Rumor: AMD "Piledriver" FX CPU production to begin Q3 2012

[Arzachel]

You are very wrong there. AMD used 15% in performance/watt. Never IPC to direct performance.

Never mind that IPC or Performance/Clocks, which is what most mean with the term, by itself meaningless, could you show me an official statement that Vishera is going to bring 15% increase in performance/watt over Zambezi?

 

[ShintaiDK]

Never mind that IPC or Performance/Clocks, which is what most mean with the term, by itself meaningless, could you show me an official statement that Vishera is going to bring 15% increase in performance/watt over Zambezi?

Bulldozer core vs Pilediver for all segments. 15% is most likely too optimistic

 

[frostedflakes]

15% is probably doable actually, resonant clock mesh alone improves perf/watt by like 5-10% IIRC, and Anand mentioned a 10%+ efficiency improvement from optimizing the design by using hard edge flip-flops instead of soft edge. And there are minor IPC improvements as well. I was kind of skeptical after the Bulldozer fiasco, but it looks like AMD will pretty easily be able to achieve their 10-15% perf/watt goal with Piledriver.

Here's the Trinity review where you can read about the under the hood changes in Piledriver.

http://www.anandtech.com/show/5831/amd-trinity-review-a10-4600m-a-new-hope/1

 

[frozentundra123456]

15% is probably doable actually, resonant clock mesh alone improves perf/watt by like 5-10% IIRC, and Anand mentioned a 10%+ efficiency improvement from optimizing the design by using hard edge flip-flops instead of soft edge. And there are minor IPC improvements as well. I was kind of skeptical after the Bulldozer fiasco, but it looks like AMD will pretty easily be able to achieve their 10-15% perf/watt goal with Piledriver.

Here's the Trinity review where you can read about the under the hood changes in Piledriver.

http://www.anandtech.com/show/5831/amd-trinity-review-a10-4600m-a-new-hope/1

Increasing perf/watt is great on a laptop, but 10-15% effiecincy gain, while nice, is not so critical on the desktop. More to the point will be the actual performance increase and will the power use still go through the roof with overclocking.

 

[Ajay]

You are very wrong there. AMD used 15% in performance/watt. Never IPC to direct performance.

I was't very wrong, I made an error - my bad. TDP is supposed to remain the same (don't know what core clocks will be) - so we should see a bump in performance and probably better overclocking or at least lower power consumption at current overclocks. 15% isn't a large enough boost for me to take notice, as I said earlier it would take a 25% jump to catch my attention (and a solid overclock to ~ 5GHz+ @ more reasonable power levels.

As things stand I'm looking to upgrade when IVB-E comes out and AMD likely won't be competitive enough for my consideration. It's too bad, I bought AMD K7s and K8s because they were competitive and I like to root for the underdog - but Intel is so far ahead now that I just can't justify buying AMD.

 

[ShintaiDK]

There is no IB-E on the roadmaps currently (And they do include HW-DT). I wouldnt be surprised if they went directly for HW-E.

 

[frostedflakes]

Increasing perf/watt is great on a laptop, but 10-15% effiecincy gain, while nice, is not so critical on the desktop. More to the point will be the actual performance increase and will the power use still go through the roof with overclocking.

Well at the same 125W TDP, 10-15% efficiency gain means 10-15% more performance. If they can clock the flagship Piledriver SKU at high 3GHz/low 4GHz base clock, I bet they'd be looking at about 15% performance increase over the FX-8150. So it definitely helps. BD clock speeds seem to scale well, only problem was high temperature and power consumption holding back how high they could clock them. But if they can keep increasing efficiency and scaling the clocks, hopefully they can get some respectable performance from the architecture.

 

[pelov]

Well at the same 125W TDP, 10-15% efficiency gain means 10-15% more performance. If they can clock the flagship Piledriver SKU at high 3GHz/low 4GHz base clock, I bet they'd be looking at about 15% performance increase over the FX-8150. So it definitely helps. BD clock speeds seem to scale well, only problem was high temperature and power consumption holding back how high they could clock them. But if they can keep increasing efficiency and scaling the clocks, hopefully they can get some respectable performance from the architecture.

