• 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.

PS4 Rumors , APU code named 'Liverpool' Radeon HD 7970 GPU Steamroller CPU 16GB Flash

Status
Not open for further replies.

RoboPlato

I'd be in the dick
FINALLY! Thought this would never end.


Now back to the topic at hand. I've been doing some research on the piledriver APU which the steamroller APU is based on. The performance doesn't seem so hot. ~30fps for Crysis 2 for ex. A lot of the articles I've read said it isn't well suited for gaming. I understand as far as power consumption its making some pretty good strides, but I dont see how this would be a good suite for PS4. I dont know if I was reading it right but the gpu's in the APU were around the 560-700 gflop range, so a 1.84 tflop gpu would be quite a lot more, but than wouldn't PS4 pretty CPU bound? These performance analysis were for the netbook version of the piledriver APU.
I may be mistaken but I think that I read that steamroller is going to be tweaked to be more suitable to gaming and similar processes than pile driver is. Sony may also be doing some customization to improve that even further. There's also the chance that Sony is really banking on the benefits of a GPGPU setup to balance out the CPU but that seems like a bit of a gamble. Sony seems to understand the importance of a strong CPU so I'm not too worried in that department.
 
Tell me:

Can SPE 3 perform a read/write operation over SPE1 local store?

Oh, and show me your Cell manufactured at 2005. Ebay will love it.

Cell was being manufactured in 2005. PS3 was supposed to launch in late 05, than it was delayed tell Sping 06, and than delayed again tell late 2006. Although I believe one of the problems was Cells yields among other things, but Cell was completed way before the end of 05.
 
Bad management. They wasted too much money buying ATI. They had to cut CPU research and sell some branches. Then they failed at planning level with the future of x86 and Bulldozer.

That, and Intel stopped screwing around and got their act together with Core 2. (Core was actually the 32-bit precursor.)

AMD has been pretty okay in the value/lower-mid tier department though. A Fusion APU should be decent enough for PS4.
 
Intel showed months ago a yield full of Haswell chips.

Release date stills at Q1 2013.

Intel-Haswell-Mobile-Chipsets-Revealed.jpg


Cell wasn't released at 2005. It's amazing how you try to rewrite history.
 

drkohler

Banned
Cell wasn't released at 2005. It's amazing how you try to rewrite history.
You really are very resistent to logic when people try to point you to your faulty logic - you simply make up another straw man argument by redefining the word "release".
Why would "the Cell be released in 2005", when Sony wasn't ready with the PS3? Who would have bought a Cell in 2005 for what purpose? That chip was designed for the PS3 and, of course, was in production in 2005. (It must have, of course. Where were all the dev units come from for the people who were using them in late 2004/2005?)
 

Argyle

Member
Stop to read at "DMA transfer".

Whats so hard to understand? Core 1 can write at L2 pos X. Core 0 can read/write at the very same X position. Cell can't.

It's easier, it's faster.

You asked

Can SPE 3 perform a read/write operation over SPE1 local store?

The question was answered, even if you didn't want to read the answer.

How much code have you written for the Cell?
 
Dunno. Some guy at internet sayz Cell was available before Conroe just pop out magically from nowhere.

You asked

The question was answered, even if you didn't want to read the answer.

How much code have you written for the Cell?

And the answer is no. You have to transfer the data.

And I didn't code anything since M68K days =)
 
Dear god man, just shut it. You single handedly derailed this freaking thread. Make a new thread about how cell sucks dick if you want but keep it out of here.
 

Argyle

Member
And the answer is no. You have to transfer the data.

And I didn't code anything since M68K days =)

If you want the data to go from one region of memory to another, isn't that the very definition of "transferring the data?"

Also, in other words, you have no idea what you are talking about? OK, then :)
 
If you want the data to go from one region of memory to another, isn't that the very definition of "transferring the data?"

Also, in other words, you have no idea what you are talking about? OK, then :)

Not really he is writing to shared memory l2 cache if im not mistaken.
So core 1 writes to point X when its done core 0 can read at point x.
Without having to copy the data from one cache to the other like i believe the DMA are doing for the SPE.

uugh im really the person to answer this hardware level stuff is kept at the minimum right now in CS classes.
 

Argyle

Member
Not really he is writing to shared memory l2 cache if im not mistaken.
So core 1 writes to point X when its done core 0 can read at point x.
Without having to copy the data from one cache to the other like i believe the DMA are doing for the SPE.

uugh im really the person to answer this hardware level stuff is kept at the minimum right now in CS classes.

He asked if SPE3 could read or write to SPE1's local store. Local store is not a traditional CPU cache, so the two local stores are different regions of memory. I didn't read all his nonsense on the previous page, maybe he was arguing about the L2 cache on the PPE or something.

