Does AMD has some future?

[jaguarskx]

Supposedly Skylake will have double digit IPC improvements up to 10% and possibly 40% or higher iGPU performance...

With regards to IGP performance, the range is reasonable since after the release of the Intel HD 3000 graphics core, newer generation IGPs have been on average around 30% - 35% more powerful than the previous generation; excluding the Iris Pro series.

The Iris Pro 5200 graphics core in Haswell generation Core i5 / i7 CPUs is more or less equivalent to the Radeon R7 integrated graphics core found in the A10-7xxx series APU. However, Intel CPUs with the Iris Pro 5200 graphics core are rather expensive.

 

[-Fran-]

And another big cheese shuffle...

Cheers!

EDIT: http://www.anandtech.com/show/8874/amd-executive-shakeup-byrne-laforce-and-naik-leave-amd

 

[anxiousinfusion]

And another big cheese shuffle...

Cheers!

EDIT: http://www.anandtech.com/show/8874/amd-executive-shakeup-byrne-laforce-and-naik-leave-amd

Again? Good lord. AMD is like the corporate equivalent to riding a motorcycle held together by duct tape and hairpins.

 

[Gam3r01]

At least they use duct tape that cost more than a dollar...

 

[juanrga]

Kaveri refresh will be a big deal if it can pull off a 20% gain at the same 95watt tdp.

How? Richland achieved about 10% higher clocks thanks to mature process and RCM improvements. Kaveri refresh only could benefit from a mature process this time. Moreover, current A10-7850k iGPU is bandwidth bottlenecked. Higher clocks will only emphasize the bottleneck.

 

[juanrga]

I have my doubts about ZEN, some claim it will be superior to Skylake... which seriously sounds like Smelly BS, Intel has already stated that Skylake will be the biggest jump in terms of performance since the last 10 years, since AMD is around 60% behind in terms of IPC i seriously don`t see Zen beating Skylake, it may however be competitive (which could either be close or worst to performance from Haswell).

AMD did the idiotic mistake of sticking with Bulldozer/Piledriver for this long, they should had dumped it as soon as FX-8150 was released. AMD can improve their archs as much as they want, but as long as they don`t fix the horrible cache systems in their Chips they will always be screwed.

The claim Zen will be faster than Skylake is pure BS. But recall whom did and which is his record of failed 'predictions'.

Leaked benchmarks show Skylake has about 20% higher IPC than Haswell which is about 60% higher than Piledriver for general scalar workloads. For SIMD/FP the gap is very superior

Skylake: 64 FLOP / core
Piledriver: 8 FLOP / core (16 FLOP / module)

AMD engineers knew that Bulldozer was going to fail. The project would have been canceled at early phase. However, management decided to continue, with the results that everyone know.

 

[jdwii]

Anyone with the exact quote from Intel?

If it is one of the "the biggest *efficiency* jump in the last 10 years" is totally not the same as "the biggest *performance* jump in the last 10 years".

Cheers!

http://www.pcworld.com/article/2683392/pc-confusion-to-linger-on-intels-quick-jump-to-skylake.html

"Skylake chips will have the biggest PC innovations in the last 10 years, Intel believes. Skylake will bring wireless charging and data transfers, and also a “significant increase in performance, battery life and power efficiency,” said Kirk Skaugen, general manager of Intel’s PC Client Group, during last week’s Intel Developer Forum."

Although they do exaggerate a lot, Intel really has no reasons to do so for the PC since they have no competition anymore. Not expecting to see anything like 20+% gains in CPU tasks on average but i expect to see at best a nehalem-sandybridge jump and at worse a Sandybridge to ivy bridge jump. Keep in mind we are talking about 14nm and a boost in the architecture itself.

Intel really cares about their igpu as well so i expect to see another 25-35% gain there as well just like the 3000 to 4000 and the 4000 to the 4400.

With Zen Amd is once again in quiet mode and with recent sales and recent money being spent on research i fully expect them to actually fall further behind Intel in many areas regrading performance.

Edit for more info
http://www.pcworld.com/article/2683392/pc-confusion-to-linger-on-intels-quick-jump-to-skylake.html
"“Intel learned a lesson in the Pentium 4 days, that you’re not far ahead of your competition,” McCarron said."

 

[juanrga]

And another big cheese shuffle...

Cheers!

EDIT: http://www.anandtech.com/show/8874/amd-executive-shakeup-byrne-laforce-and-naik-leave-amd

As I said in the broken thread AMD was actually split into two groups with very different cultures and strategies. After Rory abandoned the ship, there was a internal fight between both to get the control of the company and everyone knows who won. The loser was, according to my sources, renegotiating with the BoD his role on the post-Rory AMD and all seems to indicate that had no agreement.

 

[Sithex]

wot

 

[jaguarskx]

AMD engineers knew that Bulldozer was going to fail. The project would have been canceled at early phase. However, management decided to continue, with the results that everyone know.

I do not analyze AMD as a company, nor do I analyze their financial statements.

I would say that if AMD engineers knew Bulldozer was going to fail, they probably had no choice but to proceed since diminishing revenues mean less money to devote to R&D. Scrapping Bulldozer in favor of a different CPU architecture would likely have been very costly to AMD if they have done so. That means they would have to spend money and time developing an alternative which in turns means AMD would be reliant on the Phenom II CPUs for a longer period of time. Granted, Bulldozer was more of a "side-grade" than an upgrade to the Phenom II. But, because AMD persevered Piledriver was eventually released and it is an improvement over the Phenom II.

If AMD were to abandon development for Bulldozer and start with a clean slate, that would have meant that all the money they have spent on R&D would have to be immediately expensed instead of capitalized. Capitalizing the R&D expenses means that the total R&D expenses would have been slowly recognized as current expense over the course of 5 years or whatever timetable AMD decided. That would allow the R&D expenses to be netted against revenue over time so that if the product sold well, AMD would have been able to show a profit quarter over quarter.

As stated, abandoning Bulldozer would have meant that AMD would need to immediately expense all R&D expenditures related to Bulldozer. Not doing so would be a severe violation of SEC regulatory rules and US accounting standards. Therefore, whatever amount of money invested into R&D whether it be $500k or $1 billion, the expense would be immediately disclosed and AMD's stock price would have likely taken a severe hit in a shareholder sell off. Depending on the severity of the sell off, it is possible that the stock price could have gone so low that it jeopardizes AMD credit rating which means it makes it both more difficult and expensive to borrow money from financial institutions (which is a very common practice in business) for AMD's daily operations. That leads to even more trouble...

 

[jdwii]

Kaveri refresh will be a big deal if it can pull off a 20% gain at the same 95watt tdp.

How? Richland achieved about 10% higher clocks thanks to mature process and RCM improvements. Kaveri refresh only could benefit from a mature process this time. Moreover, current A10-7850k iGPU is bandwidth bottlenecked. Higher clocks will only emphasize the bottleneck.

juans logic- amd is dead, no way amd can pull a 20% gain

the reality a 20 % gain is possible with a mature process, that an a clock speed bump.
http://cdn.eteknix.com/wp-content/uploads/2014/03/Kaveri_overclocking_1.png
every test i have seen shows kaveri stops scaling around 2400/2600ddr3 due to the M/C
http://www.hardcoreware.net/kaveri-memory-speed-performance-scaling/
but the core clock has a lot to yield, 720 base can be pushed higher.
a move to gcn 1.2 tonga has not been ruled out , another possibility

Actually if they can pull a 20% increase(on average) in 1 generation with just a mature process i'll be willing to buy you that exact model. You understand how much 20% is right? However if Amd updated cache/branch predictor-shrunk steamroller to 22nm i'd be willing to bet they could pull it off with excavator.

When Juan says 20% IPC gain i must ask in what instruction set? To see those level of gains is extraordinary if on average.

 

[etayorius]

I don`t expect more than 10% compared to Haswell, but it is quite good none the less considering how far ahead is Intel compared to AMD, that would put AMD at least 70% behind Intel, if Intel manages up to 20% that would be HUGE... i really don`t like Intel and i rather give my money to AMD, but i have no use for AMD products in regards of upgrade... sadly i am forced to move to Intel in regards to CPU, i will stick to Radeons GPU though.

I am still holding to a PhenomII 965BE OCed at 4GHZ, a Skylake CPU would be almost double the speed compared to my current one.

News about more execs leaving makes me believe that AMD is about to sink... i hope i am wrong.

 

[juanrga]

When Juan says 20% IPC gain i must ask in what instruction set? To see those level of gains is extraordinary if on average.

the near 20% gain is for x86 workloads

http://wccftech.com/intel-skylake-benchmarks-leaked-sisoftware-sandra/

For AVX kind of workloads the max. IPC gain is 2x because Skylake brings new 512bit SIMD units (Haswell uses 256bit units).

 

[juanrga]

AMD engineers knew that Bulldozer was going to fail. The project would have been canceled at early phase. However, management decided to continue, with the results that everyone know.

I do not analyze AMD as a company, nor do I analyze their financial statements.

I would say that if AMD engineers knew Bulldozer was going to fail, they probably had no choice but to proceed since diminishing revenues mean less money to devote to R&D. Scrapping Bulldozer in favor of a different CPU architecture would likely have been very costly to AMD if they have done so. That means they would have to spend money and time developing an alternative which in turns means AMD would be reliant on the Phenom II CPUs for a longer period of time. Granted, Bulldozer was more of a "side-grade" than an upgrade to the Phenom II. But, because AMD persevered Piledriver was eventually released and it is an improvement over the Phenom II.

If AMD were to abandon development for Bulldozer and start with a clean slate, that would have meant that all the money they have spent on R&D would have to be immediately expensed instead of capitalized. Capitalizing the R&D expenses means that the total R&D expenses would have been slowly recognized as current expense over the course of 5 years or whatever timetable AMD decided. That would allow the R&D expenses to be netted against revenue over time so that if the product sold well, AMD would have been able to show a profit quarter over quarter.

As stated, abandoning Bulldozer would have meant that AMD would need to immediately expense all R&D expenditures related to Bulldozer. Not doing so would be a severe violation of SEC regulatory rules and US accounting standards. Therefore, whatever amount of money invested into R&D whether it be $500k or $1 billion, the expense would be immediately disclosed and AMD's stock price would have likely taken a severe hit in a shareholder sell off. Depending on the severity of the sell off, it is possible that the stock price could have gone so low that it jeopardizes AMD credit rating which means it makes it both more difficult and expensive to borrow money from financial institutions (which is a very common practice in business) for AMD's daily operations. That leads to even more trouble...

When I wrote "early phase" I did mean drawing board. They could have canceled Bulldozer before start to design the transistors, save time and money, and use that for an update of their previous arch. Yes, the previous arch was near its limits, but could have given a nice 10% gain, whereas an entirely new architecture was planned.

 

[juanrga]

AMD confirms Carrizo is not coming to FM2+ platform:

With regards to your specific question, we expect Carrizo will be seen in BGA form factor desktops designs from our OEM partners. The Carrizo project was focused on thermally constrained form factors, which is where you'll see the big differences in performance and other experiences that consumers value.

 

[gamerk316]

Dang it, beaten by Juan:

http://www.anandtech.com/show/8871/amds-carrizo-not-on-the-desktop-depends-what-you-define-as-desktop

Carrizo = BGA Only.

 

[etayorius]

Juan was Right... again.

 

[jdwii]

http://www.extremetech.com/gaming/197450-amds-r9-380x-finalized-could-include-new-memory-interface-and-dramatic-performance-improvements

Wow 300 Watts that's twice the power consumption as the 980.

 

[8350rocks]

I am posting here because the AMD thread gives continuous server error 500.

Before Rory departure, my sources said me that K12/Zen was going to be made on 14FF and Keller was happy with the new node. However, AMD's SVP & CFO has announced that AMD will finally use 28nm for K12

http://seekingalpha.com/article/2739835-advanced-micro-devices-amd-2014-raymond-james-systems-semiconductors-software-and-supply-chain-conference

http://wsw.com/webcast/rj94/amd/

And if K12 is 28nm then I assume that Zen will be as well.

Adds that the AMD tablet project was silently canceled a week ago and that Carrizo D/E is in the air, and the conclusion is that AMD doesn't have a winning strategy for the future.

AMD doesn't have plans/projects for phones or tablets. The new 28nm K12/Zen will be uncompetitive on laptops and servers against 14/16nm parts. AMD has currently abandoned HEDT to Intel, and there is no short term plans for supercomputers. The promised new semicustom wins are never announced. What market remain to AMD? The desktop GPU division, which even AMD admits is not in good shape? The semicustom division, which will only provide about 3 billions on revenue on next years? Is this the end of AMD?

This is all rubbish and speculation.

Reality is juanrga has no knowledge of the situation, and has been proven to be of no technical expertise about anything.

He is an intel fan boy, and ARM nut...he thinks that Qualcomm will overtake Intel in the desktop PC and laptop market soon, and that somehow, ARM will win over x86 and everything that entails.

He even got into an argument with Linus Torvalds on a Linux Forum where Linus told him he was a complete idiot and not worth his time.

http://www.realworldtech.com/forum/?threadid=146066&curpostid=146770

 

[8350rocks]

I have my doubts about ZEN, some claim it will be superior to Skylake... which seriously sounds like Smelly BS, Intel has already stated that Skylake will be the biggest jump in terms of performance since the last 10 years, since AMD is around 60% behind in terms of IPC i seriously don`t see Zen beating Skylake, it may however be competitive (which could either be close or worst to performance from Haswell).

AMD did the idiotic mistake of sticking with Bulldozer/Piledriver for this long, they should had dumped it as soon as FX-8150 was released. AMD can improve their archs as much as they want, but as long as they don`t fix the horrible cache systems in their Chips they will always be screwed.