It's difficult to make a prediction because

- AMD has been overstating their supposed increases
- Resonant clock mesh tech has diminishing returns past 4ghz
- There will supposedly be some more architectural tweaks to the Piledriver cores in Vishera
- Nobody has bothered to compare their Trinity laptops with an equivalent Bulldozer or Llano at the same clock speeds so we have no idea just how the Piledriver cores fare with respect to IPC.

Perf-per-watt has improved from Llano > Trinity, which is saying quite a lot as Llano wasn't exactly a power hungry chip and they're both at the same 32nm node. The battery life improvement and load power consumption were both significantly better (more than 10-15%) so a big Piledriver core might be much more efficient than Bulldozer considering how crappy that chip is in power consumption -- if it's over 10-15% Trinity>Llano then Trinity>Bulldozer would be even better.

There's definitely some wrenches to throw into the above figure that might screw with those estimates, but I'd at least expect Vishera to drastically improve upon perf-per-watt of Bulldozer in both multi and single threaded workloads. It won't match SB and definitely won't catch IB as far as perf-per-watt or gross performance is concerned but it could be a good chip

 

[Ajay]

There is no IB-E on the roadmaps currently (And they do include HW-DT). I wouldnt be surprised if they went directly for HW-E.

Well there have been some rumors on the web (softpedia was one, calling out 2Q13, but I can't find it atm). More importantly, Anand mentions it in the IB review - seeing how he is pretty tight with Intel, I considered that to be an approved 'leak' (which Intel could back away from if things work out differently). I think the main thing is that a small vocal minority will be very angry if there is no follow on processor for their high end LGA 2011 boards, so if it's going to be HW-E, it better come out around or shortly after HW-DT and be socket compatible w/LGA 2011.

I'd be cool with the latter, especially if Intel has a better chipset than X79 for Haswell -E.

 

[sequoia464]

especially if Intel has a better chipset than X79 for Haswell -E.

More than two SATA 3 ports would be welcome.

 

[borisvodofsky]

Hoping for the best, I hate how I can't overclock celerons anymore.

 

[ShintaiDK]

Well there have been some rumors on the web (softpedia was one, calling out 2Q13, but I can't find it atm). More importantly, Anand mentions it in the IB review - seeing how he is pretty tight with Intel, I considered that to be an approved 'leak' (which Intel could back away from if things work out differently). I think the main thing is that a small vocal minority will be very angry if there is no follow on processor for their high end LGA 2011 boards, so if it's going to be HW-E, it better come out around or shortly after HW-DT and be socket compatible w/LGA 2011.

I'd be cool with the latter, especially if Intel has a better chipset than X79 for Haswell -E.

Well the lastest confidential Intel roadmap that covers H1 2013 doesnt contain IB-E. However it contains HW-DT.

And I really dont put any credibility to random rumours when we got official documents.

Oh, and I cant see how HW-E should be LGA2011 compatible. The whole socket upgrade is a joke anyway. Look at all the AMD owners who replaced their 8xx boards with almost identical 9xx just for the upcoming Bulldozer. Or all those 6 series owners that went 7 series.

 

[guskline]

I'll wait till the reviewers put the Piledriver Vishera through it's paces before I decided whether or not to part with my 1100 Thuban in rig 3. Glad I didn't fall for the AMD marketing department per-release hype on Bulldozer and sell my Thuban in anticipation. Did CEO Rory fire the entire marketing department?

 

[Soulkeeper]

when it comes to AMD
"I'll wait for the reviews" is like "Don't hold your breath"
you'll literally die on this one

 

[podspi]

- Nobody has bothered to compare their Trinity laptops with an equivalent Bulldozer or Llano at the same clock speeds so we have no idea just how the Piledriver cores fare with respect to IPC.

