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 ...

 

[jdwii]

Looks like PD is already doomed. This one benchmark on this one website showed that a dual module cpu is slower than a quad core cpu in this one particular benchmark and the dual module cpu draws just as much power as the quad module big brother when running Their "burn in test"

isn't it funny that the 8150 and the 4170 have the same power draw?

http://static.techspot.com/articles-info/452/bench/Power.png

198W to 252W ... or 212 to 214... So which one is wrong, they can't both be right...

Unless your not being totally cereal i'm pretty sure ALL the Power consumption benchmarks show Bulldozer to consume WAY to much Power. Not to mention the idle on the 4170 is less and the max power consumption? The basics are the 955 and the 4170 are basically the same CPU in terms of overall performance but the 4170 uses more power consumption.

Plus this means nothing for Piledriver since Clock mesh technology and any other improvements Amd and Global foundries did improved the power consumption we hope anyways.

The whole reason Bulldozer was created was for servers and Performance/ per watt and if you ask me Bulldozer is a mediocre server CPU and a bad performance/ per watt CPU.

depends on the benchmark. In winrar,

Everyone knows Bulldozer is perfect for Winrar This is most likely because of a certain instruction set that works so well on the BD.

Can't find where I read it on this post, someone said that what if all of AMD's chips were marketed at stock speeds of 4.0GHZ+, and that it would be a good marketing strategy. I agree to an extent, but the secret is out on what BD can and cannot do, and if PD is only 10-15% better overall, 4.0GHZ still won't measure up to the next best intel chip running on less power and lower clock speeds. They need to redesign everything from the ground up, I know that it would cost the company millions upon millions, but in order for them to compete in intels market, they need to rebuild.

Personally I think AMD is content with the market share they have at this point, they put out a product that does perform well (to people who don't live their lives running benchmarks all day lmao) and they cost somewhat less than intels.

90% of the people who bought Bulldozer was because of Amd's high Core count and high clock rate i read all the reviews on newegg and usually people say things like "3.6Ghz Quad core for 119.99$ sign me up" or "8 cores for less then 200$".

So having all the Piledriver processors at release have a clock rate of 4.0ghz or higher is a fantastic idea. Plus a 10-15% increase in IPC and a 10% increase in clock speed as well as less power consumption is a great start, As long as Amd can price it right this time(Price as always is Number 1 on most peoples minds) The Phenom was slower then the I7 but most people didn't care to much since Amd had way better price/performance and the Phenom II was still better then the first Phenom not worse. Plus Amd left the High-end market and their most likely not coming back. I'm one of the few people who think 10 years from now Amd will only make APU's including on the server.

But by looking at how Amd priced their 7000HD series video cards/Trinity Suggested pricing) i'm going to now make a prediction on price. Also Based on my estimated 20% performance boost and my estimated 5-10% less power consumption when compared to Bulldozer then i compared to Intel's line-up.

129.99$ 4 core PD
179.99$ 6 core PD
239.99$ 8 core PD

Retail prices will most likely be 10-20$ more then this.

Lets see if my estimates are correct at launch

Were i think they should price piledriver to have great Price/Performance.
99.99$ 4 core PD
149.99$ 6 core PD
199.99$ 8 core PD

They need to price their 8 core at 199.99$ so Normal pricing can be around the same price as a 4 core I5 Ivy.

 

[axe1592]

Totally agree with that pricing. If they can get the top of the line Piledriver out for $200 and its within 10-15% of Intel, then we can have a pretty competitive processor market again. Thats something we havent had in a while and will benefit us enthusiast a ton.

PD looks to be 10-15% better IPC based on Trinity. Higher clocks, l3 cache, and any small tweaks made will improve it even more overall.

I think Vishera can come out better than Trinity because Trinity had to worry about energy efficiency, sharing half its die space with an IGPU and not having an L3. Vishera wont have any of these limitations so I think 15% will be the minimum performance jump we'll see over Bulldozer. If it can come in at 20% then it will be close enough to Intel that with some creative pricing like mentioned above, we'll have us a ballgame.

 

[esrever]

Unless your not being totally cereal i'm pretty sure ALL the Power consumption benchmarks show Bulldozer to consume WAY to much Power. Not to mention the idle on the 4170 is less and the max power consumption? The basics are the 955 and the 4170 are basically the same CPU in terms of overall performance but the 4170 uses more power consumption.

