Are we expecting Piledriver to be a refined 32nm part or the introduction of 22nm?

Apostrophe

n00b
Joined
Nov 15, 2005
Messages
36
Everything I read about Bulldozer seems to indicate that it is far from impressive. I am, however, perhaps more optimistic about the future of the micro-architecture than most.

I believe that if they can knuckle down and reduce the uncore to a more reasonable size and introduce a die shrink to 22nm, the processor will reach around 5Ghz overclocks on a decent watercooling setup.

Question is, are they going to 22nm in Piledriver or in Steamroller?

Another question is, when are they going to move past AM3+? If Piledriver ends up being the final AM3+ processor, I'm just going to skip AM3+ and wait for whatever comes after it.
 
32nm refined part with an estimated 10~15% performance per core and it's going to be an AM3+ part so the platform has some life left in it. It is not expected to get anymore cores as they will be tweaking performance. To be fair my 45nm part is pushing 4.3GHz so a refinment can help.
 
roadmap indicates release next year so 32mm part maybe on a mature process to gain advantage.
 
I highly doubt that Piledriver will be based on 22nm, AMD is slow when it comes to adopting new nano meters for their CPU's, lets hope that Piledriver will be a huge improvement over Bulldozer.
 
I highly doubt that Piledriver will be based on 22nm, AMD is slow when it comes to adopting new nano meters for their CPU's, lets hope that Piledriver will be a huge improvement over Bulldozer.

maybe AMD should switch to Intel's tick-tock strategy :D
 
There will be no 22nm for a long long time. 32nm was just released and it is has not even begun to mature.
 
GloFo's 32nm process are not yet optimized. No way will the Piledriver come out as 22nm.
Given the current performance of Orochi (Bulldozer), I expect Piledriver's schedule to be pulled-in.
 
maybe AMD should switch to Intel's tick-tock strategy :D

Has nothing to do with AMD at this point. AMD is fabless, they rely on other companies to offer that process before AMD can switch. AMD will be stuck on 32nm for awhile seeing how they just moved to it.

Their GPUs on the other hand will be coming out on 28nm shortly.
 
Has nothing to do with AMD at this point. AMD is fabless, they rely on other companies to offer that process before AMD can switch. AMD will be stuck on 32nm for awhile seeing how they just moved to it.

Their GPUs on the other hand will be coming out on 28nm shortly.

I think they GPU's have been delayed till mid next year. 28nm has problems too sigh.
 
I am not too sure that I'm gonna live long enough to see GloFo crank out 22nm CPUs.
 
This is Sooo much like the Phenom I(aka Barcelona) it's not funny. Deneb fixed all the issues with Barcelona, and sounds like Piledriver will do the same thing. Something's I'd like to recomend to AMD:

Raise the Base Hyper-Transport! Your raised it a little for the 8120 and 8150 from 2000Mhz to 2200Mhz, but take it to 3000Mhz or 3200Mhz and you'll definitely see some performance gain.

When it comes to turbo, have the chip disable every other core as not required that way the core that is doing the turbo would have more available cache available. I hope to experiment with that when I get my 8120 back from RMA.

Enable Quad-Channel on Piledriver, I believe the server parts have it and the FX just have it disabled. A lot of enthusiast are OCD (I may be speaking for myself) and like filling those slots, even if it's wholly unecesary (come on, do I need 12GB and 16GB of ram?). So, lets put it to good use!

Recommend that it best be used with 1866Mhz memory.

Together, these May net AMD 15-20% Improvement alone.

Also, when it came to selling the 8120 and 8150, I think to better justify the price jump from the 8120 to the 8150 would have been to ship the 8150 with the Water Cooler. That alone would justify the $80 difference between them, and for people with their own WC or bad ass air-coolers, can get the cheaper 8120.
 
Raise the Base Hyper-Transport! Your raised it a little for the 8120 and 8150 from 2000Mhz to 2200Mhz, but take it to 3000Mhz or 3200Mhz and you'll definitely see some performance gain.

are you talking about non-coherent HT? non-coherent HT is 2.6Ghz while coherent HT (for G34 server) is 3.2Ghz (6.4GT/s).
 
This is Sooo much like the Phenom I(aka Barcelona) it's not funny. Deneb fixed all the issues with Barcelona, and sounds like Piledriver will do the same thing. Something's I'd like to recomend to AMD:

Raise the Base Hyper-Transport! Your raised it a little for the 8120 and 8150 from 2000Mhz to 2200Mhz, but take it to 3000Mhz or 3200Mhz and you'll definitely see some performance gain.

When it comes to turbo, have the chip disable every other core as not required that way the core that is doing the turbo would have more available cache available. I hope to experiment with that when I get my 8120 back from RMA.

Enable Quad-Channel on Piledriver, I believe the server parts have it and the FX just have it disabled. A lot of enthusiast are OCD (I may be speaking for myself) and like filling those slots, even if it's wholly unecesary (come on, do I need 12GB and 16GB of ram?). So, lets put it to good use!

Recommend that it best be used with 1866Mhz memory.

Together, these May net AMD 15-20% Improvement alone.

Also, when it came to selling the 8120 and 8150, I think to better justify the price jump from the 8120 to the 8150 would have been to ship the 8150 with the Water Cooler. That alone would justify the $80 difference between them, and for people with their own WC or bad ass air-coolers, can get the cheaper 8120.
Under very specific work loads you might see an improvement, but I doubt you will ever see 10% from doubling the bandwidth. Going from 1333 to 1866 only sees around 2% so why would you think doubling something that isn't a bottleneck equal 15%!? I doubt they will have anything more than 5% IPC improvements, maybe 10%. Realistically I can see there being no improvements, maybe just making it smaller, like deciding to chop off the l3 cache and then clocking up 15% higher to get those numbers.
 
Yeah, from the people who brought you BD in 2009... IMHO GloFo has zero chance of shipping 22nm silicon in 2012. Period.

I meant the 2012 doomsday claims to the "I won't live long enough" part.
 
This is Sooo much like the Phenom I(aka Barcelona) it's not funny. Deneb fixed all the issues with Barcelona, and sounds like Piledriver will do the same thing. Something's I'd like to recomend to AMD:

Raise the Base Hyper-Transport! Your raised it a little for the 8120 and 8150 from 2000Mhz to 2200Mhz, but take it to 3000Mhz or 3200Mhz and you'll definitely see some performance gain.

The motherboard-bound HT links don't really like to work much higher than 2200MHz. Like empoy said, the coherent (socket-to-socket) HT links already run up to 3200MHz

When it comes to turbo, have the chip disable every other core as not required that way the core that is doing the turbo would have more available cache available. I hope to experiment with that when I get my 8120 back from RMA.

Enable Quad-Channel on Piledriver, I believe the server parts have it and the FX just have it disabled. A lot of enthusiast are OCD (I may be speaking for myself) and like filling those slots, even if it's wholly unecesary (come on, do I need 12GB and 16GB of ram?). So, lets put it to good use!

That kind of Turbo functionality would require integrated logic to have that kind of granularity.

None of the current dies, and possibly none of the upcoming dies have quad memory channels. Magny-Cours (2x Lisbon), Interlagos (2x Orochi) and Terramar (2x Komodo?) are MCM packages of two dies, where each die has only two memory channels, that is where Socket G34 chips get quad-channel support from.

Recommend that it best be used with 1866Mhz memory.

Together, these May net AMD 15-20% Improvement alone.

Also, when it came to selling the 8120 and 8150, I think to better justify the price jump from the 8120 to the 8150 would have been to ship the 8150 with the Water Cooler. That alone would justify the $80 difference between them, and for people with their own WC or bad ass air-coolers, can get the cheaper 8120.

The current FX chips already support DDR3-1866 memory...
Do you mean support higher memory frequencies, such as DDR3-2000/2133/2400/2667?
 
Guys, AMD's 22nm is scheduled to come sometime in 2013.

PD will refine what is in BD 2011 in 2012...hopefully it's a refinement :D

Even @ 15% improvement clock for clock, and core for core it will be tantamount to an 8 core Thuban. 2 years after the 1100T release. AMD really needs PD to hit a 25 percent improvement from an OC'd 1100T in Single Core and Multi Core performance to begin to interest me to change my mobo to AM3+ with PD 8xxx.

1100T OC'd to 4,2 scores are 1.26 and 7.40 in CB 11.5.

PD is going to have to get 1.58 and a 9.25 CB 11.5 scores, for me to think that it doesn't completely suck.

I might wait for PD or simply grow tired, and just look into SB/IB for my next CPU upgrade. That's how I see it. AMD is at a crossroads.
 
Guys, AMD's 22nm is scheduled to come sometime in 2013.