It really blows my mind this hasn't happened yet. I want to see some benchmarks done with a bunch of chips, constant clock speed, single-core.

The Tomshardware article that did this was great, but I'm not even sure they updated that article to include Bulldozer, let alone Trinity...

And yes, we all know that IPC (is meaningless, doesn't translate into performance/bit your grandmother), but I'd still like to know!

 

[tulx]

A 26% improvement is not enough for AMD. Intel will still have the performance crown.

I suppose it's a good thing that the "performance crown" is absolutely meaningless, then.
Companies don't earn money by selling the fastest processors, they earn money by selling processors of a certain performance range cheaper than their competitors. Whoever does that best, wins.

AMD don't need to beat the fastest intel CPU, they need to make CPU's with a better price-performance ratio. Right now, they're not doing so well in the performance desktop CPU department. Which is why my last two CPU's were Sandy Bridge.

For a mid-range system, though, I'd buy an AMD Trinity APU over an Intel counterpart any day.

 

[Arzachel]

Bulldozer core vs Pilediver for all segments. 15% is most likely too optimistic

As per the Anandtech review, Trinity already beats Llano by 15% on the CPU side at the same TDP except in tasks with purely FP load and that's with a large increase in GPU power at the same time. If Piledriver cores are more efficient than Stars on 32nm, I'm pretty sure that the 15% delta from Bulldozer to Piledriver is actually very pessimistic.

That said, from a purely gaming perspective they would need a far larger increase to beat the 2500k price/perf. As it stands now, the only point where it's worth to go for Zambezi is if you need a ton of cores for cheap (VMs) and I don't think that'll change with Vishera, unless AMD sells 3 module chips for silly low prices.

 

[Phynaz]

Tulx,
When AMD has the fastest processor is when they have the financial success. Quit drinking the marketing coolaid and just look at history.

 

[Arzachel]

Tulx,
When AMD has the fastest processor is when they have the financial success. Quit drinking the marketing coolaid and just look at history.

Not really, the 8ghz Bulldozer didn't really help them

AMD will have financial success when they find more niches where they are ahead of Intel instead of being second in every metric. It doesn't have to be pure performance, performance/W, performance/$ and highly threaded performance would also do. It's just that they're behind on all those metrics. They're pretty well ahead on the iGPU front and Llano seems to have sold very well. They need more of that.

 

[Bill Brasky]

Just in time for Haswell. Good luck with that.

:thumbsup:

 

[Idontcare]

I suppose it's a good thing that the "performance crown" is absolutely meaningless, then.

The performance crown is only absolutely meaningless to those individuals who have no idea what it means.

Having the performance crown means you have a production pipeline that is yielding wafers with a speedbin distribution whose right-hand tail includes a high enough volume of chips that have the desired power-consumption and clockspeed attributes to offer a performance crown SKU in the first place.

This is important because of what it means for the remainder of the distribution of chips that lie to the left of that performance crown SKU. They are yielding in even higher numbers. That mean volume products of the SKU's you want to sell at high ASPs.

If your distribution does not lie far enough to the right such that you have a performance crown SKU binning out then that means your distribution of yielding chips is pushed to the left, lower numbers of yielding chips at even the lower ASP SKU's.

That is your entire gross margin and profit margin ecosystem being shifted to the left (lower numbers) there.

To say "the 'performance crown' is absolutely meaningless" is basically saying "profits and margins are absolutely meaningless"...which is absurd I am sure you'll agree.

AMD has no choice but to price their top-end SKU at a mere $200, which means they scrape together the chips that yield even lower on the wafer (disabling cores, lower clockspeed, power issues) and sell them for even less.

That is not what happens when your chips are yielding such that you have the performance crown SKU included in the natural distribution of yielding bins on your wafers.

That's not to say that having the performance crown means everything either, you can surely have the performance crown and still make unwise business decisions that result in your downfall. AMD choosing to delay 65nm development when they had the 90nm performance crown is an example, DEC going bankrupt while selling the 21164 is another example.