Cell is cache coherent so if you DMA the memory out of local store back to main memory, the copy in L2 cache will be invalidated if the PPE tries to access that same memory.

https://en.wikipedia.org/wiki/Direct_memory_access#Cell
 

KageMaru

Member
So I guess WiiU can be considered to be at launched state since there is DevKits already.

I'm not here to debate launch dates of actual systems or what CPU was out before what other CPU, I'm just saying the CPU was being manufactured in 2005. I never said it was in mass production, I never said it was available to the public, and none of that was ever my point.
 
All we need is a new leak on specs for this thread to get back on track.

German Gameswelt has "learned from realiable sources" that PS4 will be released in Q1 2014.

Translated article here.

Why Google translates "Gameswelt" to "Gamespot" is beyond me, I don't think they are affiliated.

As for this. It may be possible. I've been predicting Q1 2014 release for PS4 in Japan for a while now. Actually thought it would release around March in Japan which would still be Q4 2013. Going by history it wouldn't release in NA tell at least 3 months after Japan.
 

double jump

you haven't lived until a random little kid ask you "how do you make love".
wish I understood half of this stuff. when are we gonna get a leaked cell phone pic of a prototype? ugh.
 
He asked if SPE3 could read or write to SPE1's local store. Local store is not a traditional CPU cache, so the two local stores are different regions of memory. I didn't read all his nonsense on the previous page, maybe he was arguing about the L2 cache on the PPE or something.

Cell is cache coherent so if you DMA the memory out of local store back to main memory, the copy in L2 cache will be invalidated if the PPE tries to access that same memory.

https://en.wikipedia.org/wiki/Direct_memory_access#Cell

I should give hardware soms more reading time. Im busy with that, halfway through Code from Charles Petzold. Any recommedaations of books or resources i could order or read.
 

AmFreak

Member
wish I understood half of this stuff. when are we gonna get a leaked cell phone pic of a prototype? ugh.

Da fuck?
Cell phone - rly?
Didn't you learn anything from the doc the last 10 pages?!?
Cell is crap, it has no integers, what you want is Xenon phone - it's like 3 x the Cell one launched 1983 (3 months before Cell phone) and it has cache!
 

Argyle

Member
I should give hardware soms more reading time. Im busy with that, halfway through Code from Charles Petzold. Any recommedaations of books or resources i could order or read.

Honestly, I don't know what is good these days. This probably dates me, but the book that comes to mind that I read personally was this one:

http://www.amazon.com/dp/156592312X/?tag=neogaf0e-20

It's going to be quite dated although it might still be useful.
 
He asked if SPE3 could read or write to SPE1's local store. Local store is not a traditional CPU cache, so the two local stores are different regions of memory. I didn't read all his nonsense on the previous page, maybe he was arguing about the L2 cache on the PPE or something.

Cell is cache coherent so if you DMA the memory out of local store back to main memory, the copy in L2 cache will be invalidated if the PPE tries to access that same memory.

https://en.wikipedia.org/wiki/Direct_memory_access#Cell

I was trying to argue why Cell in any ot it's form is unsuitable for a next gen PS4, then a wild horde of Cellvangelist appeared.

And they call me fanboy because they are more, so they can deny the most basic stuff:

As an example usage of DMA in a multiprocessor-system-on-chip, IBM/Sony/Toshiba's Cell processor incorporates a DMA engine for each of its 9 processing elements including one Power processor element (PPE) and eight synergistic processor elements (SPEs). Since the SPE's load/store instructions can read/write only its own local memory, an SPE entirely depends on DMAs to transfer data to and from the main memory and local memories of other SPEs. Thus the DMA acts as a primary means of data transfer among cores inside this CPU (in contrast to cache-coherent CMP architectures such as Intel's coming general-purpose GPU, Larrabee).
DMA in Cell is fully cache coherent (note however local stores of SPEs operated upon by DMA do not act as globally coherent cache in the standard sense). In both read ("get") and write ("put"), a DMA command can transfer either a single block area of size up to 16KB, or a list of 2 to 2048 such blocks. The DMA command is issued by specifying a pair of a local address and a remote address: for example when a SPE program issues a put DMA command, it specifies an address of its own local memory as the source and a virtual memory address (pointing to either the main memory or the local memory of another SPE) as the target, together with a block size.
 

KageMaru

Member
I was trying to argue why Cell in any ot it's form is unsuitable for a next gen PS4, then a wild horde of Cellvangelist appeared.

And they call me fanboy because they are more, so they can deny the most basic stuff:

I honestly think posts and points were not communicated properly and that is in part your fault. Both sides made correct points, but I'm not sure you both were talking about the same things specifically. Either that or you were trolling and when proven wrong, you moved the goal post as if your original point did not come through clear enough.

I'm the furthest thing from a Cellvangelist, I believe it saved the PS3 and without it the system could not keep up with the competition. However at the same time I've never drank the cool-aid that some other members here have and think it's something of extreme power that only Sony developers can properly yield.

