AMD Zambezi news, info, fans !

[Smartidiot89]

much info here also:

http://translate.google.com/translat...%2Findex6.html

You say in that thread that AMD officially have stated that one module is the same as one core that can run two threads simultaneously. Who's the joker at AMD who says such things?

[Formula350]

The specification really throws me off here, none of my AMD ES chips have anything which remotely resembles that. I look forward to seeing TRUE ES pictures

So what are the chances we could get a partial CPU-Z shot? One that only shows a bit of info, such as the biggest conflicting bits compared to the fake? (lol Why is that emote called 'yawn'??)

True, but lets assume current beta BIOSes are not allowing OC.

True, but one must remember that the chips are running on current motherbooards. These are the C32, G34 and AM3/+ boards we all can buy right now, so crippling the BIOS would be weird. Not saying it can't be done, would just mean beta BIOS files made and sent to specific people, but that would also keep from testing the BIOS in a release-like environment. They could have it running perfectly stable, and then for some reason re-enabling the tweaking options 'breaks' things :\ Plus, it wouldn't be a stretch for someone to just slip in the AGEIA CPU info to a release BIOS to gain the CPU support Either way, my friend's Super Micro lets him OC his.

I was talking about this section of CPU name:

ZD322046W8K44_36/28/20_2/8 A

This is my speculation, but to me it reads 3.6GHz Turbo. 2.8GHz Stock. I'm not sure how to interpret 20_2 and 8 A.
Once again my speculation

I admit, that was EXACTLY something I was looking at too, but for whatever reason I had disregarded it! So I'm onboard with your suggestion The only thing I figure 8 A, or the 8, is core count. A maybe being Single "chip", where perhaps B would be MCM dual 'chip': 8core + 8core for 16core Interlagos. Could also be A B C, for AM3+. C32 or G34 platform designation
This might be just a total coincidence... BUT:
20-2=18 -- 18x Multiplier: 18x200=3600MHz OR the "36" boost speed.
20-8=12, 12+2=14 -- 14x Multiplier: 14x200=2800MHz OR the "28" default speed.
lol Again, total guess and I don't see it holding water since it required the addition steps to get 2800MHz :P

(20/2)+8 also is the boosted 3.6GHz, but that leaves me with nothing on how to get the 2800....

[sarcasm]
IIIII'VE GOT IT! hahah OK stick with me here!
The APUs run on 100MHz reference clock, so:
20+8=28. x100 = 2800!
((20/2)+8)x2=36. x100 = 3600! lol
[/sarcasm]

I want one of these boards so badly!

Quad-G34, Quad Channel DDR3-32DIMM . . . . . . . Dual-G34, Quad Channel DDR3-16DIMM
2x PCIe x16@16! POWARRR!!! . . . . . . . . . . . . . . 3x PCIe x16@16! :drool:


Dual-C32, Dual Channel DDR3-8DIMM
QUAD PCIe x16@16

[Formula350]

You say in that thread that AMD officially have stated that one module is the same as one core that can run two threads simultaneously. Who's the joker at AMD who says such things?

Yea someone I think misunderstood him lol 1 Module = 2 cores, capable of 4 threads (2 per core, similar to HyperThreading, but not quite).
(I'm agreeing with you BTW, Smartidiot, and expounding on what you said)

[[XC] Oj101]

What sort of specifications are you referring to? (if you're allowed to say)

Of course I can, I'm not under NDA

The specification line in CPU-Z which reads "AMD Eng Sample, ZD322046W8K44_36/28/20_2/8 A" doesn't line up with ANY other ES I've seen.

Have a look at this though:



Swiped from the Bulldozers first screens thread

[Apokalipse]

Is it total die size that's 12% larger than a hypothetical 4 (standalone) core BD chip?
Or is it just the module itself that's 12% larger than a single hypothetical standalone BD core (not counting all other hardware, like IMC, cache, etc..)?

