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 ...

 

[-Fran-]

This sounds like music to my ears!

I want to get back on AMD if i can, reliability seems to be a bit less for me with intel atm, on my 2nd motherboard in 9 months 🙁
Might just be bad luck however.

I hope AMD can pull a fast one tbh, i want competition, and i am sure many other avid fans do too.

In all fairness, you could talk about reliability only regarding chips with AMD or Intel (northbridge or CPUs), but not MoBos or full components. So it's just bad luck, mate.

And in regards with the Anand article; I already made a comment in the server post (since it was a server oriented article, lol) and I won't repeat it here 😛

It's quite amazing that Anand recognizes there is hope in Pilediver, lol.

Cheers!

 

[jdwii]

but you dated that power hog super 😉


not always
i never use paging 😛
then why ddr4? (other than power saving)


(assumption)
4ghz 4 core pd will be equal to phenom 2 core at 4ghz in ipc (single core performance). But due to module design, it will be equal to 3.2 ghz phenom 2 (in highly multithreaded like 7zip bench). But it may use 65-80w of power in comparison to 125w of phenom 2 😗

I know what your saying i'd say on average you might be right but actually the one thing you forget is 7-zip is one area were Piledriver would still pull ahead because of the newer design but on programs such as handbrake you might be right since Amd's cores scale 80% not 180% per module sorry to anyone who thinks this because under benchmarks you can clearly tell its 80% as Amd slides indicate.

A 8 core BD would Equal 6.4 and a Phenom II has 93% scaling on average(So good!) so a 6 core phenom would have around 5.6 scaling. This allowing the 8 core module design to still have more throughoutput compared to the Phenom. When it comes to multithreading the module design has around the same throughoutput as a 7 core phenom.
Under FP operations its been seen that 4 FP equal 6 that is found on the Phenom, But the bulldozer supports new instruction sets such AVX.

Amd is increasing the L1 cache on Piledriver and also doing other small improvements that will help its IPC a little probably around 7-10% we will have a huge clue when Trinity(desktop version) is released. I hear June 6 is when benches are coming i hope its right!

 

[-Fran-]

Are you sure, jdwii? The 6th of June (hahaha) is right around the corner, so it will make this week very interesting 😛

Cheers!

 

[gamerk316]

Gamer it's already been said Amd is also improving other aspects as well, Plus look here
http://www.tomshardware.com/forum/335164-28-trinity-4100

Trinity is already 10% faster per clock compared to Bulldozer and this is without L3 cache. With this info alone we can hint that IPC will go up! Theirs going to be other improvements as well that will help performance such as a higher clock speed. If Amd prices their parts right Intel will have competition. Will it hurt their I7 or High-end I5's No i don't think it will.


Edit look here as well

http://semiaccurate.com/assets/uploads/2012/05/Piledriver_core_improvements.png

So...10% in best case benchmarking? Thats not exactly a good thing, since 10% over BD still only makes the chip slightly better then a PII X6...

If the chief cause of performance defficencies is in fact a very large penalty for a cache miss, as Anand claims, and this is not addressed, then I fail to see how PD is going to be a significant speedup. Even if you take the 10% improvement as a god given fact, then all PD does is close the gap against IB, without actually catching up any.

 

[-Fran-]

So...10% in best case benchmarking? Thats not exactly a good thing, since 10% over BD still only makes the chip slightly better then a PII X6...

If the chief cause of performance defficencies is in fact a very large penalty for a cache miss, as Anand claims, and this is not addressed, then I fail to see how PD is going to be a significant speedup. Even if you take the 10% improvement as a god given fact, then all PD does is close the gap against IB, without actually catching up any.

Like Anand pointed out in the article, they're compensating that with a larger cache (L2 in particular). Problem is, for desktop usage, you want better hit instead of a larger cache. So yeah, most prolly for the desktop world, the improvements will be close to nothing unless they really improve the hit numbers for the cache.

Once the benchies start rolling out, we'll see for sure. Best thing to do is keep low expectations on desktop loads (games and single threaded apps).

Cheers!

 

[jdwii]

So...10% in best case benchmarking? Thats not exactly a good thing, since 10% over BD still only makes the chip slightly better then a PII X6...

If the chief cause of performance defficencies is in fact a very large penalty for a cache miss, as Anand claims, and this is not addressed, then I fail to see how PD is going to be a significant speedup. Even if you take the 10% improvement as a god given fact, then all PD does is close the gap against IB, without actually catching up any.

10% per clock, Its also been stated by several sites that Piledriver will have a 4.0Ghz+ clock speed. IPC improvements plus the new Clock mesh tech not to mention anything else Global foundries improved should make Piledriver look 15-20% better on benchmarks not to mention make Piledriver around 10-20% more efficient as well. If priced correctly this should be a decent chip Amd does not need to "Top" Intel all they need to do is make a price/performance killer which is going to be hard since the I5 ivy is so good.

