• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

AMD Details Jaguar; Preps Chip for PlayStation 4

T

televator

Member
amstradcpc said:
AMDs new Asynchronous Compute Engines serve as the command processors for compute operations on GCN. The principal purpose of ACEs will be to accept work and to dispatch it off to the CUs for processing. As GCN is designed to concurrently work on several tasks, there can be multiple ACEs on a GPU, with the ACEs deciding on resource allocation, context switching, and task priority. AMD has not established an immediate relationship between ACEs and the number of tasks that can be worked on concurrently, so were not sure whether theres a fixed 1:X relationship or whether its simply more efficient for the purposes of working on many tasks in parallel to have more ACEs.

One effect of having the ACEs is that GCN has a limited ability to execute tasks out of order. As we mentioned previously GCN is an in-order architecture, and the instruction stream on a wavefront cannot be reodered. However the ACEs can prioritize and reprioritize tasks, allowing tasks to be completed in a different order than theyre received. This allows GCN to free up the resources those tasks were using as early as possible rather than having the task consuming resources for an extended period of time in a nearly-finished state. This is not significantly different from how modern in-order CPUs (Atom, ARM A8, etc) handle multi-tasking.

from http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/5

So, allows to allocate instructions at your will, compute or rendering ones, and the order to be executed.Fine tuning this order manually you could avoid stalls in the alus and improve their efficiency.PS4'S CERNY SAUCE.
Click to expand...

I like this post. Great info! Thanks for posting.
 
B

Biggzy

Member
StevieP said:
In floating point operations, absolutely. In general computing no. Cell and Xenon were not very good in that regard. Deep pipeline and all that
Click to expand...

The CU will take care of the floating point operations deficit.

velociraptor said:
We've had developers developing games on the Cell for almost 7 years now...

It's quite sad that we are taking a backwards step with the CPU.
Click to expand...

Judging by the way Cerny was apologising for the Cell, I would make an educated guess that developers moaned to Sony about it.
 
A

aeolist

Banned
FyreWulff said:
Xbox 360 already had six hardware threads. (2x each core, unless you were silly and used XNA, then you only got access to four)

Problem is telemetry will pretty much show a majority of PCs out there are on 2 cores or less. Even the Steam HW survey can't break >50% with their 4-core userbase. So PC devs are content in having shitty multicore support.

I'd like to put my 8-core bulldozer to some real work
even though it's technically actually 8 half-cores. Fuck you AMD, my next computer is an i7
)
Click to expand...

360 cores were PPC with deep integer pipelines, no branch prediction, and heavy floating point resources. They were more often used to beef up the graphics since they weren't really good at the kind of logic that would run on a PC CPU.

PS4/Durango are standard x86-64 which will translate over to PC much more nicely, and lots of people have 8 threads available these days.
 
M

Manmademan

Member
Biggzy said:
Judging by the way Cerny was apologising for the Cell, I would make an educated guess that developers moaned to Sony about it.
Click to expand...

If the goal is also to get more indie devs on board and enable self publishing, the cell probably wouldn't have been the best option there, either.
 
K

KidBeta

Junior Member
aeolist said:
360 cores were PPC with deep integer pipelines, no branch prediction, and heavy floating point resources. They were more often used to beef up the graphics since they weren't really good at the kind of logic that would run on a PC CPU.

PS4/Durango are standard x86-64 which will translate over to PC much more nicely, and lots of people have 8 threads available these days.
Click to expand...

No branch predictor in PPC?.

Not likely.
 
I

i-Lo

Member
amstradcpc said:
Well just found this about the gpu:
I'm hearing that PS4 GCN has 8 ACE's, each capable of running 8 CL's each. I believe Tahiti is 2 per ACE, 2 ACE's.

http://forum.beyond3d.com/showpost.php?p=1712819&postcount=1188

This info,appears also in the sea islans gpu series isa document leaked last week, but refering to an "upcoming device".So gpu seems sea islands but with many more ACEs.This would debunk the 14+4 division as with so many aces and command ques you could divide the cus type of execution threads at your will between rendering/compute tasks.
Click to expand...

Interesting (went over my head).

amstradcpc said:
AMDs new Asynchronous Compute Engines serve as the command processors for compute operations on GCN. The principal purpose of ACEs will be to accept work and to dispatch it off to the CUs for processing. As GCN is designed to concurrently work on several tasks, there can be multiple ACEs on a GPU, with the ACEs deciding on resource allocation, context switching, and task priority. AMD has not established an immediate relationship between ACEs and the number of tasks that can be worked on concurrently, so were not sure whether theres a fixed 1:X relationship or whether its simply more efficient for the purposes of working on many tasks in parallel to have more ACEs.