[BeepBeep2]

Congrats guys

You just speculated for a whole day over a fake photoshop I made in about four minutes.

[demonkevy666]

^ lol run the bann hammer will hunt you down XD

[Formula350]

Congrats guys

You just speculated for a whole day over a fake photoshop I made in about four minutes.

Um... We know dude lol At least, I knew that's what was going on I was speculating as to what it exactly was you changed, and you still haven't said what is is either


Oj101 beat me to posting the screenie too lol
My guess of cache multiples was actually right, though :O lol

[Olivon]

Congrats guys

You just speculated for a whole day over a fake photoshop I made in about four minutes.

You mean you are 2good4you ?

[informal]

No,I think BeepBeep was referring to faked picture in this post.

[FlanK3r]

Informal:
but what about this photo? Look at some infos...L1 data cache, diferent L2 (2048KB from Chinas vs 2MB here), black place in CPU-Z version looks not as example 1.57.x, but only 1.57 or older, too small for 4 chars.

[Formula350]

Informal:
but what about this photo? Look at some infos...L1 data cache, diferent L2 (2048KB from Chinas vs 2MB here), black place in CPU-Z version looks not as example 1.57.x, but only 1.57 or older, too small for 4 chars.

All looks correct to me? L1 is 16kb per core (data), L2 is 64 per module (instruction), L2 is 2MB per module (L3 is per chip and varies dependently)) So 8 core = 8x L1 Data, 4x L1 Instr., 4x L2

Chinas shows what it does due to only reading one core, so everything is not a multiple and thought of a singular. This depicts 8 cores and so it references everything as such. EDIT: That is why it shows 2048KB instead of 2MB, it would run out of room trying to display "4 x 2048KB"

Yea, it's a 2-digit version number. So it'd have to be 1.XX, likely being either 1.57 or maybe a newer 1.58 (internal?). If it's an older picture, it might be an AMD release and so a special version? *shrug*

[informal]

Informal:
but what about this photo? Look at some infos...L1 data cache, diferent L2 (2048KB from Chinas vs 2MB here), black place in CPU-Z version looks not as example 1.57.x, but only 1.57 or older, too small for 4 chars.

Hey FlanK3r mate ,just saw your post. The only weird thing in that screenshot is the L1 data cache associativity,it should be 4way not 2way.Everything else "seems" to be correct as far as I can tell.

[FlanK3r]

yes, exactly... 4-way. And strange is "CPU-Z version 1.5x" 1.58 doesn relase and 1.57 show no correct number of cores. 1.57.x it can not be, because black area in CPU-Z is too small for example 1.57.3

[2good4you]

from the cpu-z shot i posted @ sweclockers, I do not now much about it and comes from imageshack.

The official information about how we should see the core/module count is still the way I wrote. It is in many ways strange and in many ways not.

It should kick ass anyways because most apps is still singelthread or no more than dual/quad threaded.

I don't like to see a 4 module Bulldozer compared to an 8 core Intel. It's absolutely not a good comparison. AMD may have a 6-core Bulldozer ready, or an MCM for desktop like Interlagos as an anser.

Also like I wrote, one module is in almost all ways more like one core, than two. It shares _all_ except L1 datacache which is quaite impressive. It's very like Intel's Hyperthreading, but in a better way.

[Formula350]

from the cpu-z shot i posted @ sweclockers, I do not now much about it and comes from imageshack.

Then you need to take it with a grain of salt, and be skeptical of it's legitamacy, as it could be faked. All it takes to fake that is pshopping the screen shot, putting it as the desktop and hiding icons/task bar (or just end task on Explorer). From there, take a pic with the camera and submit to ImageShack :P

The official information about how we should see the core/module count is still the way I wrote. It is in many ways strange and in many ways not.

It should kick ass anyways because most apps is still singelthread or no more than dual/quad threaded.

I don't like to see a 4 module Bulldozer compared to an 8 core Intel. It's absolutely not a good comparison. AMD may have a 6-core Bulldozer ready, or an MCM for desktop like Interlagos as an anser.

