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 ...

 

[gamerk316]

As I said though, the point is speed and efficiency. Power usage for total system would drop with on die memory and performance would jump up since latency would drop tremendously.

Well, at the very least, moving RAM onto the chip would seriously eliminate the need for the L3, if not the L2 cache, depending on how low latencies can be driven. So thats one area where you can pick up the die space needed to put memory directly on the chip itself.

 

[gamerk316]

is anyone interested in discussing tdp reduction in trinity, that means if amd have managed to get clock speed above 3ghz in under 40w tdp then it can be a +point for amd to competete against intel's lead (in smaller fabrication process).

After this much power saving, how much clock speed you are assuming at what tdp and in which config (x2,x4,x6 or x8).

I'll say it again: Since AMD is going from llano [Athlon arch] to Trinity [BD arch], expect a slightly higher jump in performance then we saw between Phenom II and BD. And that assumes that they can increase the clock by as much in Trinity without running into thermal limits.

So as far as the CPU is concerned, I'm predicting a 10-15% boost on average; higher in threaded applications, lower otherwise. GPU will likely be more powerful then this though; 30% isn't outside the realm of possibility.

 

[fazers_on_stun]

The Sultan of Brunei has solid gold toilets in his palace

Hmm, you actually been there and used his facilities?? 😛

I'd much rather have a smart, porcelain Japanese toilet/bidet to squirt my arse for me and play me some iTunes or browse the net, using a 4100 or an i3 😀.

But if those gold toilets come with harem girls, then OK I'll go for a dumb solid-gold toilet instead 😗 😀..

 

[king smp]

so liquid cool it!!!

water cooling especially the self contained H-series has drastically gotten cheaper and easier

a liquid cooled TWK 42 CPU would be freaking awesome
just for the experience
and for the bragging rights

right now your 965be on air with a mild OC is at 3.8
you know with even a good H-series like the H80 you could probably go up to
4.5 with that TWK and wouldnt have to go over 1.48 on volts

a Deneb can safely go to 1.55v according to many reviews and even Toms
with that TWK 42 you could stay under 1.50 and still be in the 4+ghz range
darn push it a little and you have a chance of hitting 5ghz

as a tech and enthusiast why would you not do it?

 

[Cazalan]

Didn't Elpida just go out of business?

They filed bankruptcy protection. Some company would end up buying the fabs and taking over.

 

[Cazalan]

Thermal dynamics doesn't work that way. If it did then BD's thermal issues could be solved overnight and we'd have much faster CPU's then we do now.

I suppose you've done the research?

Google it, there's like 100 papers on the subject.

"TSV thermal array"
"Thermal-Aware Cell and Through-Silicon-Via Co-Placement for 3D ICs"

http://web.mst.edu/~yshi/DAC2011_placement.pdf

"Since TSVs are the major channel for heat flow, their distribution also has a
significant impact on the temperature. A survey on concurrent TSV planning within thermal-aware 3D floorplanning and 3D routing is given in [19]. We first devise a simple criterion to guide the placement of TSVs for achieving the lowest temperature."

 

[Cazalan]

That is why I know Caz's got the wrong idea. On-chip memory isn't capable of replacing main desktop memory. Right now is 512MB on a low heat chip, could you Run windows 7/8 on 512MB of memory? Not comfortably, which destroys any performance advantage you'd get from it. In two to four years, when it becomes possible to put 2GB of memory on a slow chip, we'll be using 32~64GB of main memory and you'll be faced with the same problem. No matter how much you pile into the CPU it'll always be cheaper and more economical to put it outside, and your requirements will grow such that it won't be possible to meet with what is technically available at that time. Need I remind everyone of Bill Gate's famous quote of 640KB being "enough" memory. Caz is basically making the exact quote, that 2GB is "enough" when we'll be running 32~64GB.

Try to put words in my mouth all you want but it doesn't change the fact that the mainstream market is heading towards more compact and efficient systems. I never said the high end chips and discrete GPUs would go away. That's just asinine.

When you cut the max capacity in half it makes 2GB seem small. Now try 4GB and it doesn't sound so bad. To enthusiasts is does but go to Walmart and see how many netbook/laptop computers ship with just 1GB-2GB of memory.

