AMD Piledriver rumours ... and expert conjecture

Page 271 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
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 ...
 
Larger Address Space, allowing for more then 2GB of RAM to be accessed. Even on Win64, the old 2GB Address Space limit remains in effect (4GB, if the application is compiled as LAA). This limits what you can do, especially since textures take up so much of that space...You also get the benefit of all the extra Registers that Win64 offers, which can give a significant boost to performance, depending on how register intensive the application in question is. Finally, you remove the need to go through WOW32, which carries a slight performance hit (2% or so).

I said general applications. They dont need the 64 bits. They barely use 32bit registers as it is.

Which is idiotic. Its basically a cludge within the kernel, basically a revere WOW32. Essentially, you run the CPU in long mode, but put an artificial limit on memory usage. Idiotic concept that was developed so the devs could drop support for 32-bit linux.

its not a kludge. Its for those applications that require 64bit compoutation but dont need that much RAM. So the ALU is 64 bit, but the memory pointers are 32 bit only. Its not as if every CPU bound app needs multi Terabyte memory.Linux is run on mobile devices also, where every MB RAM saved matters.
And there is no conspiracy by kernel devs to root out x86 compatibility.
 



I said that a while back, AM3+ is far to old and too overpriced to be failing out where they are, it would be nice to see a bit more competitive pricing like seen with Radeons.

The article is good, the Trinity is nothing short of a fantastic product, I have not quelms suggesting it over i3's and i5's in the HTPC realm, for budget gaming just consider the flexibility. I have personally seen the results, reading other results one can see its IGPU solution is capable of modest res and modest presets, add to that hybrid xfire and even further testing it with a HD7950 didn't show off much drop in performance bar CPU bound games but still maintained hefty frames. In short the trinity packs a lot of potency and its about time reviewers suck up their pride and actually give it the credit it deserves.
 


Ummm...yes they do. Registers are the primary bottleneck in computations. More registers can have a huge impact, especially in anything that does mathematical computation.



Umm...you've had 64-bit datatypes in 32-bit OS's for some time now. That's why calculator.exe can handle calculations with results greater then 4,294,967,295. And when writing applications for platforms with server memory constraints (like I do), you typically use custom sized datatypes as needed, rather then pack everything into a 32/64 bit word. [For example, if you have a lot of for loops that iterate 10 times, you might create a 4 bit integer type, just to save the extra 28 bits.] The ONLY space savings would be on the 32-bit pointers; that's it. There is nothing else that you are saving by having a separate 32-bit ABI within the OS.
 


Take into account the power used on each. For the 7970, I'm sure those 250W+ were not all on the GPU, but like 85%+ of them were.

Point is: with a big power envelope you can do stuff like that (big computing power), off course. The trick is getting the power down in CPUs on each gen (for whatever reason) and not breaking the 300W barrier for PCIe 2 (more in PCIe 3 AFAIK).

So, if the power used is proportional to what the 3960X consumed, then they're not so different. Also, there are some more layers the GPU suffers at the moment to actually be used.

I wonder how HSA will do that, precisely.

Cheers!
 
http://www.phoronix.com/scan.php?page=news_item&px=MTIwNDY

interesting. AMD is adding Steamroller optimisation patches to the GCC.
Instructions supported : BMI, TBM, F16C, FMA, AVX, XOP, LWP, AES, PCL_MUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, and ABM

The bdver3 contains three pipelined FP units and two integer units. Fetching and decoding logic is different from previous fam15 processors. Fetching is done every two cycles rather than every cycle and two decode units are available. The decode units therefore decode four instructions in two cycles.

Three DirectPath instructions decoders and only one VectorPath decoder is available. They can decode three DirectPath instructions or one VectorPath instruction per cycle.