My reading seems to have led us to believe they will show up as two threads per core, but that would mean an 8 core = 16 threads. I don't see AMD going about it that way, so I might be understanding it slightly wrong, and it is 1 core = 1 thread, but each core processes as if it was two threads due to the split architecture. Still, I'm 98% sure ONE module = TWO physical cores, which is why an E350 Bobcat is dual core and only uses one module

I don't like the comparison of AMD to Intel either, but we've just got to look at it on the scale of performance. If it turns out an 8-threaded Intel performs as well as an 8core AMD, then what are we left to say? :\

Your best bet would be Valencia for desktop, as it'd be cheaper since you'd only be going with a C32 board instead of G34. It'd only be a max of 8 cores still, but hopefully the memory performance and HyperTransport can make up for that a bit If you've got the cash though, I suppose you could liken the 12/16core Interlagos like the Extreme i7 chips, and G34 to the x58 since the costs will be close.

Also like I wrote, one module is in almost all ways more like one core, than two. It shares _all_ except L1 datacache which is quaite impressive. It's very like Intel's Hyperthreading, but in a better way.

Again, I'm pretty sure it's 2 physical cores, sharing only the 64KB of L1 Instruction and 2MB of L2 cache per module. The L3 is shared amongst the whole CPU, same as it has been so far.


I'm very open to being correct though if I'm misunderstanding anything! I only want to spread the facts, not misconceptions (be that my own or read from others insisting they are facts).

[BeepBeep2]

Then you need to take it with a grain of salt, and be skeptical of it's legitamacy, as it could be faked. All it takes to fake that is pshopping the screen shot, putting it as the desktop and hiding icons/task bar (or just end task on Explorer). From there, take a pic with the camera and submit to ImageShack :P




My reading seems to have led us to believe they will show up as two threads per core, but that would mean an 8 core = 16 threads. I don't see AMD going about it that way, so I might be understanding it slightly wrong, and it is 1 core = 1 thread, but each core processes as if it was two threads due to the split architecture. Still, I'm 98% sure ONE module = TWO physical cores, which is why an E350 Bobcat is dual core and only uses one module

I don't like the comparison of AMD to Intel either, but we've just got to look at it on the scale of performance. If it turns out an 8-threaded Intel performs as well as an 8core AMD, then what are we left to say? :\

Your best bet would be Valencia for desktop, as it'd be cheaper since you'd only be going with a C32 board instead of G34. It'd only be a max of 8 cores still, but hopefully the memory performance and HyperTransport can make up for that a bit If you've got the cash though, I suppose you could liken the 12/16core Interlagos like the Extreme i7 chips, and G34 to the x58 since the costs will be close.



Again, I'm pretty sure it's 2 physical cores, sharing only the 64KB of L1 Instruction and 2MB of L2 cache per module. The L3 is shared amongst the whole CPU, same as it has been so far.


I'm very open to being correct though if I'm misunderstanding anything! I only want to spread the facts, not misconceptions (be that my own or read from others insisting they are facts).

Woah, woah woah there are some wrong things in this post....(To my knowledge!)
E350 is Fusion yes...but it doesn't have this module thing going on. Core structure is closer to what we have seen in the past.
ONE Module in Bulldozer does include TWO (Physical, if you can call it that and) LOGICAL cores, yes...AMD has already stated, the 8 core Desktop model will include four "modules"

Here is some more (fudge) to speculate on if one wishes to XD

[2good4you]

Then you need to take it with a grain of salt, and be skeptical of it's legitamacy, as it could be faked.
All it takes to fake that is pshopping the screen shot, putting it as the desktop and hiding icons/task bar (or just end task on Explorer).
From there, take a pic with the camera and submit to ImageShack :P

Of course it can be, or it is some kind of shot saying very little. The info plus some details seems/can be correct.
Like aero, and aero effect at the pointer and closingicon is lightening etc.

My reading seems to have led us to believe they will show up as two threads per core, but that would mean an 8 core = 16 threads.
I don't see AMD going about it that way, so I might be understanding it slightly wrong, and it is 1 core = 1 thread, but each core processes as if it was two threads due to the split architecture.
Still, I'm 98% sure ONE module = TWO physical cores, which is why an E350 Bobcat is dual core and only uses one module

I did not wrote that and has never done.
2 integer execution-cores per module=2 threads per module, and the floating point unit can execute 4x256-bit instructions in one cycle or 8x128-bit instructions in one cycle.
I have wrote AMD's latest official statement, and also my own opinion of the design, which cover a lot more than the execution-logic.

If it turns out an 8-threaded Intel performs as well as an 8core AMD, then what are we left to say? :\

It will not.

Your best bet would be Valencia for desktop, as it'd be cheaper since you'd only be going with a C32 board instead of G34.
It'd only be a max of 8 cores still, but hopefully the memory performance and HyperTransport can make up for that a bit
If you've got the cash though, I suppose you could liken the 12/16core Interlagos like the Extreme i7 chips, and G34 to the x58 since the costs will be close.

It's still up to AMD to launch a primary server part cpu for desktop use.

Again, I'm pretty sure it's 2 physical cores, sharing only the 64KB of L1 Instruction and 2MB of L2 cache per module.
The L3 is shared amongst the whole CPU, same as it has been so far.

And still as I have stated, they share _all_ units in the module except the L1 datacache. The fetch, branchpredictor and all the schedulers are all shared.
Some details are for sure still unclear. Bulldozer will (must) have a extremely powerful front end to decode and schedule for two integer execution cores.
It also have dedicated scheduler for floating point operations which seems good. Bulldozer will because of this statement shine in singlethreaded applications.

[Formula350]

Woah, woah woah there are some wrong things in this post....(To my knowledge!)
E350 is Fusion yes...but it doesn't have this module thing going on. Core structure is closer to what we have seen in the past.
ONE Module in Bulldozer does include TWO (Physical, if you can call it that and) LOGICAL cores, yes...AMD has already stated, the 8 core Desktop model will include four "modules"

That's what I get then for having gathered most of info back in 2010 when they were calling them modules still I do admit to having created some self-confusion between Bulldozer and Bobcat modules. I had thought that a module of either was 2 cores, but Bobcat is only a single, which makes sense given the single core APUs. And yea, my terminology of "physical" is indeed not quite correct, which Logical should be actually used. My only reason for using it would be to more easily differentiate between a multi-threaded single-core (HyperThreading) and single threaded core-multi CPU.

I was also under the impression current Bobcats were K10.5, but now I'm seeing it be referred to as K14. CPU-Z is in line with that, though strangely Windows7 is not, showing actually a family number of 20 for the E350 >_> lol That might be because of the Stars and in that time it's gotten corrected *shrug* It's hard when the manufacture isn't giving any info heh Not that I don't blame them though, just sucks for us lol

Of course it can be, or it is some kind of shot saying very little. The info plus some details seems/can be correct.
Like aero, and aero effect at the pointer and closingicon is lightening etc.

I'll shop you up a good one here on what I mean that it's easy to do.

I did not wrote that and has never done.

Didn't mean for it to sound that way, just meant that is what I've been reading the past few months on the net.

The fetch, branchpredictor and all the schedulers are all shared.
Some details are for sure still unclear. Bulldozer will (must) have a extremely powerful front end to decode and schedule for two integer execution cores.
It also have dedicated scheduler for floating point operations which seems good. Bulldozer will because of this statement shine in singlethreaded applications.

Good info to know, thanks

[informal]

Orochi definitely has 8 cores,it's just the way they are organized(modules) that is bringing confusion.You have 8 very real integer cores and 8 very real 128bit fmac units.Those 8 fmacs are in reality 4 double fmacs(256bit) that are called Flex FP by AMD.One 256bit FlexFP can be allocated to one thread only(AVX or single threaded code),or it can be split into 2 and each dedicated to it's "parent" integer core(2 fp threads running in parallel in module).

Shared parts of the design are saving die space and increasing efficiency.This is why AMD can cram 8 cores with 16MB of cache(total size) in less than 300mm^2.

[Formula350]

Orochi definitely has 8 cores,it's just the way they are organized(modules) that is bringing confusion.You have 8 very real integer cores and 8 very real 128bit fmac units.Those 8 fmacs are in reality 4 double fmacs(256bit) that are called Flex FP by AMD.One 256bit FlexFP can be allocated to one thread only(AVX or single threaded code),or it can be split into 2 and each dedicated to it's "parent" integer core(2 fp threads running in parallel in module).

Shared parts of the design are saving die space and increasing efficiency.This is why AMD can cram 8 cores with 16MB of cache(total size) in less than 300mm^2.

Pretty nifty stuff... Is it me or, maybe with the exception of HyperThreading, Intel not do much interesting designing stuff like that, instead concentrating on making stuff that "has more torque" (if you will)?


So I was digging into making that photoshopped CPU-Z image to show 2good, when I noticed something about the ORIGINAL... Wouldn't the "Package" line be blue like the rest of the labeled of filled boxes? Yet, it shows Gray, as if it was originally empty...

EDIT: While not picture since it's on a different tab, do we know if the CPU-NB on the AM3+ Dozers is going to be 2600MHz? I'd assume HT-Link as well, as it has mirrored the NB clock in the past, but could stay at 2000MHz I guess...

[BeepBeep2]

when I noticed something about the ORIGINAL... Wouldn't the "Package" line be blue like the rest of the labeled of filled boxes? Yet, it shows Gray, as if it was originally empty...

EDIT: While not picture since it's on a different tab, do we know if the CPU-NB on the AM3+ Dozers is going to be 2600MHz? I'd assume HT-Link as well, as it has mirrored the NB clock in the past, but could stay at 2000MHz I guess...

That image is fake, the word would not be gray if the line is filled.
Its a really nice photoshop, but then again I could change the frequency to 3400 on my last photoshopped image and then proclaim its true.

[Smartidiot89]

from the cpu-z shot i posted @ sweclockers, I do not now much about it and comes from imageshack.

The official information about how we should see the core/module count is still the way I wrote. It is in many ways strange and in many ways not.

It should kick ass anyways because most apps is still singelthread or no more than dual/quad threaded.

I don't like to see a 4 module Bulldozer compared to an 8 core Intel. It's absolutely not a good comparison. AMD may have a 6-core Bulldozer ready, or an MCM for desktop like Interlagos as an anser.

Also like I wrote, one module is in almost all ways more like one core, than two. It shares _all_ except L1 datacache which is quaite impressive. It's very like Intel's Hyperthreading, but in a better way.

You might think of it that way, but facts are AMD now and always have counted 1 module = 2 cores and they will in the future. The mentality has always been this way from AMD and you stated in that Sweclockers thread specifically that AMD officially counts 1 module as 1 core which is wrong. The top die on Bulldozer gen.1 will have 4 modules, 8 cores, 8 threads.

But I agree on what you have said that comparing an 8 core AMD vs. 8 core Intel isn't "fair" not cause AMD took the module route. I never compare products like that, I compare it they way you should which provides the best bang for my buck?

[[XC] Oj101]

Informal:
but what about this photo? Look at some infos...L1 data cache, diferent L2 (2048KB from Chinas vs 2MB here), black place in CPU-Z version looks not as example 1.57.x, but only 1.57 or older, too small for 4 chars.

The info displayed is about in line with what I'd expect from 1.57

[FlanK3r]

mabye only one reason, could not be fake is: it was CPU-Z 1.56 or 1.57 and sample was in AM3 board (not AM3+). But from my side, I thinking, its fake . We can only waiitng for relase date