One effect of having the ACEs is that GCN has a limited ability to execute tasks out of order. As we mentioned previously GCN is an in-order architecture, and the instruction stream on a wavefront cannot be reodered. However the ACEs can prioritize and reprioritize tasks, allowing tasks to be completed in a different order than theyre received. This allows GCN to free up the resources those tasks were using as early as possible rather than having the task consuming resources for an extended period of time in a nearly-finished state. This is not significantly different from how modern in-order CPUs (Atom, ARM A8, etc) handle multi-tasking.


from http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/5

So, allows to allocate instructions at your will, compute or rendering ones, and the order to be executed.Fine tuning this order manually you could avoid stalls in the alus and improve their efficiency.PS4'S CERNY SAUCE.
Click to expand...

*still over my head*. *Beginning to understand certain things*

JonathanPower said:
How can 8 jaguards be weak? Do you really think you could stand against 8 jaguars?

oncinha.jpg
Click to expand...

Soft kitty, warm kitty, little ball of fur.... oh hell naw.
 
A

aeolist

Banned
Wait no, Cell and Xenon have simple branch predictors but don't support out of order execution

My bad, but point still stands
 
Z

zoku88

Member
aeolist said:
Wait no, Cell and Xenon have simple branch predictors but don't support out of order execution

My bad, but point still stands
Click to expand...

Oh, ok. I was thinking some weird misunderstanding was going on.

I would be surprised if any chip from the last 20 years didn't have branch prediction.
 
G

goomba

Banned
Considering Intel has murdered AMD in the PC CPU market , I guess thats why they are now focused on low power, low cost game console CPU's
 
R

rudieboy77

Member
Anyone remember the rumor from a year ago about a PowerPC chip that was 8 core and 8 threads ? That would be interesting...
 
T

TOAO_Cyrus

Member
velociraptor said:
We've had developers developing games on the Cell for almost 7 years now...

It's quite sad that we are taking a backwards step with the CPU.
Click to expand...

Modern GPU's have great GPGPU capabilities so it makes sense to drop the CELL with its great floating point performance and poor general purpose performance in favor of a more balanced design.
 
D

DSN2K

Member
interesting but all 3 have taken the same approach to their hardware this time round just on different levels. these Console(720,PS4,WiiU) are built to stress the GPU/Ram while less on the CPU.
 
I

IHaveCandy

Banned
TOAO_Cyrus said:
Modern GPU's have great GPGPU capabilities so it makes sense to drop the CELL with its great floating point performance and poor general purpose performance in favor of a more balanced design.
Click to expand...

Yeah, wasn't there all this talk about a "physics processor" that would be added to PCs back around 2005-05? Video cards weren't up to snuff for a lot of physics tasks, and people thought we'd have to have a CPU, GPU, and PPU in PCs. Then GPUs started enhancing physics capabilities, and that is where we are today. The PS3 and Cell seem to be born from a time where GPUs couldn't muster the same kind of raw calculation power that they can today.
 
B

Biggzy

Member
DSN2K said:
interesting but all 3 have taken the same approach to their hardware this time round just on different levels. these Console(720,PS4,WiiU) are built to stress the GPU/Ram while less on the CPU.
Click to expand...

This is the direction the industry is taking.
 
Trickster

Trickster

Member
Clint Beastwood said:
I feel the same.

Unless devs take advantage of this stuff it really doesn't matter what the specs are. PS4 reveal shows this isn't going to happen. Everyone is jizzing over KZ and all it did was go from sub-HD to 1080p. Gamers have sent a message, pretty rehashes = STFU N TAKE MY MONIES
Click to expand...

No offense, but this post is bullshit.

let's ignore the fact that the new KZ games is not just going from sub hd to 1080p ( killzone 3 wasnt even sub hd... ).

You're making a sweeping generalization that anyone that's been reading gaf since the ps4 reveal would tell you is dead wrong. A lot of people are annoyed that the actual killzone SF gameplay, looks like what we saw in killzone 2 and 3.

It's also downright comical that your conclusion to the fact that a lot of people think killzone 3 looks amazing. Means that all gamers want is pretty rehashes of existing IP's.
 
J

JohnnySasaki86

Member
amstradcpc said:
AMDs new Asynchronous Compute Engines serve as the command processors for compute operations on GCN. The principal purpose of ACEs will be to accept work and to dispatch it off to the CUs for processing. As GCN is designed to concurrently work on several tasks, there can be multiple ACEs on a GPU, with the ACEs deciding on resource allocation, context switching, and task priority. AMD has not established an immediate relationship between ACEs and the number of tasks that can be worked on concurrently, so were not sure whether theres a fixed 1:X relationship or whether its simply more efficient for the purposes of working on many tasks in parallel to have more ACEs.