That said I agree it isn't the best choice for the PS4, and you really could have made that point without these debates.
 

double jump

you haven't lived until a random little kid ask you "how do you make love".
Da fuck?
Cell phone - rly?
Didn't you learn anything from the doc the last 10 pages?!?
Cell is crap, it has no integers, what you want is Xenon phone - it's like 3 x the Cell one launched 1983 (3 months before Cell phone) and it has cache!

this legit made me laugh.
 
Da fuck?
Cell phone - rly?
Didn't you learn anything from the doc the last 10 pages?!?
Cell is crap, it has no integers, what you want is Xenon phone - it's like 3 x the Cell one launched 1983 (3 months before Cell phone) and it has cache!

Well played if this forum had a karma system I would +1 you.
 
I can't think about Cell as PS3 saviour because it have the fault of system weakness. More cheap conventional CPU could have allowed a stronger GPU and/or more RAM at any given budget. There wouldn't be any problem to start off.

I have to admit I trolled a bit here and there, but not because anyone proved me wrong. Just because of that "OMG CELL >>> Xbox" argument. And far from admit it, they want another Cell based PS4 instead of realize Cell BE is at legacy stage.
 

i-Lo

Member
If cell was indeed as unsuitable for console gaming as some here make out to be then why was it in PS3? Wouldn't the plethora of devs cried foul when the specs were first disseminated? Has PS3 been a vanity project made to hoodwink consumers when IBM knowing full well of the allegedly cheaper and "more" efficient architecture/processors out there?

Oh well, it's irrelevant for PS4 anyway since so far there has been no indication of a CELL2 or its derivative for PS4. If Sony are to continue as they did with Vita AMD's APU sounds like a more dev friendly alternative.
 

coldfoot

Banned
Yep. Super unfortunate. I've learned a lot about tech in this thread and really enjoyed it before.
Ignore option is your friend :) People I put on ignore tend to get banned within a couple months for whatever reason.

Still can't believe there aren't any more leaks yet.
 

KageMaru

Member
I can't think about Cell as PS3 saviour because it have the fault of system weakness. More cheap conventional CPU could have allowed a stronger GPU and/or more RAM at any given budget. There wouldn't be any problem to start off.

I have to admit I trolled a bit here and there, but not because anyone proved me wrong. Just because of that "OMG CELL >>> Xbox" argument. And far from admit it, they want another Cell based PS4 instead of realize Cell BE is at legacy stage.

Granted I've made the same point myself in the past but when I say Cell saved the PS3, I mean with it's current configuration. IIRC Sony and Nvidia signed the deal in 2004, but we don't know what options Sony had or the resources Nvidia was willing to invest. The fact that it's based on the 7800 but cut down to almost 7600 level indicates that time and resources were not a luxury the two companies enjoyed.

Besides, given how ahead of the time Xenos was compared to other GPUs, Sony likely made the right choice in the end. It's debatable if Nvidia could have provided a similar GPU for Sony even if they did invest more towards the GPU instead of the CPU. More memory was out of the question due to the number of chips necessary back then IMO.
 

10k

Banned
If cell was indeed as unsuitable for console gaming as some here make out to be then why was it in PS3? Wouldn't the plethora of devs cried foul when the specs were first disseminated? Has PS3 been a vanity project made to hoodwink consumers when IBM knowing full well of the allegedly cheaper and "more" efficient architecture/processors out there?

Oh well, it's irrelevant for PS4 anyway since so far there has been no indication of a CELL2 or its derivative for PS4. If Sony are to continue as they did with Vita AMD's APU sounds like a more dev friendly alternative.
Devs bitched about cell because they had to do some extra work to port to it or develop ground up for it. There was no "Port Button" on the PS3. Ah well, it's irrelevant now since it looks like PS4 will use a more standard CPU.
 

Reiko

Banned
Ignore option is your friend :) People I put on ignore tend to get banned within a couple months for whatever reason.

Still can't believe there aren't any more leaks yet.

Someone isn't willing to risk their job yet. But like all leaks for next generation consoles, more info will come in due time.
 
Granted I've made the same point myself in the past but when I say Cell saved the PS3, I mean with it's current configuration.

I agree with you and that's what I said at my earlier posts. Once the damage it's done, let's see what we can do.

As much as I don't agree with Cell inclusion at PS3, I loved what Dice did with Frostbite.

I enjoyed this thread so much back in the day:

http://forum.beyond3d.com/showthread.php?t=59766
 

KageMaru

Member
That's what I said at my earlier posts. Once the damage it's done, let's see what we can do.

As much as I don't agree with Cell inclusion at PS3, I loved what Dice did with Frosbite.