The load/store queue unit is not attached to the schedulers but communicates with all the execution units separately instead.

bdver3 belong to fam15 processors. We use the same insn attribute that was used for bdver3 decoding scheme.

more details in the article.
 
http://semiaccurate.com/2012/10/11/what-is-the-latest-on-amds-sea-islands/

Regarding AMD HD8xxx
Sea Islands is not going to be a 2012 product, it is looking like a March 2013 release for now. Things could of course change, but that is unlikely given what the card is. CI is a 28nm evolution of the current 7xxx line of cards, think optimizations rather than massively new features.

As far as performance goes, AMD was promising AIBs roughly a 30% speed boost from the new silicon a few months back. That we hear is fairly optimistic, with more measured sources saying +15% is more likely. AMD is power bound and die size equals cost, so what you are looking at is process improvements and algorithm optimizations rather than any new big bangs.
 
BD in Linux is actually almost even with Sandy and IB, for the most part, after GCC 4.7 included most optimizations for it.

I'm guessing PD could even be a little ahead of Sandy at least in Linux when it arrives. Too bad 99% of Windows programs cannot be compiled (actually, won't let you) on your local machine to take advantage of it...

Cheers!
 


I dunno anybody, except maybe Sarinaide who insists that Trinity is great for desktop 😛..

However the point of the article is not to recommend a build :pt1cable: but instead to show that the i3 is much further up the "CPU-bound" curve in gaming than Trinity is. As you probably already knew 😀, you have to take the GPU out of the equation if you want to compare CPUs when gaming.
 


Well the Cinebench score only improved maybe 5% multithreaded but we should wait a (hopefully) couple weeks and see how the actual PD product reviews go.
 


IIRC they were saying the same thing about Bulldozer just before its release a year ago..
 


USD$253, huh... I might grab one in day 1 then. Hopefully Frys or BestBuy get them on day 1 for me to pick them up; otherwise, Amazon would be the one.



Well, the only desktop usage for Llano or Trinity has always been around HTPC labor with a little flavor of gaming. For a mini ITX or micro ATX board, with 1 PCIe x16 slot, I'd say that using Trinity is better suited to get something else in that slot. Like a good sound card or more storage and not a video card.

For every other econobox use out there, an i3, lower priced i5s, Pentium G or a A6+ is as good as it gets and they can all be recommended IMO. Like I said, Excel is still single threaded AFAIK 😛

Cheers!
 


Newegg? Try microcenter i'm sure they already have some but can't sell them yet
 


Actually I did that about 4 pages ago, somewhere back there.

skyrim%201280.png


A10 to A8, you take an 8% performance hit when you lose 33% of your shader cores.

Take those 128 shader cores and run FP calcs, how much performance is gained? shouldn't be hard to guess that it can easily be more than 8% since thats basically giving you 2 full SIMD engines (wich im guessing will be = to 2 fpu, but if a shader core = fpu ... ouch, 128 extra fpus) ... you just went from 2 physical fpu to 4.

So an A10, running 2 SIMD engines for FPU will likely be faster than the A10 with 0 boost to fpu, esp since games are generally fp heavy.

Obviously you can't use all your shaders for fp calcs until you have a discrete gpu.
 


Honestly, why bother? The guys at Xtremesystems (and a few other places, too) have already concluded that all Cinebench versions up to R11.5 don't use the proper instruction sets for AMD CPUs. Then, the question is: why should people still bother so much with this software after it has been proved an unreliable means of comparing CPUs? Once R14 is superseded, it might be proved just as unreliable as the previous versions. Just follow up the thread at Xtremesystems; a 2600K gets a 10% improvement in going from R11.5 to R14, whilst a Phenom II X4 gets 20% and an 8150 gets around 25%. People who keep on drawing conclusions mainly from this software should know better, and it wouldn't matter a bit if Piledriver scored better than any other CPU in Cinebench.

http://www.xtremesystems.org/forums/printthread.php?t=276245&pp=100&page=9
 


7. Integrated PCH or south bridge (Already mentioned for Broadwell)

8. DRAM stack


Intel has been putting multiple die in the same package since the Pentium days.

293px-Pentiumpro_moshen.jpg



Here the L2 cache is on a separate die in the same package. Now we have integrated L2 and L3 cache, opening room for eDRAM/DRAM depending on how much capacity they want.
 