I wish Bulldozer was priced as low as it is today(8150 189.99$ and the 8120 priced at 169.99$) at release i would of given Amd a more laid back response to their newer design. I had to spend 179.99$ to get my 1100t back in October which means the 8150 would of been worth it for me as well.

I hope Rory read does not let Amd's marketing team handle the Piledriver release with Upto statements.

Are you sure, jdwii? The 6th of June (hahaha) is right around the corner, so it will make this week very interesting 😛

Something like that i hope so.

 

[jimmysmitty]

This sounds like music to my ears!

I want to get back on AMD if i can, reliability seems to be a bit less for me with intel atm, on my 2nd motherboard in 9 months 🙁
Might just be bad luck however.

I hope AMD can pull a fast one tbh, i want competition, and i am sure many other avid fans do too.

Its EVGA mobos 😛

EVGAs cards are great and their support is great. But I have seen a lot of dead EVGA mobos. Not as many as some other brands but way more than say Asus, which personally I have not had any die on me other than one that was 8 years old and even that was still going I just saw thw caps starting to buldge.

Like Anand pointed out in the article, they're compensating that with a larger cache (L2 in particular). Problem is, for desktop usage, you want better hit instead of a larger cache. So yeah, most prolly for the desktop world, the improvements will be close to nothing unless they really improve the hit numbers for the cache.

Once the benchies start rolling out, we'll see for sure. Best thing to do is keep low expectations on desktop loads (games and single threaded apps).

Cheers!

The other issue with a larger cache is a larger area to waste power. More leakage. While it would help to keep the chip cooler, it may also use more power and a lot more transistors as cache tends to do.

But masking the problem is not a solution. Of course I don't think AMD can wait it out and not release a PD in lieu of a new arch. Better to get something that may be a bit better out now than lose sales and fall even more behind.

 

[anxiousinfusion]

10% per clock, Its also been stated by several sites that Piledriver will have a 4.0Ghz+ clock speed. IPC improvements plus the new Clock mesh tech not to mention anything else Global foundries improved should make Piledriver look 15-20% better on benchmarks not to mention make Piledriver around 10-20% more efficient as well.

I've seen a lot of posts now that recognize a set of the improvements being made in Piledriver. Can we maybe get a master list together so people can just reference that?

 

[noob2222]

Like Anand pointed out in the article, they're compensating that with a larger cache (L2 in particular). Problem is, for desktop usage, you want better hit instead of a larger cache. So yeah, most prolly for the desktop world, the improvements will be close to nothing unless they really improve the hit numbers for the cache.

Once the benchies start rolling out, we'll see for sure. Best thing to do is keep low expectations on desktop loads (games and single threaded apps).

Cheers!

they were concentrating on server usage for their article.

We do agree that it is a serious problem for desktop applications as most of our profiling shows that games and other consumer applications are much more sensitive to L2 cache latency. It was after all one of the reasons why Nehalem was not much faster than the older Penryn based CPUs. Lowly threaded desktop applications run best in a large, low latency L2 cache. But for server applications, we found worse problems than the L2 cache.

Gamerk left that part out.

low latency cache is not BD's strong point by any means. PII was 15 clocks vs 21 in BD, hence the "PII has better IPC".

Then there is the penalty for running CMT vs 1 alu in a module.

Another significant problem is that the L1 instruction cache does not seem to cope well with 2-threads.

 

[fazers_on_stun]

From the article:

The Real Shortcomings: Branch Misprediction Penalty and Instruction Cache Hit Rate

Bulldozer is a deeply pipelined CPU, just like Sandy Bridge, but the latter has a µop cache that can cut the fetching and decoding cycles out of the branch misprediction penalty. The lower than expected performance in SAP and SQL Server, plus the fact that the worst performing subbenches in SPEC CPU2006 int are the ones with hard to predict branches, all points to there being a serious problem with branch misprediction.

So...if the branch predictor is being blamed as the primary cause of performance woes, and AMD admits very little performance increase of PD is comming out of branch predictor improvements, where exactly is PD getting its IPC improvements from? Based on this analysis, you'd have to conclude that PD is basically going to gain performance through clock speed increases, which would NOT be a good sign, as it indicates AMD is either unable or unwilling to make major improvements to the underlying architecture.

Sheesh - that article is a real eye-opener about not only Bulldozer, but AMD in general!

Some, including sources inside AMD, have blamed Global Foundries for not delivering higher clocked SKUs. Sure, the clock speed targets for Interlagos were probably closer to 3GHz instead of 2.3GHz. But that does not explain why the extra integer cores do not deliver. We were promised up to 50% higher performance thanks to the 33% extra cores, but we got 20% at the most.