Windows 8 (32-bit) requires 1GB memory
Windows 8 (64-bit) requires 2GB memory

How many millions of people are happy with their iPad that has only 512MB RAM?

 

[Cazalan]

Now what's really funny is that your so caught up with vertical stacking that you never sat back and saw what ~is~ possible, horizontal placement of a 3D memory stack. They stack them vertically to better make use of limited physical space, this is a problem in mobile applications (and real estate out here). In a desktop computer you have plenty of room on the CPU mounting board. The CPU die is often 20% or less of the actual size of the socket, which means tons of horizontal space is being wasted. Instead of stacking memory ontop of the CPU you put it horizontal with the CPU with the interconnects being on the bottom layer and fusing to the side of the CPU. Fundamentally it's the same concept, two separate die's but instead of them connecting in a vertical manor you have the bottom most layers connect in a horizontal manor. This allows the thermal load from the CPU unimpeded access to the heat plate while leaving you plenty of room for a larger memory stack. Still not enough to compete with main memory, but more then enough to be GPU memory or a large cache. Requires more engineering work, especially as you'll need a finely fitted heat plate, but more then doable.

Now you're getting it! Apparently you missed when I wrote:

"* Another couple years out the "main memory" will be stacked on or next to the CPU but in the same package (similar to how Core2Quad and Interlagos is made now) "

3D-IC is more than just singular stacks.

 

[jdwii]

I'll say it again: Since AMD is going from llano [Athlon arch] to Trinity [BD arch], expect a slightly higher jump in performance then we saw between Phenom II and BD. And that assumes that they can increase the clock by as much in Trinity without running into thermal limits.

So as far as the CPU is concerned, I'm predicting a 10-15% boost on average; higher in threaded applications, lower otherwise. GPU will likely be more powerful then this though; 30% isn't outside the realm of possibility.

That's what i'm thinking!

 

[jdwii]

was there really a jump in performance over say a Deneb C3 or Thuban E0...? no.
just higher clocks.
overclocking crap just makes hotter and louder crap..

My pc is pretty quiet! But then again i'm under stock voltage, And i have the X6 OC pretty good.

 

[dogman_1234]

Mal, any news on PD yet, or is AMD holding out again on us? I may have to agree with the poster who said that AMD will wait until IB comes out to give us a 'hint'. 😉

 

[dogman_1234]

PD doesn't need to come out roaring like the enthusiasts want it to. What I want PD to do is fix the...was it cache and scheduler issues...whatever issues it had. ( Sorry, can't remember at the moment. 🙁 )

Also, I think the implemented prices for BD was bad as well, from its performance perspective.

Oh, and make the silicon cooler. I want to Fold w/o burning my house down!

 

[king smp]

this might be old
http://semiaccurate.com/2012/03/12/why-did-tsmc-stop-28nm-production/

this might be interesting
the whole reason I went out and bought 8gb recently
http://www.xbitlabs.com/news/memory/display/20120313194311_DRAM_Pricing_Continues_to_Climb_Slightly_Market_Observers.html

also this
http://www.xbitlabs.com/news/graphics/display/20120308194917_Nvidia_Shows_GeForce_Kepler_in_Action_as_Photos_of_GeForce_GTX_670_Ti_Hit_the_Web.html


and while I know this is a PD thread
check out the competition
http://ivybridgeprocessor.com/2012/03/14/ivy-bridge-processor-die-layout-estimated/


and take a look at AMD stock
http://ycharts.com/companies/AMD

compared to Nov 2011 they are kicking butt (well not Intels LOL)

 

[Blandge]

Oh, and make the silicon cooler. I want to Fold w/o burning my house down!

I hear they tried to etch pictures of dragons onto the surface of the chip but that only made the problem worse.

But serously, I believe it was cache latency and thread/core scheduling in the OS. Obviously AMD doesn't have much control over the scheduling issues.

Also, I'm curious as to what you guys think the successful branch predition rate of AMD and Intel CPUs is?

 

[dogman_1234]

Also, I'm curious as to what you guys think the successful branch prediction rate of AMD and Intel CPUs is?

Um...hopefully more than 10% respectfully...

 

[Blandge]

Um...hopefully more than 10% respectfully...