FYI, anybody here unfortunate enough to own AMD stock might wanna get up early tomorrow and put in a sell order:

http://seekingalpha.com/article/919691-advanced-micro-devices-warns-on-weak-demand-what-s-next?source=msn

Advanced Micro Devices (AMD) announced preliminary financial results on 10/11, noting that it would see a 10% sequential drop in revenues against the guided 1% drop, +/- 3%. Further, the company announced that its gross margins would be around 31%, rather than the previously guided 44%, due to both a $100M inventory write down, as well as lower than expected selling prices.

With the weakness in the PC space suggested by Intel's (INTC) pre-announcement, coupled with HP (HPQ) and Dell's (DELL) fairly weak results in their most recent quarters, it was natural to expect that AMD would also suffer some weakness. The problem here is that AMD's guidance for the current quarter was already particularly weak (whereas Intel's was quite bullish and revised down to bearish), so a warning on top of an already weak guidance is particularly painful.

A Net Loss Expected In The Quarter

With this updated guidance, it is now certain that the company will experience a net loss. With gross margins of 31%, revenues 10% down from the previous quarter to roughly $1.26B, and operating expenses down 7% from the prior quarter's $557M, the operating loss will likely be about $128M, or -$0.18/share. Assuming flat interest expense compared to the prior quarter of $43M, the net loss should be even wider than the straight operating loss.

Q4 Guide - Unlikely To Be Good

With the company writing down $100M of inventory, it is unlikely that the Q4 guide will be particularly strong. On the previous call, the company had assured investors that the majority of the new inventory that had been built up was next generation "Trinity" products and not, as feared, the previous generation "Llano" products. However, the inventory that had already been built -- especially on the desktop chip side -- remained.

AMD attempted to hold off the coming of the next generation "Trinity" parts on the desktop in order to clear inventories, but it seems that the company was unable to do so. A part of this is likely due to the fact that AMD had created something of an "Osborne Effect" by talking up its upcoming products while its new products had just hit the shelves. This likely had a material effect on the demand for its current products, as AMD created the perception that better stuff was perpetually "just around the corner." Another problem, of course, is that Intel is seemingly more aggressively competing in the low end that AMD seems to dominate.

In the mobile space, it is somewhat surprising to hear that weakness continued, as the company had indicated on the previous call that shipments of its "Trinity" APUs in the mobile space had been quite healthy. Even with broad PC weakness, AMD's market share is fairly small compared to Intel's, so it was not inconceivable that the company would be able to take market share. Apparently, though, the preliminary numbers suggest that Intel may have actually gained share. The problem here is that Intel already owns about 80% of the consumer PC CPU market, so it is unfortunate that even in light of a contracting PC market, AMD is unable to make inroads.

Finally, even though AMD holds a scant 5.5% of market share in the server space, it seems that the company was unable to take share in the quarter. In its warning, Intel reported that its data center growth remained on track. Unfortunately, AMD is apparently not riding that wave.

CEO's Job In Jeopardy?

While this quarter's warn won't kill AMD -- its debt and cash positions will be able to take the net loss without causing liquidity problems -- two earnings misses in a row reflects poorly on CEO Rory Read, who has been with the company since August 2011. While the newcomer has admirably managed to cut operating expenses, the company continually misses its already conservative revenue targets.

Now, in Mr. Read's defense, the roadmap and products that are being put out now were likely not developed under his watch -- CPUs take about four years to develop -- so he is working with what he was given. Further, the broad weakness in the PC space is undeniable, and even Intel was not immune.

I suspect Read will keep his job in the near- to medium-term, but the company's strategy and its management will continue to be under severe scrutiny.
 
derp

sorry wrong post and stuff lool delete this post?

edit:

Ah ok I'll make a post.

Intel doesn't buy AMD because it is arrogant? Or it knows people won't trust Intel as much if it is the only company building cpus. Or doesn't actually have the money, because of investments. Or just doesn't want too..
 
Status
Not open for further replies.