The Front End: Shared Decoders

Quite a few reviewers, including our own Anand, have pointed out that two integer cores in Bulldozer share four decoders, while two integer cores in the older “K10” architecture each get three decoder. Two K10 cores thus have six decoders, while two Bulldozer cores only have four. Considering that the complexity of the x86 ISA leads to power hungry decoders, reducing the power by roughly 1/3 (e.g. four decoders instead of six four dual-core) with a small single-threaded performance hit is a good trade off if you want to fit 16 of these integer cores in a power envelope of 115W. Instead of 48 decoders, Bulldozer tries to get by with just 32.

IIRC a certain poster here blew his stack when I suggested the shared front end was part of BD's problem, after his long treatise on the L2 cache being the only problem 😛..

But the most telling part is this:

Integer Crunching Power

Each core has two integer executions units (EX0 and EX1) and two AGUs (Address Generation Units). For comparison, the K10 core inside Magny-Cours and Istanbul had three ports to a “Fully featured ALU + AGU” couple. AMD marketing cleverly drew four pipeline blocks inside the Bulldozer integer core, but those powerpoint blocks cannot hide the fact that each Bulldozer integer core has fewer execution resources.

In practice, the AG0 and AG1 are little more than assistants with limited capabilities to EX0 and EX1.The software optimization guide for AMD family 15h processors lists only a few instructions (page 248 in the january 2012 version) that can be processed by the AG0 and AG1 execution units and each time the remark "First op to AG0 | AG1, Second to EX0 | EX1" is made. The AG0 and AG1 execution units reduce the latency of the CALL and LEA instructions, but the maximum throughput of each integer core inside the Bulldozer module is only two integer instructions per clock cycle. It's only when a fused branch enters EX0 and another integer instruction can enter EX1 that we have a slightly higher throughput of three integer instructions.

So the Bulldozer integer core can execute one integer instruction less per cycle (2 vs 3). That doesn’t mean that the Bulldozer integer core is only 1/3 slower, however. The integer core of Bulldozer is smaller but also more flexible. The per lane dedicated 8-entry schedulers are gone, and a much larger 40 entry scheduler replaced it. This means that Bulldozer should be better at extracting ILP (Instruction Level Parallelism) out of code that has low IPC (Instructions Per Clock).

I think most of us back in the BD thread a year ago fell for this - 4 pipes per core to bring BD up to equivalence of Conroe & later Intel CPUs finally. Looks to me that AMD marketing had more control over BD than the engineers did. Hopefully PD will somehow ameliorate some of these shortcomings, but given Read's recent statements about 'not competing', I suspect no massive redesign efforts will be made..

 

[noob2222]

And the result of poor marketing and PR team prepping for bulldozer launch? YOUR FIRED!!!

AMD's initial layoff announcement yesterday implied that the dismissals would occur across the company's global sales force. While that may still be true, it has become clear that AMD has slashed its PR and marketing departments in particular

http://hothardware.com/News/AMD-Layoffs-Maul-Marketing-PR-Departments/

Marketing was the biggest disaster for Bulldozer. Get people to believe that its going to be faster than anything ever made before, then deliver something in-line with their current PII x6 ... Id fire them too.

 

[Reynod]

My apologies for the mass delete but I removed all references to two personal discussions with users not present on these forums, or able to represent themselves personally, or wanting their personal discussions regarding a product discussed on this forum.

My apologies again.

 

[truegenius]

🙁 but you deleted my link of amd bd hitting 9 ghz
why
i guess due to lazzyness 😛

 

[Reynod]

Yes ... feel free to add it again though.

There was too many other posts to delete it would have taken me quite a while otherwise.

My apologies again.

 

[de5_Roy]

"PCIe3 support vanishes from AMD roadmaps"
http://semiaccurate.com/2012/06/01/pcie3-support-vanishes-from-amd-roadmaps/
"AMD delays desktop Trinity one quarter"
http://semiaccurate.com/2012/06/01/amd-delays-desktop-trinity-one-quarter/
"ECS Rolls Out First Socket FM2 Motherboard - A85F2-A Deluxe"
http://vr-zone.com/articles/ecs-rolls-out-first-socket-fm2-motherboard--a85f2-a-deluxe/16119.html
"AMD gives up monthly catalyst updates"
http://www.fudzilla.com/home/item/27386-amd-gives-up-monthly-catalyst-updates
this one seemed more informative
"AMD Discontinues Monthly Driver Updates, Releases Catalyst 12.6 Beta"
http://www.anandtech.com/show/5880/amd-discontinues-monthly-driver-updates-releases-catalyst-126-beta

 

[truegenius]

"AMD delays desktop Trinity one quarter"
http://semiaccurate.com/2012/06/01/amd-delays-desktop-trinity-one-quarter/

then haswell will blow it away, thus Another Major Disappointment

 

[de5_Roy]

last rumored launch date was august iirc. the new one is september. right before windows 8's rumored launch. oh teh blasphemyyyymmdd

 

[king smp]

My apologies for the mass delete but I removed all references to two personal discussions with users not present on these forums, or able to represent themselves personally, or wanting their personal discussions regarding a product discussed on this forum.