Plus this means nothing for Piledriver since Clock mesh technology and any other improvements Amd and Global foundries did improved the power consumption we hope anyways.

The whole reason Bulldozer was created was for servers and Performance/ per watt and if you ask me Bulldozer is a mediocre server CPU and a bad performance/ per watt CPU.

depends on the benchmark. In winrar,

Everyone knows Bulldozer is perfect for Winrar This is most likely because of a certain instruction set that works so well on the BD.

I never knew bulldozer was perfect for winrar. Im not sure where you got that assumption. Its performace in real world applications are generally pretty good. In things like 7 zip and povray it does even better. It fails all the synthetic benchmarks like 3dmark06 and pcmark07 because they were optimized for intel, other benchmarks such as cinebench uses a lot of FPU resource and thus has poor scoring for bulldozer. In general, the performance is nowhere near what you think it is.

Another thing is, bulldozer doesn't consume way more power until you overclock it, its power consumption is within line with the old phenom II line pretty well, the FX4XXX consumes below the x4s and the 6XXX consumes more than the x4s and less than the x6s while the 8XXX consumes slightly above the x6s. The 4170 will use much less power than the 955 due to lower idle power consumption, your assumptions are ill founded.

 

[jdwii]

I never knew bulldozer was perfect for winrar. Im not sure where you got that assumption. Its performace in real world applications are generally pretty good. In things like 7 zip and povray it does even better. It fails all the synthetic benchmarks like 3dmark06 and pcmark07 because they were optimized for intel, other benchmarks such as cinebench uses a lot of FPU resource and thus has poor scoring for bulldozer. In general, the performance is nowhere near what you think it is.

Another thing is, bulldozer doesn't consume way more power until you overclock it, its power consumption is within line with the old phenom II line pretty well, the FX4XXX consumes below the x4s and the 6XXX consumes more than the x4s and less than the x6s while the 8XXX consumes slightly above the x6s. The 4170 will use much less power than the 955 due to lower idle power consumption, your assumptions are ill founded.

You make this sound like a good thing? The whole point of Bulldozer was to make a great Performance/ per watt CPU and they failed really badly with this first try i say fail when they have to sacrifice performance for power consumption they could of just released a Phenom Clock lower and gotten better results or Die shrink the Phenom II. Plus yes Bulldozer is somehow really good at compression such as 7-zip. Also i can care less about Synthetics. Also You get less performance on a Bulldozer 4 core then you get from a smaller 4 core Phenom II. Same goes for the X6 vs 6200 bulldozer. On top of that you get only 2 FPU on a 4 core Bulldozer vs a 4 core Phenom.

Amd can say all they want about how die size doesn't matter but their dead wrong since it matters a lot for them since their the ones losing so much money.

http://www.ocaholic.ch/xoops/html/modules/smartsection/item.php?itemid=773&page=13
Also the 4170 uses 33 watts more on load then a 955 and 5 watts more then a 1090T on idle so where are you getting your numbers from? Yes a 4100 uses less power then a 955 but its only as fast as a Athlon II x3 maybe a x4 again, all they had to do is release a Phenom without L3 cache(wait they did) and it would of lowered the TDP and lowered the cost of production dramatically and just jump to piledriver.

 

[noob2222]

You make this sound like a good thing? The whole point of Bulldozer was to make a great Performance/ per watt CPU and they failed really badly with this first try i say fail when they have to sacrifice performance for power consumption they could of just released a Phenom Clock lower and gotten better results or Die shrink the Phenom II. Plus yes Bulldozer is somehow really good at compression such as 7-zip. Also i can care less about Synthetics. Also You get less performance on a Bulldozer 4 core then you get from a smaller 4 core Phenom II. Same goes for the X6 vs 6200 bulldozer. On top of that you get only 2 FPU on a 4 core Bulldozer vs a 4 core Phenom.

Amd can say all they want about how die size doesn't matter but their dead wrong since it matters a lot for them since their the ones losing so much money.

http://www.ocaholic.ch/xoops/html/modules/smartsection/item.php?itemid=773&page=13
Also the 4170 uses 33 watts more on load then a 955 and 5 watts more then a 1090T on idle so where are you getting your numbers from? Yes a 4100 uses less power then a 955 but its only as fast as a Athlon II x3 maybe a x4 again, all they had to do is release a Phenom without L3 cache(wait they did) and it would of lowered the TDP and lowered the cost of production dramatically and just jump to piledriver.

Like I said, according to whom? sure, you found a site that sais the 4170 draws just as much power as the 8150. I already pointed out before, another website shows the 4170 draws considerably less power than the PII 980 even at full load and wayy wayy less than the 8150.

Wich website is right? I don't see running 2/4 modules pulling just as much power as 4/4 modules, obviously you do since it supports your arguement.

This is why PD will be a failure, everyone will look at one aspect from one website and ignore the rest.

I did find one thing interesting on the ocaholic site. thier cinebench test was timed and not "scored"

Everyone knows PII scores better .... but BD completes the task faster ?! How does that work..

Cinebench R10 1 CPU

AMD FX-4170 4.2 GHz DDR3-1600 209 sec 93.72 %

Intel Core i7 940 2.93 GHz DDR3-1600 207 sec 92.83 %

AMD FX-8150 3.6 GHz DDR3-1600 208 sec 93.27 %

Intel Core i7 920 2.66 GHZ DDR3-1600 223 sec 100.00 %

AMD Phenom II X6 1090T 3.2 GHz DDR3-1333 225 sec 100.89 %

AMD FX-8120 3.1 GHz DDR3-1600 230 sec 103.14 %

AMD Phenom II X4 955 3.2 GHz DDR3-1333 237 sec 106.28 %

 

[jdwii]

Like I said, according to whom? sure, you found a site that sais the 4170 draws just as much power as the 8150. I already pointed out before, another website shows the 4170 draws considerably less power than the PII 980 even at full load and wayy wayy less than the 8150.

Wich website is right? I don't see running 2/4 modules pulling just as much power as 4/4 modules, obviously you do since it supports your arguement.

This is why PD will be a failure, everyone will look at one aspect from one website and ignore the rest.

I did find one thing interesting on the ocaholic site. thier cinebench test was timed and not "scored"

Everyone knows PII scores better .... but BD completes the task faster ?! How does that work..

Cinebench R10 1 CPU

AMD FX-4170 4.2 GHz DDR3-1600 209 sec 93.72 %

Intel Core i7 940 2.93 GHz DDR3-1600 207 sec 92.83 %

AMD FX-8150 3.6 GHz DDR3-1600 208 sec 93.27 %

Intel Core i7 920 2.66 GHZ DDR3-1600 223 sec 100.00 %

AMD Phenom II X6 1090T 3.2 GHz DDR3-1333 225 sec 100.89 %

AMD FX-8120 3.1 GHz DDR3-1600 230 sec 103.14 %

AMD Phenom II X4 955 3.2 GHz DDR3-1333 237 sec 106.28 %

Total BS, That is really all i have to say some times it doesn't matter how much proof some one has some people are in 100% denial. Also show this other site and i'll take a look at it.

The 980 is also faster then the 4170 by a pretty big margin and a 955 is about equal so if Amd is supposed to have some great Performance/per watt(main reason the Bulldozer was created is for cost/power consumption)why would this be down on these newer series and it also takes more die space to produce a 4170 then it does to produce a 955 which means less profit for Amd.

Also I think that is funny that the test completed faster Toms noticed the same thing on some of their testing can't find the link.

This conversation is over though since this is a piledriver forum not to mention all were doing is running in circles and ovisially unless i see proof i'm not changing my mind and with proof you don't, but please if you find the link PM me. i mean that i hope its not a video my slow 3G internet can't load it LOL.

 

[esrever]

Can you please show how the 955 is much faster than the 4170? not just 1 benchmark please.

 

[king smp]

just to show a FX against a Thuban

this is FX-8150 vs 1100T
remember that the 1100T is 300mhz slower with two less cores

http://www.anandtech.com/bench/Product/434?vs=203

I will look for 955 vs 4170 later

 

[de5_Roy]

... did you guys look into each test system and the software used during load power measurement?
edit: since the 4170 is a 125w cpu, it might not use more than 125w at stock settings regardless of test platform. the power consumption figures from both ocaholic and techreport look like total system power consumption measurements.
i don't know how amd measures tdp, if it designates tdp using maximum usage scenario, cpu power consumption would be around 125w at stock.

 

[de5_Roy]

apple was this close to get an amd llano apu...
if amd gets the trinity ulv apus right, apple might become interested again, who knows, next gen apple macs might contain a steamroller apu!! 😀

 

[guskline]

king smp: Since I own an 1100T running well at 4 ghz, I will wait for PD to be released to see what improvements it makes. BD was the only comparison and though it won a few benchies the jump forward was not consistent. Moreover, pluncking down nearly $200 on the 1100T and then seeing the BD results sure dampens the desire to upgrade.

 

[jdwii]

If those benchmarks were false then fine but i stand by the 4170 not being slower but pretty much equal to a 955 while using more power consumption the 4170 might be 10% faster or it might be 15% slower But as for this dragging on so long like it did was not my intention and i wish a mod would delete some of the past posts. But its not like its really doing any harm if someone else would like a review its here unfortunately their is not much comparing. I can't honestly find another review of the 4170 besides from those 2. When it comes to power consumption we already now Bulldozer uses a lot of Power when clock high this will be slightly resolved with clock mesh technology.

http://www.youtube.com/watch?v=99-Nd-MNiTw

 

[jdwii]

The A10 does so well when compared to the I3 i'm really proud of Amd and their Trinity their is now way in hell i would recommend a I3 over a A10. About the same CPU performance and WAYYYYY better graphics performance. Ram speeds are holding Amd and probably even Intel back. I want my DDR4 or 5 on my desktop.


If priced right Trinity will be such a good product i might even build myself a nice A10 system its going to be my media centers best friend.

 

[Apple Debbie Bhatia]

Hi,
where i can buy AM3+ mobo n san antonio?

Apple Debbie Bhatia

 

[king smp]

If those benchmarks were false then fine but i stand by the 4170 not being slower but pretty much equal to a 955 while using more power consumption the 4170 might be 10% faster or it might be 15% slower But as for this dragging on so long like it did was not my intention and i wish a mod would delete some of the past posts. But its not like its really doing any harm if someone else would like a review its here unfortunately their is not much comparing. I can't honestly find another review of the 4170 besides from those 2. When it comes to power consumption we already now Bulldozer uses a lot of Power when clock high this will be slightly resolved with clock mesh technology.

http://www.youtube.com/watch?v=99-Nd-MNiTw

this might be an interesting read

http://www.legionhardware.com/articles_pages/amd_fx_8150fx_8120fx_6100_and_fx_4170,4.html

bunch of different benches in that article and BTW I have a Phenom II system
the issue is that the FX series isnt a great upgrade or new build option IMHO
if somebody has an AM3+ mobo then chances are they have a Phenom II or Thuban in there so cost vs performance the FX doenst make sense to do the upgrade
and for a new build if it is going to be used for multithreaded work then a FX might be worthwhile but the problem is that AM3+ is nearing the EOL
from what I understand the PD will be the last CPU on the AM3+
so recommending AM3+ is tough since upgrade options in the future are limited
true enough that 1155 is being phased out but the upgrade options are better as far as CPU performance goes
also I believe that FM2 is on the way out also
sad to say it is just not a a good time to go with AMD at this point
hopefully this will change with new sockets and CPUs in the future

 

[palladin9479]

BD on release was a bad option due to pricing. The 8120 turned out to be a pretty good option after they did price cuts. If someone had an AM3+ system with an older PII in it, I could see the 8120 being a decent upgrade for them.

PD OTH we don't know about, depending on price and scalability it may end up being a good upgrade option for those of us with AM3+ platforms and higher end OC'd PII's. It may even be a good idea for new builds, depending on price.

Also for those still playing the IPC card, I once said that 1 BD "core" was approximately 75% the processing power of a typical "core". Considering it only has 66% of the ALU's per "core" this actually speaks positively about their design. They focused a bit too much on a wide flat design instead of something to make the bench-markers happy. Lets see how PD pans out.

 

[king smp]

Isnt PD the last CPU for AM3+?

hard to recommend for a new build with no future CPU generation for an upgrade
of course the same can be said for 1155

 

[amdfangirl]

It should be.

 

[king smp]

kind of disappointed in the AM3+ socket
EOLing kind of quick isnt it?
just came out for BD and PD

 

[jimmysmitty]

If those benchmarks were false then fine but i stand by the 4170 not being slower but pretty much equal to a 955 while using more power consumption the 4170 might be 10% faster or it might be 15% slower But as for this dragging on so long like it did was not my intention and i wish a mod would delete some of the past posts. But its not like its really doing any harm if someone else would like a review its here unfortunately their is not much comparing. I can't honestly find another review of the 4170 besides from those 2. When it comes to power consumption we already now Bulldozer uses a lot of Power when clock high this will be slightly resolved with clock mesh technology.

http://www.youtube.com/watch?v=99-Nd-MNiTw

But hasn't it been said that this clock mesh tech also is only effective up to a certain speed? BDs problem was not that it used too much power. At stock it was not that bad for what it is. It was when overclocked it would jump to almost 2x the power us for the system.

