Thoughts, Rumors, or Specs of AMD fx series steamroller cpu

[Don Karnage]

i'm curious to know why you would even look at it. Whatever improvements they make won't help it catch up to IVY which you are running.

Are you planning on a second machine for a dedicated purpose?

I'm dumping Ivy and going back to 6 core sex of SB-E but if an 8 core piledriver was 160 dollars i would consider picking up a 2nd system just to play with.

 

[Cpus]

I'm dumping Ivy and going back to 6 core sex of SB-E but if an 8 core piledriver was 160 dollars i would consider picking up a 2nd system just to play with.

You might. You can find an FX-8150 for $190. Maybe Microcenter will have a special with the new Vishera CPU when it comes out.

 

[Rvenger]

I'm dumping Ivy and going back to 6 core sex of SB-E but if an 8 core piledriver was 160 dollars i would consider picking up a 2nd system just to play with.

If Piledriver is 15% faster than Thuban and can clock to 5ghz on air at lower power consumption than BD, it will replace my Intel machine.






















Wishful thinking :awe:

 

[Don Karnage]

If Piledriver is 15% faster than Thuban and can clock to 5ghz on air at lower power consumption than BD, it will replace my Intel machine.






















Wishful thinking :awe:

Almost had me till i scrolled down

 

[Homeles]

Seriously, a half node shrink? Where did you read/hear that?

A source was already linked, but AMD desperately needs to take advantage of a transistor shrink. Kaveri will be on 28nm, so there's no reason why AMD wouldn't also use the process for Steamroller. I'm sure AMD would prefer 20nm, but the process is going to be too expensive.

 

[tulx]

Lets wait to see how piledriver performs first. Intel might just buy Amd by 2014 if this next one is a flop

Intel might buy AMD stock to keep it alive, just as MS held Apple on life support, but they'd never be allowed to "buy" (gain a too significant amount of stock share) AMD.

 

[T_Yamamoto]

Failroller

 

[tulx]

Failroller

Thank you for adding value to this discussion with your educated opinion.

 

[Ajay]

A source was already linked, but AMD desperately needs to take advantage of a transistor shrink. Kaveri will be on 28nm, so there's no reason why AMD wouldn't also use the process for Steamroller. I'm sure AMD would prefer 20nm, but the process is going to be too expensive.

Interesting, too bad Piledriver wasn't going to be built on 28nm, given that it's not going into production till 3Q12. Then a 25% jump in performance might have been possible, but I doubt AMD has the manpower now to move designs along at a decent speed.

Excavator, in some slides, has been shown as taking a sharp turn upwards in performance compared to Piledriver and Steamroller, I wonder if that will be 20nm. Do you think that 20nm will remain too expensive for AMD's Excavator? Excavator was supposed to be out in late 2014, but it could be a 2015 product given the revised release dates for Bulldozer and Piledriver (Vishera).

 

[Ancalagon44]

Why are we bothering to discuss SteamRoller now? Its gonna be 2 years late and will probably still get outperformed by Thuban.

 

[Arachnotronic]

Steamroller is what powers Kaveri.

 

[Arachnotronic]

Failroller

Uh huh. I'm not an AMD shareholder anymore -- I lost a bunch of money on them recently -- so I'm definitely not AMD's biggest fan atm. But I think that it's pretty incredible that AMD can still put out products that are even remotely competitive with Intel's (and on the iGPU side, which is becoming much more important, AMD still has a huge upper hand), given their significantly smaller size & much lower R&D budget.

But hey, it's cool to just piss all over AMD.

 

[Rvenger]

Why are we bothering to discuss SteamRoller now? Its gonna be 2 years late and will probably still get outperformed by Thuban.

And C2Q.

 

[Homeles]

Interesting, too bad Piledriver wasn't going to be built on 28nm, given that it's not going into production till 3Q12. Then a 25% jump in performance might have been possible, but I doubt AMD has the manpower now to move designs along at a decent speed.

Excavator, in some slides, has been shown as taking a sharp turn upwards in performance compared to Piledriver and Steamroller, I wonder if that will be 20nm. Do you think that 20nm will remain too expensive for AMD's Excavator? Excavator was supposed to be out in late 2014, but it could be a 2015 product given the revised release dates for Bulldozer and Piledriver (Vishera).

Take a look at this article: www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless

I'll highlight the key images:

Keep in mind, Rory Read from AMD has the same sentiment. Right now, the transition to 20nm isn't looking so hot. 450mm wafers won't be available until 14nm, and I doubt they'll be very cost efficient at first. Essentially, AMD and Nvidia need to work some deal out with TSMC in order to make 20nm a viable process for them in 2013 or 2014.

 

[blckgrffn]

And C2Q.

And K6-2. SRSLY.

They should just shrink that down and use it, why bother with this new crap.