One effect of having the ACEs is that GCN has a limited ability to execute tasks out of order. As we mentioned previously GCN is an in-order architecture, and the instruction stream on a wavefront cannot be reodered. However the ACEs can prioritize and reprioritize tasks, allowing tasks to be completed in a different order than theyre received. This allows GCN to free up the resources those tasks were using as early as possible rather than having the task consuming resources for an extended period of time in a nearly-finished state. This is not significantly different from how modern in-order CPUs (Atom, ARM A8, etc) handle multi-tasking.

from http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute/5

So, allows to allocate instructions at your will, compute or rendering ones, and the order to be executed.Fine tuning this order manually you could avoid stalls in the alus and improve their efficiency.PS4'S CERNY SAUCE.
Click to expand...

So does PS4 GPU have more of these than even a 7870?
 
J

JohnnySasaki86

Member
StevieP said:
In floating point operations, absolutely. In general computing no. Cell and Xenon were not very good in that regard. Deep pipeline and all that
Click to expand...

So according to this chart it has 194,490 flops per core. But is that just mega or kilo flops? Is it basically 19glops per core? So x8= 152 flops?
 
M

MooseKing

Banned
Seems pretty garbage. Much worse than what Intel offers with Sandy/Ivy and potentially even their last generation of architecture.
 
D

Demon Ice

Banned
People assume the CPU is going to be weak because it's got a relatively low clock speed (What is it, 1.6 Ghz?)

If the architecture is efficient, and it seems to be, then the low clock speed won't be an issue.
 
N

nubbe

Member
What are CPU's used for in games nowadays?
AI? AI is useless since progress in FPS are scripted.
Why we still have CPU?
 
M

Myshkin

Member
velociraptor said:
We've had developers developing games on the Cell for almost 7 years now...

It's quite sad that we are taking a backwards step with the CPU.
Click to expand...

No, it's not sad at all. The CPU Is a specialist now, not wasting power on any FP capability. And that's OK, because it will always be paired with a big honkin' FP coprocessor. And they increase the scheduling capability in the GPGPU, in case there are a bunch of stray FP tasks. And with unified memory it could be as close to point-and-shoot programming as you could expect for those situations where you don't need to optimize.

Someone wanting to make the modern equivalent of Star Wars: Rebellion might want unlimited CPU juice and sob that it's not i7, but they still wouldn't want Cell.
 
R

RoboPlato

I'd be in the dick
I'm really curious as to what the final Jaguars for PS4 and Durango are going to wind up being. I'm sure neither of them will be stock. The CPU is the only thing Sony didn't release hard numbers on after the reveal. We know the GPU is 1.84 Tflops, 8gb of GDDR5 at 176Gb/s, but only that the CPU was 8 cores.

Jaguar should be fine for next gen systems. Not great, but fine. Especially since both systems have audio chips that take will take the whole audio rendering load off of the CPU, which will clear up a lot of power.
 
J

JohnnySasaki86

Member
K

KidBeta

Junior Member
aeolist said:
IBM has PowerPC designs with branch predictors

Cell and Xenos were not those designs
Click to expand...

The PPE in the Cell, and each of the 3 Cores in the Xenon had a branch predictor.

Just see here.

http://www.cis.upenn.edu/~milom/cis501-Fall08/lectures/12_xbox.pdf

Slide 13 / Page 4.

The branch predictor may have not been _GOOD_ but it was still there.

And for the PPE element in Cell, just see here.

http://publib.boulder.ibm.com/infoc...e/1.0/Overview/L1T1H1_10_CellArchitecture.pdf

Slide 21.

Slide 27 for the SPE branching, but IIRC its no where near as good / powerful / flexible.
 
You must log in or register to reply here.

Similar threads

AMD "Strix Halo" Zen 5 Mobile Processor Pictured
Rumor Analysis
39
3K
THE DUCK replied
AMD Zen 5 CPUs Reportedly Made On TSMC’s 3nm Process In Q2, Mass Production In Q3
2
82
6K
Gameplay Gods Bless replied
AMD RDNA 5 To Be A Completely New GPU Architecture From The Ground Up, RDNA 4 Mostly Fixes RDNA 3 Issues & Improves Ray Tracing
News Rumor Hardware 2 3
128
6K
SoloCamo replied
Another crypto-mining boom threatens CPU prices, with AMD’s Ryzen 7950X now sold out
Hardware
42
3K
MonkeyClaw36 replied
Was 8 CPU cores enough for 9th gen consoles? Strix Halo, the prefect console.
Opinion Analysis 2
62
3K
Tertullian replied
Top Bottom