The A10 does so well when compared to the I3 i'm really proud of Amd and their Trinity their is now way in hell i would recommend a I3 over a A10. About the same CPU performance and WAYYYYY better graphics performance. Ram speeds are holding Amd and probably even Intel back. I want my DDR4 or 5 on my desktop.


If priced right Trinity will be such a good product i might even build myself a nice A10 system its going to be my media centers best friend.

Memory speed is in no way holding either back. Well not Intel at least as has been shown there are almost no noticeable improvements from DDR3 1333 to DDR3 1866.

What will help is stacked RAM as it will cut latency down to near nothing.

 

[palladin9479]

But hasn't it been said that this clock mesh tech also is only effective up to a certain speed? BDs problem was not that it used too much power. At stock it was not that bad for what it is. It was when overclocked it would jump to almost 2x the power us for the system.

No a previous poster stated that due to the physics involved, the IC's could only effectively resonate within a certain frequency range. The poster was correct about his physics, not correct with how modern CPU's are made to use multiple frequencies. The tank circuits will resonate on the bclk, thus will limit overclocking via the bus and not the multiplier. This is also the reason RCM technology can only reduce power to CPU's by 10~20% (I'm expecting ~15%), whereas on an IC with a single frequency it can reduce power usage by well over 30%. There is no upper limit to the frequencies RCM can be used with, effectiveness goes down with complexity not frequency.

 

[EXT64]

The depressing part for me at least is that although PD seems to be making some good gains over BD, it is still severely hampered by its shared FPUs in float heavy computations. It (BD) is barely competitive in float with the X6 and 1st gen i7, and when you compare it with SB, Ivy, or Gulftown it is blown away. I think AMD really jumped the gun with their slashing of float resources. Yes, GPUs do offer amazing float performance, but they are currently severely hampered by limited ram and slow PCI-E and thus are only optimal for certain problems. And while software is being written to take advantage of GPUs and coprocessors, it will be a long time before it is really prevalent, at which point BD and PD will be long forgotten. APUs have an interesting future, but they are even more 'bleeding edge' than discrete GPU processing and thus are even farther out from full acceptance. Not to mention, I can't get a 4 module BD or PD with a graphics core for float, so why slash the float units in the CPU?

 

[palladin9479]

Memory speed is in no way holding either back. Well not Intel at least as has been shown there are almost no noticeable improvements from DDR3 1333 to DDR3 1866.

What will help is stacked RAM as it will cut latency down to near nothing.

Neither latency nor bandwidth have much impact on consumer computing with respect to system memory. GPU's are a different story as their immense parallel computing requires constantly feeding massive amounts of data. I believe that was what the poster was referring to, the more powerful the iGPU the more memory bandwidth / latency it needs.

Stacked RAM will be amazing as a cache or a graphics memory, though I suspect it might not be large enough with how GPU's are now needing 1GB+. I'm suspecting Intel / AMD will eventually create a form of "L2 graphics cache" for iGPU's. For costing reasons iGPU's would still use system memory (when this happens it'll be 2GB or more as standard) for data storage, with a high speed low latency 512MB+ secondary cache that would mask the slower speed / latency of the main system memory. GPU's already have instruction / data caches built into them now, this just makes sense as the natural evolutionary step to take.

 

[palladin9479]

BD's are hampered by FPU performance due to their shared front end and L2 cache system. A single BD module does have two 128-bit FPU's with each one assigned to a different "core". They can be used to process a single 256-bit instruction or two separate 128-bit instructions. Due to the front end scheduler arbitration rarely will both FPU's be used at the same time unless you deliberately separate the FPU calculations into different threads, and then you run into the caching situation.

Also make 100% sure that your not using something that's been linked to Intel's math library. Non-Intel CPU's might as well not even have modern SIMD FPU's then.

 

[EXT64]

Well, to be honest, I was more depressed by my slower clocked X6 beating it (so intel optimizations shouldn't hurt there). Have you heard of any improvements to PD that might help the FPUs work more efficiently (so, beyond just clock speed increases)?