My apologies again.

that is okay I was about two pages behind in reading and now I am caught up LOL


as far as PD goes
AMD doesnt need to beat Intel
it just needs to beat its own older architecture
BD was a disappointment to me because in alot of benches the PHII beat it
It is one thing not to beat your competitor but to not beat your own older product
that is really bad
I had a PHII 925 @ 3.5ghz that I upgraded to a PHII 965 @ 4ghz for $90 USD
and I would buy a 1090t or 1100T before going to a AM3+ FX

 

[viridiancrystal]

Since when did we have 12 quarters in a year?

 

[4745454b]

It's tough to be a fan of AMD right now. So much going wrong. Not competing with Intel, new chip that has issues keeping up with older one, GPUs that do ok, but presumably are no faster then Nvidia's midrange, driver releases that can't keep schedule so they bail on monthly releases, etc. I need some sand for my head...

 

[esrever]

It's tough to be a fan of AMD right now. So much going wrong. Not competing with Intel, new chip that has issues keeping up with older one, GPUs that do ok, but presumably are no faster then Nvidia's midrange, driver releases that can't keep schedule so they bail on monthly releases, etc. I need some sand for my head...

why would it matter whats happening. If you are a fan then it wouldn't phase you one bit.

 

[jdwii]

that is okay I was about two pages behind in reading and now I am caught up LOL


as far as PD goes
AMD doesnt need to beat Intel
it just needs to beat its own older architecture
BD was a disappointment to me because in alot of benches the PHII beat it
It is one thing not to beat your competitor but to not beat your own older product
that is really bad
I had a PHII 925 @ 3.5ghz that I upgraded to a PHII 965 @ 4ghz for $90 USD
and I would buy a 1090t or 1100T before going to a AM3+ FX

They don't have to beat Intel but my god they should at least have something new out to compete not Another Minor Delay.
I think this is what Amd means but i'm not sure what one.
Another Minor Delay
Another Major Disappointment

It's tough to be a fan of AMD right now. So much going wrong. Not competing with Intel, new chip that has issues keeping up with older one, GPUs that do ok, but presumably are no faster then Nvidia's midrange, driver releases that can't keep schedule so they bail on monthly releases, etc. I need some sand for my head...

Funny i can still find only 2 680's at newegg not to mention its only 10% faster while being around 10% more efficient while costing around 10% more compared to the 7970. But i'm real disappointed in Amd i just don't think this stuff is going to end their limited supply issue can be traced back to the Athlon days. Throughout their whole existence their usually a generation behind Intel or worse(from 1990-2012). At least they helped keep Intel honest.

 

[jdwii]

Maybe Amd should do what Apple did back in the day and hire the CEO of Pepsi. We all know how much a CEO of a soda company understands new technology.

 

[Haserath]

Aeronautical Maintenance Disaster

Flies great until the crash landing. Also known as falling with style.
--
Desktop Trinity and the driver updates don't seem like a big problem. I bet the drivers will be like Nvidia's setup. Release Beta's often and WHQL drivers when enough changes occur.

The bad part about the Trinity delay is that it's another delay... Mainstream desktop isn't that important anymore, but every little thing helps.

And Agreeing with Semi that Pci-e3 is a "checkbox, 'must have'" item for desktop at least(which isn't important). HPCs won't like this though...

Let's see how AMD flies before giving up on it entirely.

 

[r0ck3tm@n]

Aeronautical Maintenance Disaster

Flies great until the crash landing. Also known as falling with style.
--
Desktop Trinity and the driver updates don't seem like a big problem. I bet the drivers will be like Nvidia's setup. Release Beta's often and WHQL drivers when enough changes occur.

The bad part about the Trinity delay is that it's another delay... Mainstream desktop isn't that important anymore, but every little thing helps.

And Agreeing with Semi that Pci-e3 is a "checkbox, 'must have'" item for desktop at least(which isn't important). HPCs won't like this though...

Let's see how AMD flies before giving up on it entirely.

I encourage every desktop/gaming enthusiast to buy Intel for this generation. AMD clearly is walking away from this area of development and I do not like the direction the CEO is leading the company. When AMD is again serious as Intel is about their high performance CPUs, I will be a serious AMD fan again. In the meanwhile I will keep reading and dreaming about the AMD that could be.

To the fans who believe AMD is going in the right direction, I will ask this question: How can one expect a first class APU from a company who is not investing in first class CPU technology?