The claim Zen will be faster than Skylake is pure BS. But recall whom did and which is his record of failed 'predictions'.

Leaked benchmarks show Skylake has about 20% higher IPC than Haswell which is about 60% higher than Piledriver for general scalar workloads. For SIMD/FP the gap is very superior

Skylake: 64 FLOP / core
Piledriver: 8 FLOP / core (16 FLOP / module)

AMD engineers knew that Bulldozer was going to fail. The project would have been canceled at early phase. However, management decided to continue, with the results that everyone know.

Oh really...?

I have been wrong with predictions? When have I predicted anything? I speculate as I would...here and there.

However, as someone under NDA, predictions are not wise for such a man.

As someone who is not under NDA, juan, why not tell them how your predictions for Kaveri failed almost across the board?

 

[juanrga]

http://www.extremetech.com/gaming/197450-amds-r9-380x-finalized-could-include-new-memory-interface-and-dramatic-performance-improvements

Wow 300 Watts that's twice the power consumption as the 980.

Brute force approach.

 

[8350rocks]

I really don't have details, etayorius, but I'm pretty sure changing a design that fails (Zambezi) is not an economically feasible option. Specially when money is not a resource your company has in abundance. They just had to make do with the bad calls from previous management and all that.

In particular to Zen, I haven't seen any documents around it's design, so I don't know what it will bring and change. Even more, wasn't Zen for an ARM design?

Cheers!

PS: Cazalan, the "official" thread is dead. Do you know if we can just get a new one and close the old one?

*sigh* we killed it... 🙁

 

[jdwii]

I am posting here because the AMD thread gives continuous server error 500.

Before Rory departure, my sources said me that K12/Zen was going to be made on 14FF and Keller was happy with the new node. However, AMD's SVP & CFO has announced that AMD will finally use 28nm for K12

http://seekingalpha.com/article/2739835-advanced-micro-devices-amd-2014-raymond-james-systems-semiconductors-software-and-supply-chain-conference

http://wsw.com/webcast/rj94/amd/

And if K12 is 28nm then I assume that Zen will be as well.

Adds that the AMD tablet project was silently canceled a week ago and that Carrizo D/E is in the air, and the conclusion is that AMD doesn't have a winning strategy for the future.

AMD doesn't have plans/projects for phones or tablets. The new 28nm K12/Zen will be uncompetitive on laptops and servers against 14/16nm parts. AMD has currently abandoned HEDT to Intel, and there is no short term plans for supercomputers. The promised new semicustom wins are never announced. What market remain to AMD? The desktop GPU division, which even AMD admits is not in good shape? The semicustom division, which will only provide about 3 billions on revenue on next years? Is this the end of AMD?

This is all rubbish and speculation.

Reality is juanrga has no knowledge of the situation, and has been proven to be of no technical expertise about anything.

He is an intel fan boy, and ARM nut...he thinks that Qualcomm will overtake Intel in the desktop PC and laptop market soon, and that somehow, ARM will win over x86 and everything that entails.

He even got into an argument with Linus Torvalds on a Linux Forum where Linus told him he was a complete idiot and not worth his time.

http://www.realworldtech.com/forum/?threadid=146066&curpostid=146770

I actually read all that and for Amd's sake lets hope linus is wrong. It's something many people have been stating over and over again we need high single core performance. I'm guessing anything beyond 8 cores is just silly for normal consumers, and this includes-gamers.

 

[8350rocks]

Dang it, beaten by Juan:

http://www.anandtech.com/show/8871/amds-carrizo-not-on-the-desktop-depends-what-you-define-as-desktop

Carrizo = BGA Only.

I said Carrizo was mobile only and 28nm months ago when it was leaked...back in July/August last year to be precise...

Then Juan comes out with "Oh look Carrizo DT parts...You were wrong!"

I still said Carrizo was mobile only...the DT part is a completely different codename I have not seen mention of anywhere on the entire internets...(not Zen either).

So...no worries.

This does not deviate from anything I was told many moons ago.

EDIT: You know what...nevermind...keep listening to Juan...evidently Juan can forecast the end of everything, and yet it never comes...and people just keep eating the crow stew like it is spicy chicken and dumplings.

 

[etayorius]

http://www.extremetech.com/gaming/197450-amds-r9-380x-finalized-could-include-new-memory-interface-and-dramatic-performance-improvements

Wow 300 Watts that's twice the power consumption as the 980.

Brute force approach.

OMG more idiotic Bulldozer approach... seriously is AMD Retarded?

If industry asks for more IPC AMD adds more Cores, if Industry asks for more efficient GPUs AMD adds more GPU Cores with more power draw...