But in general if you find a company that has a "crown" in their industry (be it CPU's, or memory or boats or race horses) then you will also find a company that has superior financials in comparison to their competition because having a performance crown is not the cause, it is the effect of a cause, and the cause that effects a performance crown is also the same cause that effects a rise in all boats for the company, all chips are cheaper to produce at the volumes they want to sell, etc.

No leader in any industry which has the crown will ever go on record as saying it is meaningless, it is a harbinger of good margins, who doesn't want that?

 

[guskline]

when it comes to AMD
"I'll wait for the reviews" is like "Don't hold your breath"
you'll literally die on this one

I understand your point but think back to the Bulldozer. Almost no one had a pre-release of this chip yet the AMD marketing gurus release that video of the "secret lab" where they OC'd the Bulldozer chip on 1 of the 4 modules to over 8 Ghz to set a "new" Guinness record. WHY??? Why not let 25 to 50 of the chips out to reviewers for an honest appraisal?

I think I know the reason. Moreover the Hype created had many, myself included go out and buy the 990FX chipset mbs in anticipation of this new UBER chip. Thank goodness I snagged a 1100 Thuban before they stopped making them.

I realize the Intel SB and now IB processors are better gamer chips and all I was hoping was the Bulldozer would be an improvement over the Phenoms. I hope AMD has learned from this debacle but I'm not sure. I'll wait for the reviews thank you.

 

[Idontcare]

I understand your point but think back to the Bulldozer. Almost no one had a pre-release of this chip yet the AMD marketing gurus release that video of the "secret lab" where they OC'd the Bulldozer chip on 1 of the 4 modules to over 8 Ghz to set a "new" Guinness record. WHY??? Why not let 25 to 50 of the chips out to reviewers for an honest appraisal?

What was really sad in my eyes was not that AMD pulled that stunt but that they managed to convince so many so-called enthusiast OC'ers to tow the company line (essentially turned a bunch of credible XS members into shills) right up until bulldozer was finally released.

That was disgusting, and only harmed the credibility of the enthusiast OC'ing community. That damage won't be undone, ever.

The suspicions will forever taint every future sponsored OC'ing event, and forever dog those poor chaps who got roped into doing it the first time around. And that is ALL on AMD.

 

[podspi]

What was really sad in my eyes was not that AMD pulled that stunt but that they managed to convince so many so-called enthusiast OC'ers to tow the company line (essentially turned a bunch of credible XS members into shills) right up until bulldozer was finally released.

That was disgusting, and only harmed the credibility of the enthusiast OC'ing community. That damage won't be undone, ever.

The suspicions will forever taint every future sponsored OC'ing event, and forever dog those poor chaps who got roped into doing it the first time around. And that is ALL on AMD.

There is a reason there was a rumor that their entire marketing department got canned. Either it happened, or somebody wishes it did


That being said, while I don't disagree that being the best is, well, the best, only one firm is going to be the best. Unless we're saying there is only room for one semiconductor company, (ie: it is meaningfully a natural monopoly on the output scale we expect), there will be room for others.

I also agree that AMD, like the ARM err, army, is going to have a tough time moving forward. But AMD's execution as of late has been excellent. Things are getting released on time, etc. It bodes as well as it is going to...

 

[Ajay]

Well the lastest confidential Intel roadmap that covers H1 2013 doesnt contain IB-E. However it contains HW-DT.

And I really dont put any credibility to random rumours when we got official documents.

Oh, and I cant see how HW-E should be LGA2011 compatible. The whole socket upgrade is a joke anyway. Look at all the AMD owners who replaced their 8xx boards with almost identical 9xx just for the upcoming Bulldozer. Or all those 6 series owners that went 7 series.

Well, I seem to recall Intel promising that IB-E would work on LGA2011 boards. I came up with this slide in a quick search: