anyone who praises Android, explain why toms forums doesn't work properly with any browser, or any mobile website that refuses to redirect to DT for that matter. Open a program, do another task, then swap back and see if the program is still running or if it restarted.

Heck, do a simple task, print a picture.

Android is good for what it does properly, but sucks for what it doesn't do at all. Its not a full OS.

 

[gamerk316]

http://www.computingcompendium.com/p/arm-vs-intel-benchmarks.html

To anyone who praises Android, explain why toms forums doesn't work properly with any browser, or any mobile website that refuses to redirect to DT for that matter. Open a program, do another task, then swap back and see if the program is still running or if it restarted.

Heck, do a simple task, print a picture.

Android is good for what it does properly, but sucks for what it doesn't do at all. Its not a full OS.

Displaying webpages is not a function of the OS; that's an application level job. So blame the applications, not the kernel.

As for multiple apps, given the form factor, that isn't exactly something most users are going to do. Though the interfaces could do with a makeover in that regard...