AMD Piledriver rumours ... and expert conjecture

[Reynod]

We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...

 

[Blandge]

You might be assuming that will lead directly to 2.5x performance, which isn't very accurate to assume. Using that reasoning, A Radeon 7970 would perform >2x as well as a Radeon 7850. It has twice the shader cores, and a higher clock rate (thus the >). Add extra memory everything, and you still don't even come close to 2x performance.

I agree.

Intel also seems to be relying heavily on gpu clock rates, which will not turn out well for them in the long run.

Unlike in the early 2000s where Intel prioritized frequency over architecture, I think they are using their process advantage to play catchup in their GPU architecture because they are so far behind. Just my 2 cents. On CPU it was a choice, in GPU it is a necessity.

 

[cgner]

5.5GHz chip...by IBM:

http://www.computerworld.com/s/article/9230661/IBM_goes_big_on_new_mainframe_with_fastest_chip_flash_memory

"fastest cpu ever" eh? i guess we are still on the hertz war :lol:

 

[Blandge]

Wow, So why can't Amd's design meant for high clocks do this?

If you thought Intel dwarfed AMD's revenue 😵.

 

[GrandAndreasK]

5.5GHz chip...by IBM:

http://www.computerworld.com/s/article/9230661/IBM_goes_big_on_new_mainframe_with_fastest_chip_flash_memory

lol... Riiiiiiiiiiiiiiiiiight......how are someone suposed to afford that.....


They are made for 1$ million servers


http://www.gadgetlite.com/2010/09/06/ibm-5-2-ghz-z196-worlds-largest/

http://www.tomshardware.com/news/z196-mainframe-microprocessor,11168.html

 

[sarinaide]

So OBR laid to rest with AMD already talking about the SteamRoller microarchitecture.....I would very much like to see this "TOP SECRET : SUPER DUPER SECRET DOCUMENT" he is refering too.

People should start to learn how to ignore the biggest internet troll out there.

 

[noob2222]

from the looks of whats going on with AMD's recent announcements about not competing at high end and looking at what BD stuntshow was about, Read is putting a dead stop to that crap, not comparing cpus to Intel's 980x gimmick that brought the napalm to the fight.

TBH thats a good thing, promote under-performing expectations then over-deliver.

BD was stupid promote over-performing expectations then under-deliver. Again, thats why marketing got FIRED.

 

[sarinaide]

Question is how far can they go? They keep shrinking their die size but its getting harder and harder look at how the 22nm turned out. I heard from some where that Intel is being a little light on their current designs(their size is small when compared to Amd) and if they wanted to they probably could of made some beast Intel 5000HD graphics.

Is their any way to shrink after 10-14nm?

Its not a case of "if they wanted" its a case of they cannot, AMD have the leverage of the ATI aquisition which made fusion possible, making X86-64 processors and making graphics cores are not the same thing and it is unlikely that HD 5000 will beat Llano let alone what AMD is offering by Steam Roller. It is also folly to assume incremental rates of improvement, Intel may just have the perfect iGPU but with the lack of support in the form of drivers it will continue to be a fail.

from the looks of whats going on with AMD's recent announcements about not competing at high end and looking at what BD stuntshow was about, Read is putting a dead stop to that crap, not comparing cpus to Intel's 980x gimmick that brought the napalm to the fight.

TBH thats a good thing, promote under-performing expectations then over-deliver.

BD was stupid promote over-performing expectations then under-deliver. Again, thats why marketing got FIRED.

There is no market for the $550 plus chip anymore other than rich folk looking to blow cash, that is not AMD's target market and is certainly not condusive to cloud computing which AMD is after.

 

[de5_Roy]

recent amd related links compiled
setamroller
http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture
http://www.tomshardware.com/news/AMD-Steamroller-Piledriver-Kaveri-processors,17217.html

AMD Explains Advantages of High Density (Thin) Libraries
http://www.tomshardware.com/news/Steamroller-High_Density_Libraries-hot-chips-cpu-gpu,17218.html

amd future 'vision'
http://www.tomshardware.com/news/amd-hot-chips-Surround-Computing-steamroller,17221.html

AMD Hires Ex-Intel Labs Architect, John Gustafson, As Chief Graphics Product Architecture
http://www.anandtech.com/show/6202/amd-hires-exintel-labs-architect-john-gustafson-as-chief-graphics-product-architecture

bobcat 2
http://semiaccurate.com/2012/08/28/amd-let-the-new-cat-out-of-the-bag-with-the-jaguar-core/
amd's 40nm hondo is almost ready
http://www.fudzilla.com/home/item/28540-amd%E2%80%99s-40nm-hondo-almost-ready
piledriver rehash...again..
http://www.cpu-world.com/news_2012/2012082801_Some_details_of_AMD_Piledriver_CPUs.html
i just noticed that that list in the article has the k series unlocked apus missing. so.. was that obr guy right? will the unlocked dt trinities ship with new steppings after bug fix? 😉

 

[sarinaide]

One common theme is it seems like they are switching to TSMC for the fab process which does tend to further the suggestion that higher end GCN GPU integration is in process, that would change a great many things.