PD will refine what is in BD 2011 in 2012...hopefully it's a refinement :D

Even @ 15% improvement clock for clock, and core for core it will be tantamount to an 8 core Thuban. 2 years after the 1100T release. AMD really needs PD to hit a 25 percent improvement from an OC'd 1100T in Single Core and Multi Core performance to begin to interest me to change my mobo to AM3+ with PD 8xxx.

1100T OC'd to 4,2 scores are 1.26 and 7.40 in CB 11.5.

PD is going to have to get 1.58 and a 9.25 CB 11.5 scores, for me to think that it doesn't completely suck.

I might wait for PD or simply grow tired, and just look into SB/IB for my next CPU upgrade. That's how I see it. AMD is at a crossroads.

Just to correct, AMD's next process stepping is 28nm, as was displayed in all of the roadmaps from the recent FAD: http://www.anandtech.com/show/5491/amds-2012-2013-client-cpugpuapu-roadmap-revealed
As far as we know, AMD may not use 22nm, since they are depending on whatever process technologies GF will use.
 
Didn't AMD at one point have their own fab?

certainly did.
They spun it off to fund (in part) their purchase of ATI.
I believe a big chunk of that cash also came from the settlement from Intel via anti-competition suite.
 
To answer the OP's question, there was a slide about a year ago showing AMD's proposed timeline of the next processors from Bulldozer onward. This was way before AMD FX (Bulldozer) was released.

Now, AMD FX (Bulldozer module/Zambezi die) is 32 nm as we all know. Piledriver module/Vishera die about a half a year to a year later after BD is 32 nm. Both are socket AM3+. This is according to that roadmap. Piledriver module is second generation Bulldozer with improved performance. It isn't specific whether this is per core performance, per CPU performance, or improvement in IPC or all of the above.

Steamroller will supposedly be a new socket and new process size-- 28 nm, and improved performance of the original Piledriver (2nd gen Bulldozer) modules. After Steamroller, we get Excavator in 2014-ish.

To put it into perspective, look at it this way if AMD's old roadmap holds true:
- When AMD reached 32 nm with Bulldozer modules/Zambezi die, Intel was already on 32 nm (Sandy Bridge) and moving into 22 nm (Ivy Bridge).

- By the time 32 nm Piledriver module/Vishera die is out, Intel will be just starting 22 nm process for its next processor-- Ivy Bridge architecture. Intel will have started using tri-gate transistors, and AMD will still be on traditional CPU transistors.

- When AMD releases their next processor based on Steamroller modules, they will be on 28 nm. Intel will have moved on or will be starting 22 nm Haswell architecture-based production on its next CPUs after Ivy Bridge-- the "tick" in its tick-tock cycle.

- By the time AMD gets around to Excavator modules on a new process (possibly 28 nm) and socket (maybe same socket as Steamroller-based CPUs), Intel will be on its "tock" cycle following Haswell with Broadwell, which should be 14 nm process.​
After Excavator, who knows? I think AMD is going from 40 nm to 32 nm to 28 nm to maybe 20 nm (if I recall) afterwards after four hops. Intel will be 40 nm to 32 nm to 22 nm to 14 nm to 10 nm in five hops.

Hopefully that answers your original question. I am not sure if this roadmap has been updated since Rory took over as CEO.

Their latest slide only shows up to Piledriver module/Vishera die and their Trinity APU.
 
Last edited:
Guys, AMD's 22nm is scheduled to come sometime in 2013.

PD will refine what is in BD 2011 in 2012...hopefully it's a refinement :D

Even @ 15% improvement clock for clock, and core for core it will be tantamount to an 8 core Thuban. 2 years after the 1100T release. AMD really needs PD to hit a 25 percent improvement from an OC'd 1100T in Single Core and Multi Core performance to begin to interest me to change my mobo to AM3+ with PD 8xxx.

1100T OC'd to 4,2 scores are 1.26 and 7.40 in CB 11.5.

PD is going to have to get 1.58 and a 9.25 CB 11.5 scores, for me to think that it doesn't completely suck.

I might wait for PD or simply grow tired, and just look into SB/IB for my next CPU upgrade. That's how I see it. AMD is at a crossroads.

Agreed.....every enthusiast should be hoping for Piledriver to live up to its name. If it isn't atleast a 25-30% improvement over the Thubans, I fear for the future of the market.
 
Back
Top