 

[podspi]

There are other advantages to die shrinks than just cost-per-transistor, though. But yes, we may (finally) be reaching the end of Moore's Law. Physically we'll have to eventually, (though hopefully it'll be able to continue on in spirit).

 

[Ajay]

Take a look at this article: www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless


Keep in mind, Rory Read from AMD has the same sentiment. Right now, the transition to 20nm isn't looking so hot. 450mm wafers won't be available until 14nm, and I doubt they'll be very cost efficient at first. Essentially, AMD and Nvidia need to work some deal out with TSMC in order to make 20nm a viable process for them in 2013 or 2014.

I recall, now, reading that article and it does seem like NV is freaking out about cost for 20nm and lower. Based on it's huge volume, Intel could be the only company able to afford smaller process nodes - with others having to wait a considerable period of time for costs to come down. This is a sad state of affairs visa vi the CPU/GPU competition.

AMD is about 1/8th of Intel's size (gross sales $$s); if they can't afford the 20nm process node - who can? Nvidia will need a 50% increase in yields to produce affordable 20nm GPUs!!

Moore's law isn't going to slow process development - cost will. Who would have thought, at the turn of the millennium, that cost would be the biggest problem we would face?

This is a bit depressing

 

[Ajay]

Why are we bothering to discuss SteamRoller now? Its gonna be 2 years late and will probably still get outperformed by Thuban.

I think now that AMD has transitioned to a new architecture the delays won't be so long as they were with Bulldozer and Steamroller will be faster than Thuban. At 28nm, they will have enough transistors available to finally fix all the shortcomings of Bulldozer. Steamroller should be Bulldozer done right, of course, Not that everything will be rosy - Intel will still be walking away from AMD, performance wise, at a brisk pace.

 

[kernelc]

Take a look at this article: www.extremetech.com/computing/123529-nvidia-deeply-unhappy-with-tsmc-claims-22nm-essentially-worthless

Keep in mind, Rory Read from AMD has the same sentiment. Right now, the transition to 20nm isn't looking so hot. 450mm wafers won't be available until 14nm, and I doubt they'll be very cost efficient at first. Essentially, AMD and Nvidia need to work some deal out with TSMC in order to make 20nm a viable process for them in 2013 or 2014.

Hi,
while I understand that smaller shrinks progressively become less advantageous from a cost perspective, I think that at least the GPU guys need them because the size / power contrains of new design (think to Big Kepler and its 7.1B transistors, for example).

On the CPU side, things are a little different: at the moment, in the desktop space we have no benefit for more then 4 cores. Most transistors are nowadays spent on cache and/or uncore or, for integrated design, for the iGPU.

Regards.

 

[kernelc]

I recall, now, reading that article and it does seem like NV is freaking out about cost for 20nm and lower. Based on it's huge volume, Intel could be the only company able to afford smaller process nodes - with others having to wait a considerable period of time for costs to come down. This is a sad state of affairs visa vi the CPU/GPU competition.

Yeah, this is quite probable. After all, Intel is using HKMG since 2007, while others only recently started to use this process tech (and, with the exception of Samsung, with significant problems).

Regards.

 

[kernelc]

I think now that AMD has transitioned to a new architecture the delays won't be so long as they were with Bulldozer and Steamroller will be faster than Thuban. At 28nm, they will have enough transistors available to finally fix all the shortcomings of Bulldozer. Steamroller should be Bulldozer done right, of course, Not that everything will be rosy - Intel will still be walking away from AMD, performance wise, at a brisk pace.

I'm quite curious to see if AMD will continue to use its write-through L1 design + WCC or if it will revert back to a write-back L1.

Maybe they will simply increment WCC size

 

[pelov]

Intel will have volume issues as everyone else will. That's why they're opening up their fabs and that will only further. Intel has a lot of money but not enough such that they can maintain their fab advantage by selling only their own chips. Cost affects everyone and not just TSMC/GloFo.

Haha, I'm glad someone mentioned the write-through approach I think that's going to heavily depend on whether they can increase their clock speeds and lower their misprediction penalty (as per Johan's article). The WCC is essentially the L2 cache so the L1 concerns are directly translated to the L2 problems.

 

[Maximilian]

I might just buy AMD if their next chip is a flop, i found some loose change down the sofa.

I loled.

 

[blckgrffn]

Intel will have volume issues as everyone else will. That's why they're opening up their fabs and that will only further. Intel has a lot of money but not enough such that they can maintain their fab advantage by selling only their own chips. Cost affects everyone and not just TSMC/GloFo.

Haha, I'm glad someone mentioned the write-through approach I think that's going to heavily depend on whether they can increase their clock speeds and lower their misprediction penalty (as per Johan's article). The WCC is essentially the L2 cache so the L1 concerns are directly translated to the L2 problems.

Write-through cache on a CPU makes me shudder.

*shudders*

 

[Joseph F]

Wasn't the last Intel CPU to use write-through L1 the 80486? :awe: