Advanced Micro Devices has released a document that is supposed to assist software developers to optimize their programs for AMD’s microprocessors based on Bulldozer micro-architecture and forthcoming chips. Among other things, AMD describes peculiarities of Piledriver micro-architecture as well as chips on its base, including Trinity, Vishera, Terramar and Sepang. The document also mentions Steamroller and Excavator micro-architectures

AMD has a rather strong and clear roadmap for the period of the following three years with the aim to increase performance-per-watt of its high-performance cores by approximately 10% - 15% every year. In practice, this transforms into 33% - 52% speed boost of Excavator micro-architecture compared to Bulldozer. Technologically, AMD wants to improve IPC (instructions per clock) performance and reduce power consumption at the same time, which transforms into higher clock-speeds.

Among the revelations regarding Piledriver, noted by CPU-World web-site, are increased number of entries of L1 TLB, deeper FPU load queue, FMA3, BMI, TBM and other new instructions.
Interestingly, but the maximum amount of cores inside Vishera (next-gen FX-class desktop-chip), Sepang (1-die Opteron chip for 1-2 socket servers) and Terramar (2-die Opteron MCM for 2-4 socket servers) microprocessors is considered to be ten, not eight in case of Vishera as noted earlier. Moreover, it looks like the silicon behind all three of the products seems to be very same.
Another notable news is that the AMD family 15h models 2xh chips will physically have quad-channel memory controller, which lets creation of server platforms with extreme memory bandwidth (even eight-channel memory is theoretically possible in case of 4-socket machines). Unfortunately, since Vishera is supposed to be socket AM3+ compatible, it will not support quad-channel DDR3 memory.
AMD calls its Bulldozer microprocessors as AMD family 15h. The already available models have model numbers 00h - 0fh (0xh), Trinity [with Piledriver inside] are believed to carry model numbers 10h – 1fh (1xh), more advanced Piledriver generation of chips will have 20h - 2fh (2xh) model numbers. The document also mentions of 30h - 3fh, and 40h - 4fh processors, but it is unclear whether they belong to Steamroller and Excavator families, or just to Steamroller for clients and Steamroller for servers.
AMD did not comment on the news-story.
Tags: AMD, Bulldozer, Piledriver, Excavator, Steamroller, Trinity, Vishera, Terramar, Sepang
Comments currently:
21
Discussion started: 01/16/12 07:17:49 PM
Latest comment: 10/10/12 06:31:30 PM
Expand all threads |
Collapse all threads
[1-6]
1.
30h-3Fh will be Kaveri
40h-4Fh name will be announced at AMD Financial Analyst Day(February 2nd)
http://www.xbitlabs.com/i...09/amd_roadmap_mobile.jpg
30h-4Fh = Steamroller
50h-6Fh = Excavator
10h-1Fh = Trinity APUs
20h-2Fh = Viperfish CPUs
20h+ Models AGLUs include execution of these instructions:
BEXTR reg, reg, reg
MOV reg, reg
XADD reg, reg
XCHG reg32, reg32
XCHG reg64, reg64
4
0
[
Posted by: seronx

|
Date: 01/16/12 07:17:49 PM]
+ expand thread (13 answers)
- collapse thread
so, where's piledriver
0
0
[
Posted by: madooo12

|
Date: 01/17/12 08:37:38 AM]
June for Trinity and November for Viperfish
Unless they are going for a quick release with no paper launch
3
0
[
Posted by: seronx

|
Date: 01/17/12 11:44:48 AM]
I don't understand, so viperfish is the AMD High-end Piledriver, like the current FX-series
0
0
[
Posted by: madooo12

|
Date: 01/17/12 01:05:00 PM]
Yep, Trinity = Early Piledriver/ Viperfish = Late Piledriver
Late Piledriver(20h-2Fh) will be the natural extension for Bulldozer(00h-0Fh)
With Kaveri(30h-3Fh) having what (20h-2Fh) has and some of what (40h-4Fh) Streamroller will have
CPUs will most likely be even numbers
APUs will most likely be odd numbers
With 2015+ bringing the age of Full Fusion no CPUs
3
0
[
Posted by: seronx

|
Date: 01/17/12 03:21:15 PM]
I hate the full fusion no CPUs thing, it means the high end get nothing, are you sure about that
Piledriver looks promising but we'll have to see it in action once it comes out
BTW i'm waiting eagerly for the AMD Financial Analysts day (2/2) they should uncover their new plans then
0
0
[
Posted by: madooo12

|
Date: 01/17/12 04:25:32 PM]
Full Fusion will get a high-end version you get to overclock still and will get GPU extensions(RAID 5 or 0 setup with the GPUs instead of the current RAID 0+1 Setup)
Current GPUs use the same memory
Next Generation GPUs will use separate memory
1
0
[
Posted by: seronx

|
Date: 01/17/12 04:33:40 PM]
Are there any difference between "early" Piledriver and "late" Piledriver in terms of execution? There should not be.
"With Kaveri(30h-3Fh) having what (20h-2Fh) has and some of what (40h-4Fh) Streamroller will have"
Are we heading towards different types of x86 cores inside APUs and CPUs? K10.5+ "Husky" is different from Phenom, but it is clearly does not have Bulldozer features (except, maybe, when it comes to power consumption).
0
1
[
Posted by: Anton