 

[sarinaide]

recent amd related links compiled
setamroller
http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture
http://www.tomshardware.com/news/AMD-Steamroller-Piledriver-Kaveri-processors,17217.html

AMD Explains Advantages of High Density (Thin) Libraries
http://www.tomshardware.com/news/Steamroller-High_Density_Libraries-hot-chips-cpu-gpu,17218.html

amd future 'vision'
http://www.tomshardware.com/news/amd-hot-chips-Surround-Computing-steamroller,17221.html

AMD Hires Ex-Intel Labs Architect, John Gustafson, As Chief Graphics Product Architecture
http://www.anandtech.com/show/6202/amd-hires-exintel-labs-architect-john-gustafson-as-chief-graphics-product-architecture

bobcat 2
http://semiaccurate.com/2012/08/28/amd-let-the-new-cat-out-of-the-bag-with-the-jaguar-core/
amd's 40nm hondo is almost ready
http://www.fudzilla.com/home/item/28540-amd%E2%80%99s-40nm-hondo-almost-ready
piledriver rehash...again..
http://www.cpu-world.com/news_2012/2012082801_Some_details_of_AMD_Piledriver_CPUs.html
i just noticed that that list in the article has the k series unlocked apus missing. so.. was that obr guy right? will the unlocked dt trinities ship with new steppings after bug fix? 😉

+1 Good reads, the firing and hiring is a common process when underperformers need to be shown the door and introduce new ideas, many criticized the move but honestly AMD needed to flush the stench of Hector Ruiz and his cash cow cronies.

 

[gamerk316]

You might be assuming that will lead directly to 2.5x performance, which isn't very accurate to assume. Using that reasoning, A Radeon 7970 would perform >2x as well as a Radeon 7850. It has twice the shader cores, and a higher clock rate (thus the >). Add extra memory everything, and you still don't even come close to 2x performance.

Intel also seems to be relying heavily on gpu clock rates, which will not turn out well for them in the long run.

The issue is one of bandwidth. Because memory access isn't getting significantly faster compared to the speed of the CPU [nevermind that latencies keep going up each generation...], there aren't many ways to increase memory bandwidth. That necessitates higher clock rates to improve performance.

 

[fazers_on_stun]

Its not a case of "if they wanted" its a case of they cannot, AMD have the leverage of the ATI aquisition which made fusion possible, making X86-64 processors and making graphics cores are not the same thing and it is unlikely that HD 5000 will beat Llano let alone what AMD is offering by Steam Roller. It is also folly to assume incremental rates of improvement, Intel may just have the perfect iGPU but with the lack of support in the form of drivers it will continue to be a fail.

Hmm, I haven't had any issues with my HD4K drivers, with a couple updates from Intel by now since May. What lack of driver support are you referring to?

 

[sarinaide]

I hardly believe you are running HD4000 when you have a HD7970, they have drivers but I really doubt it turns HD4000 into a highly playable eyecandy graphics core, its barely playable at modest res and lowest presets. I still don't see HD4K or HD5K being the stuff dreams are made of, and probably the very last reason anyone will buy a Intel setup.

 

[fazers_on_stun]

I hardly believe you are running HD4000 when you have a HD7970, they have drivers but I really doubt it turns HD4000 into a highly playable eyecandy graphics core, its barely playable at modest res and lowest presets. I still don't see HD4K or HD5K being the stuff dreams are made of, and probably the very last reason anyone will buy a Intel setup.

Actually I use both simultaneously with Virtu MVP.

I have run some games on the HD4K by itself (connected directly to a 50" plasma HDTV), such as Dragon Age and Torchlight, and yes no way it compares with the 7970. But the point I was making is that I haven't encountered any driver issues. Intel appears to be taking graphics support seriously nowadays.

 

[fazers_on_stun]

The issue is one of bandwidth. Because memory access isn't getting significantly faster compared to the speed of the CPU [nevermind that latencies keep going up each generation...], there aren't many ways to increase memory bandwidth. That necessitates higher clock rates to improve performance.

If S/A and AT are correct about Haswell using TSV interposers for L4 cache for the GPU, then supposedly the bandwidth would go way up and latency way down. I don't see AMD (or GF or TSMC) being able to do this for a number of years, despite their saying otherwise..

Anyway, seems Steamroller and maybe Excavator will be on 28nm strained silicon, so Intel might be on 14nm Bridgewell - half the node size and quadruple the transistor budget - by the time Steamroller rolls out.

 

[sarinaide]

Wisecracker states Excavator will be on 22nm process.

 

[gamerk316]

If S/A and AT are correct about Haswell using TSV interposers for L4 cache for the GPU, then supposedly the bandwidth would go way up and latency way down.

In theory. Depends on the size of the L4.

Also, am I the only one worried about cache failing simmilar to RAM as we continue to increase the size of the caches?

 

[truegenius]

^ cache limit (32/64 bit) or bad prefetching

 

[fazers_on_stun]

Wisecracker states Excavator will be on 22nm process.

Read AT's article in the meantime and it seems Steamy will be on TSMC's 28nm strained silicon, in Q4 of next year. So Exxy would be in 2014, I'm guessing 20nm, again on TSMC's process. Just wonder if TSMC (and/or GF) will have problems and delays like they did with 40nm and 28nm.

 

[fazers_on_stun]

In theory. Depends on the size of the L4.

Also, am I the only one worried about cache failing simmilar to RAM as we continue to increase the size of the caches?

Think cache is static RAM, as opposed to dynamic RAM as used in DIMMs, plus there's probably a certain amount of redundancy built-in to bypass any bad cells or column/row drivers or other circuitry.

 

[cgner]

recent amd related links compiled
setamroller
http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture
http://www.tomshardware.com/news/AMD-Steamroller-Piledriver-Kaveri-processors,17217.html

AMD Explains Advantages of High Density (Thin) Libraries
http://www.tomshardware.com/news/Steamroller-High_Density_Libraries-hot-chips-cpu-gpu,17218.html

amd future 'vision'
http://www.tomshardware.com/news/amd-hot-chips-Surround-Computing-steamroller,17221.html

AMD Hires Ex-Intel Labs Architect, John Gustafson, As Chief Graphics Product Architecture
http://www.anandtech.com/show/6202/amd-hires-exintel-labs-architect-john-gustafson-as-chief-graphics-product-architecture

bobcat 2
http://semiaccurate.com/2012/08/28/amd-let-the-new-cat-out-of-the-bag-with-the-jaguar-core/
amd's 40nm hondo is almost ready
http://www.fudzilla.com/home/item/28540-amd%E2%80%99s-40nm-hondo-almost-ready
piledriver rehash...again..
http://www.cpu-world.com/news_2012/2012082801_Some_details_of_AMD_Piledriver_CPUs.html
i just noticed that that list in the article has the k series unlocked apus missing. so.. was that obr guy right? will the unlocked dt trinities ship with new steppings after bug fix? 😉

Thanks for good links mang.

It looks like steamroller might actually beat Phenom II performance by a thin margin. 5 years after P2 was developed holy fail.

Also they needed a clean up in that company. good to hear they swept them loserz outs.

 

[davemaster84]

Thanks for good links mang.

It looks like steamroller might actually beat Phenom II performance by a thin margin. 5 years after P2 was developed holy fail.

Also they needed a clean up in that company. good to hear they swept them loserz outs.

I think you're really confused, first steamroller is scheduled for October 2013, the upcoming architecture is called piledriver, and second PD is expected to be around 15% better than bulldozer, and since bulldozer as in it's worst appearances just the same, or 2% below phenom II we can say PD is going to be at least 10% over Phenom, which is not bad if you consider that Ivy bridge is only 5% to 8% better than sandy bridge.

 

[mayankleoboy1]

^ except that Intel got that 5%-8% from a die shrink and ~100 mhz speed increase only. Not an arch change.

 

[cgner]

I think you're really confused, first steamroller is scheduled for October 2013, the upcoming architecture is called piledriver, and second PD is expected to be around 15% better than bulldozer, and since bulldozer as in it's worst appearances just the same, or 2% below phenom II we can say PD is going to be at least 10% over Phenom, which is not bad if you consider that Ivy bridge is only 5% to 8% better than sandy bridge.

I am not confused, but maybe my wording is a little off. 15% improvement in SR after 5 years of development. lame yo

 

[davemaster84]

I am not confused, but maybe my wording is a little off. 15% improvement in SR after 5 years of development. lame yo

Yeah perhaps AMD isn't doing things right or whatever, it is clearly that FX had a great disadvantage over SB. But isn't good enough that piledriver is here to correct what FX failed to do so? Besides if it wasn't for AMD it would be left alone with Intel's high prices. Just figure this, with 360 dollars you can buy an 8150 and one very well suited 990fx board, while on Intel you'll barely get enough for a 3770k and even not taking into account that a z79 mobo is around 250. AMD = Best bang for the buck