Haha well flipping a coin would get you about 50% 😛

 

[braincruser]

fliping a coin, while more efficient in prediction rates each takes supstantial amout of time, mainly in waiting for the coin to fall back and reading the data, considering a processor works around 3Ghz it would take 6-7 gigacoins flying in the air in one moment, as well as one huge chasis for all those coins to flip in.

 

[palladin9479]

Well, at the very least, moving RAM onto the chip would seriously eliminate the need for the L3, if not the L2 cache, depending on how low latencies can be driven. So thats one area where you can pick up the die space needed to put memory directly on the chip itself.

Your not moving it onto the chip, your moving a second chip ontop of the first. Both chips are made by difference process's and bonded together prior to being placed on the substrate. Their doing this because space and power are big issues in mobile applications, specifically phones and tablets. They remove the external memory bus and replace it with a local one that has interconnects on top of the CPU die. The DRAM chip is then laid directly onto the CPU die with the two points being bonded together to form a high speed data path. Thus you don't need to worry about external pins from the CPU being routed to a separate memory chip on the board along with termination and signal reflection. This reduces heat and power requirements while increasing bandwidth and lowering latency. It's still no where near what L2 or even L3 can achieve, but significantly faster then what an external memory bus would achieve. It will not replace L2, though could serve as L3/L4 for low power chips.

The two big drawbacks, something Caz is dancing around and refuse's to answer, is that your limited to ONE memory chip at most and to lower clock speeds due to thermal insulation. Both of these are acceptable in the mobile space as total thermal output is what is measured not per-device thermal output. Removing the external system memory bus reduces heat more then the insulating factor of one semi-conductor on top of another.

In the desktop world, which is the context of this entire discussion, these two issues become magnified. I seriously doubt anyone in this thread, nor anyone else for that matter, desires to run Windows 7 on 512MB or even 1GB. No matter how fast the local memory is, it won't be big enough to actually do the work you need the system to do. Right now 4GB has become the minimum for a low end system, and 8GB is common. By the time you can put 4GB of a memory on top of a CPU 32GB will be the minimum and 64GB will be common. Simply put, it can never replace main memory.

The thermal issue also raise's a problem. We cool devices individually and high performance chips are 70W+. Thermal insulation of the CPU results in lower cooling performance which in turn results in lower clock speeds and chip performance. We make our CPU cooling blocks out of copper and aluminum and not silicone for a reason. Take a CPU cooling block and put a thin layer of silicon epoxy, then put it on the CPU. See what happens when you turn on that computer and watch what your temps will look like. The only way to mitigate this in desktops is to horizontally stack the memory, something that hasn't even been started yet. You'd still be faced with the memory limitation but at least it wouldn't be slowing down your CPU.

 

[palladin9479]

Try to put words in my mouth all you want but it doesn't change the fact that the mainstream market is heading towards more compact and efficient systems. I never said the high end chips and discrete GPUs would go away. That's just asinine.

When you cut the max capacity in half it makes 2GB seem small. Now try 4GB and it doesn't sound so bad. To enthusiasts is does but go to Walmart and see how many netbook/laptop computers ship with just 1GB-2GB of memory.

Windows 8 (32-bit) requires 1GB memory
Windows 8 (64-bit) requires 2GB memory

How many millions of people are happy with their iPad that has only 512MB RAM?

Haven't put a single word in your mouth, you even repeated it in this thread.

Another couple years out the "main memory" will be stacked on or next to the CPU but in the same package (similar to how Core2Quad and Interlagos is made now) "

Actually you claimed stacked memory, you didn't even know about horizontal stacking yet. It hasn't appeared in any news releases and won't for another year at least.

That diagram still shows a memory chip vertically placed on top of a CPU.

Windows XP
Required: 64MB
Recommended: 128MB

So your going to tell me and everyone on this board that you'd be happy running XP with 64MB of memory? And that you'd be happy running Windows 7/8 with 2GB? Come on now your being watched.

And what applications will you be running on your 2GB Windows 7 PC? Office, Metro 2066, Dirt 4, BF4? (Various potential future games)

I want to hear, from you, what you actually expect to do with this 2GB desktop PC in a world that will be full of 16 ~ 32GB PC's. (2GB * 8 = 16GB per DIMM stick at approx $30~40 USD).