|
Date: 01/17/12 08:03:25 PM]
Are there any difference between "early" Piledriver and "late" Piledriver in terms of execution? There should not be.
20h+ Models AGLUs include execution of these instructions:
BEXTR reg, reg, reg
MOV reg, reg
XADD reg, reg
XCHG reg32, reg32
XCHG reg64, reg64
^-- Late Piledriver, Early Steamroller, and newer etc.
Other than Bextr, Mov, Xadd, Xchg being allowed to be executed in the AGLU pipes there is no difference between Early Piledriver and Late Piledriver
But, Only Late Piledriver will be getting an IPC increase(4 64bit Movs per clock vs 2 64bit Movs per clock)
But mov is supposedly very important in x86 as it is the most used instruction:
http://www.strchr.com/media/top20_instructions_x86.png
If you mean execution as marketing then I would rather wait for February 2nd before I determine my actual answer
----
Are we heading towards different types of x86 cores inside APUs and CPUs? K10.5+ "Husky" is different from Phenom, but it is clearly does not have Bulldozer features (except, maybe, when it comes to power consumption).
No, all new APUs and CPUs will have Bulldozer Family Cores...Llano was just a stop gap and there is only one version that reached design goals for Llano actually which is the A8-3870 Black Edition
APUs:
"Early" Piledriver -> "Early" Steamroller -> "Early" Excavator
CPUs:
Bulldozer -> "Late" Piledriver -> "Late" Steamroller -> "Late" Excavator
A/C-PU:
Bulldozer -> Piledriver v1 + GPU -> Piledriver v2 + No GPU -> Steamroller v1 + GPU -> Steamroller v2 + No GPU -> etc.
3
0
[
Posted by: seronx

|
Date: 01/17/12 08:22:37 PM]
Thanks a lot!
Did you get that from the document linked, or from other sources?
0
1
[
Posted by: Anton

|
Date: 01/18/12 08:53:26 PM]
The new instructions from the SW guide
The names I somehow had to explain the difference Piledriver 10h-1Fh and Piledriver 20h-2Fh(10h-1Fh being June~ and 20h-2Fh being November~)
Llano is from ISSSC and AMD Slides of Llano which said Llano was meant to be 3.0+GHz
Then we have February 2nd
1
0
[
Posted by: seronx

|
Date: 01/18/12 10:11:22 PM]
Dont feed shillov. He hates AMD.
1
2
[
Posted by: bereft

|
Date: 01/26/12 05:43:11 AM]
it will again be late. trinity was supposed to be out by 1st quarter this year.
show the post
0
5
[
Posted by: dudde

|
Date: 01/17/12 10:16:09 PM]
nope nothing official came out,
rumors expected a Q3 launch
0
0
[
Posted by: madooo12

|
Date: 01/18/12 03:29:28 AM]
2.
As suspected by many, trinity does not have any L3 cache..probably worth noting in the article!
1
1
[
Posted by: turtle

|
Date: 01/16/12 10:33:03 PM]
3.
Unfortunately, since Vishera is supposed to be socket AM3+ compatible, it will not support quad-channel DDR3 memory.
Is this a fact or just a guess? There is nothing that stops an AM3+ system to just use one memory channel, so why wouldn't a system designed to be backward compatible with AM3+ be able to use more memory channels in a newer socket and motherboard that supports it?
1
0
[
Posted by: Snigel

|
Date: 01/17/12 12:28:24 AM]
+ expand thread (1 answer)
- collapse thread
AM3+ is a Dual Memory socket.
To give it Quad Memory driver capability (which it desperately needs) would require a new socket ie; AM4
0
0
[
Posted by: fdunn

|
Date: 10/10/12 06:31:30 PM]
4.
In practice, this transforms into 33% - 52% speed boost of Excavator micro-architecture compared to Bulldozer.
excavator is bulldozer, i think you mean zambezi
0
2
[
Posted by: madooo12

|
Date: 01/17/12 08:32:01 AM]
5.
10 cores, quad channel memory, double the L1 cache in Piledriver
Just Add USB 3.0 and PCIe 3.0 to that
If the IPC is 20%-30% better than zambezi then there would be no reason to not buy it
1
2
[
Posted by: madooo12

|
Date: 01/17/12 08:44:54 AM]
6.
so in 4 years time AMD will be at the same IPC and performance as what intel is today at the SAME price as what AMD is today! LOL
Good luck with that haha AMD should put bulldozer in the trash and work on a new design
show the post
2
6
[
Posted by: vid_ghost

|
Date: 01/17/12 09:34:58 PM]
+ expand thread (1 answer)
- collapse thread
How so? Do you have a design better than the Senior fellows at AMD? Do submit and you will for sure be given a job that pays no less than 150-200K per year USD. Guaranteed.
If you are not able to do that, you should not punch above your weight with pointless ignorant illiterate advice.
Basement arm chair commentary is a great pass-time, but don't try make it look like cutting edge electronics is the same as managing a burger king store where any random person/customer might have some sort of valid comment to make.
1
0
[
Posted by: vanakkuty

|
Date: 06/26/12 03:35:42 AM]
[1-6]