What "performance advantage" will you have when your PC is forced to trash the HDD to swap out the other 4GB of game data that won't fit into your 1.5GB of usable memory? We're not talking now, we're talking 2+ years from now.

 

[king smp]

didnt somebody say that you could just go side by side with the memory and cpu in one package
possibly use CPU HSF to cool mem also?

 

[palladin9479]

And BTW, 16Gb DRAM chips are at least two years away from being cheap. Samsung, the people pioneering this technology, are making 4Gb right now. That's 512MB per stack, not even enough to do something on Windows XP, much less Windows 7/8 x86. They are demoing 8Gb technology now and predict to have it available for server use in Q4 this year. That would make 1GB per stack or cheap 8GB sticks. 16Gb chips aren't even being made in the labs and aren't expected for at least two years, 32Gb is something that haunts the dreams of their engineers at night.

Now assuming 6months to 1yr development time from commercial production to when a CPU manufacturer can design, implement, fabricate and release it to market. We're looking at 512MB "on chip" memory next year, 1GB "on chip" memory two years from now, and 2GB "on chip" memory three to four years from now. 4GB "on chip" memory would be five or more years from now.

Thus your saying 512MB will be "enough" for next years high performance PCs, 1GB for high performance PC's in two years, 2GB for high performance PCs in three to four years and 4GB memory for high performance PCs in five or more years.

Please share with us what your smoking cause that sounds a whole lot like "640KB will be enough for anyone".

 

[palladin9479]

didnt somebody say that you could just go side by side with the memory and cpu in one package
possibly use CPU HSF to cool mem also?

I did actually. Their not even trying to engineer it right now because their focusing everything on the next Samsung mobile CPU and horizontal mounting wouldn't work there. I posted this info to demonstrate that Caz doesn't actually know about this and is just regurgitating what he heard from internet news articles.

It would solve the thermal insulation issue, requires a more precise heat plate, but that's not hard. It does nothing to solve the scaling issue in that your limited to the density of one stack.

 

[jimmysmitty]

Haven't put a single word in your mouth, you even repeated it in this thread.

Another couple years out the "main memory" will be stacked on or next to the CPU but in the same package (similar to how Core2Quad and Interlagos is made now) "

Actually you claimed stacked memory, you didn't even know about horizontal stacking yet. It hasn't appeared in any news releases and won't for another year at least.

That diagram still shows a memory chip vertically placed on top of a CPU.

Windows XP
Required: 64MB
Recommended: 128MB

So your going to tell me and everyone on this board that you'd be happy running XP with 64MB of memory? And that you'd be happy running Windows 7/8 with 2GB? Come on now your being watched.

And what applications will you be running on your 2GB Windows 7 PC? Office, Metro 2066, Dirt 4, BF4? (Various potential future games)

I want to hear, from you, what you actually expect to do with this 2GB desktop PC in a world that will be full of 16 ~ 32GB PC's. (2GB * 8 = 16GB per DIMM stick at approx $30~40 USD).

What "performance advantage" will you have when your PC is forced to trash the HDD to swap out the other 4GB of game data that won't fit into your 1.5GB of usable memory? We're not talking now, we're talking 2+ years from now.

Actually 7 runs fine with 2GB of RAM for most common users. Its only advanced users that need 4GB+, like us.

XP ran great on 128MB on release, but the SPs really pushed it to 1GB. I ran it on release with 192MB of RAM on a Pentium II 333MHz and it was fine.

Minimum is running, recommended is smooth, beyond recommended is butter.

 

[king smp]

"Minimum is running, recommended is smooth, beyond recommended is butter."

Can I use that quote JS?
I will give you credit LOL

 

[palladin9479]

Actually 7 runs fine with 2GB of RAM for most common users. Its only advanced users that need 4GB+, like us.

XP ran great on 128MB on release, but the SPs really pushed it to 1GB. I ran it on release with 192MB of RAM on a Pentium II 333MHz and it was fine.

Minimum is running, recommended is smooth, beyond recommended is butter.

Running what exactly? OS without applications is as useful as a car without wheels. You can look at it all day long, but it won't actually accomplish anything.

Also keep in mind the memory density available when Windows XP was published. 64MB was standard stick size, that's 8MB per chip or 64Mb fab technology. So this super CPU with local memory be running Windows XP with 8MB.