I enjoyed this thread so much back in the day:

http://forum.beyond3d.com/showthread.php?t=59766

Yes but as I explained earlier, it was probably the best choice in the end given their options. Do you really think Nvidia could have provided a comparable GPU? I'm not saying that Nvidia, or Sony, is incapable, but AMD had advantages with working closely with MS and already having experimented with unified shaders in the past.

That is a great thread though. There are plenty of great threads in that forum. The people here who wish to learn by threads like this one should go read some stuff there instead IMO.
 
I'm just not sure. It's true G80 was ready at PS3 launch, but it is also true that it wasn't the planned schedule for Sony. It had several delays cause of unavailability of Cell and BR.

What I think, it's once Sony invested that much in Cell they had no other option even when Xenos was launched. They could have discarded it and go with a cheap OoO dual core + G80. I really don't know if it was pride, bad business decisition or just too much money invested.

As much as I can have some fun with Cellvangelist, I don't believe Cell was a bad idea either. I think it was really cool and innovative project at early 2000. It's just once Xenon+Xenos could do the same with smaller transistor count it just lost the long term race.

Oh, btw, I was expecting something like that thread here. I love how so many guys know what they are talking about but differ at their opinions. I'm a bit sad =(
 
I agree with you and that's what I said at my earlier posts. Once the damage it's done, let's see what we can do.

As much as I don't agree with Cell inclusion at PS3, I loved what Dice did with Frosbite.

I enjoyed this thread so much back in the day:

http://forum.beyond3d.com/showthread.php?t=59766

Its funny how a lot of things in that thread go against things you've said about Cell. No one ever said Cell>>>Xbox, just that Cell>Xenon. And is the reason why PS3 and 360 have roughly the same output. Without Cell, and the same configuration with RSX, PS3 would of not being able to compete with 360. Its debatable whether or not they could of achieved much better than RSX with nvidia anyways. I remember reading countless posts on beyond3d that RSX is the best nvidia could have done for Sony at that time, and was supported with multiple reasons why that I dont remember. Xenos was ahead of its time, just as Cell was ahead of its time. No one is trying to say if it was a good decision to include Cell in the first place or not, or debate that 360 was a better deal when it comes to the silicon budget. The fact is, it was included and thats that.

I dont think many people would disagree a full fledged "Cell v2" with 32 SPU isnt the best bet for PS4, and I dont think many devs would want that. But I always thought a portion of it, or an aspect of the tech would of appeared in some form. Maybe Jeff_rigby's 1PPU4SPU to accompany the x86 cpu.

I dont get why the PowerXCell 8i was never considered. x86 from what I read doesnt seem like the best bet for a gaming console, but who knows it probably wont just be a simple x86 as it will probably be a custom x86 CPU tailored for a console.
 
I do not understand this. A DX11 Compute Shaders GPU render it useless. Not being the full-fledged Cell BE deny retro. And think about the nightmarish motherboard and added cost.

How about no?

I always thought it would just be there for mainly BC, and maybe as an audio processor. Surely a 1PPU4SPU with more memory, and combind with the rest of the hardware could accomplish being fully BC with PS3 games. Might require some emulation. Is that not possible?

Yes it would increase motherboard complexity. but so did having GS+EE on the launch PS3 motherboard.

Can anyone comment on why the PowerXCell 8i might of never been considered? The thing seemed like much beefier and better designed Cell, and surely a modern updated revision of that today would be quite capable, no?
 
Then go the crazy way. Add a full CellBE able to retro PS3 and to be used as a custom co-processor for PS4. Such as Z80 in Megadrive.

Free physics or things like that.

Expensive, though.

PowerXCell 8i is just a server upgrade for Cell.
 

patsu

Member
I'm just not sure. It's true G80 was ready at PS3 launch, but it is also true that it wasn't the planned schedule for Sony. It had several delays cause of unavailability of Cell and BR.

What I think, it's once Sony invested that much in Cell they had no other option even when Xenos was launched. They could have discarded it and go with a cheap OoO dual core + G80. I really don't know if it was pride, bad business decisition or just too much money invested.

Ha ha, G80 was not ready at PS3 launch. It's too hot, too hungry, and has limited quantity. Being ready doesn't mean "it exists". It means they need to make enough, runs cool and available cheaply.

Sony would have an even bigger problem on hand if they had included G80. They will be tied down by nVidia's business terms as well.

As much as I can have some fun with Cellvangelist, I don't believe Cell was a bad idea either. I think it was really cool and innovative project at early 2000. It's just once Xenon+Xenos could do the same with smaller transistor count it just lost the long term race.

Oh, btw, I was expecting something like that thread here. I love how so many guys know what they are talking about but differ at their opinions. I'm a bit sad =(

Cell is interesting in the sense that it delivers more with less. In PS3, it is limited by the memory size. That's why they had to juggle stuff around and work on sophisticated streaming system. And then it also needs to help out RSX, acting as a second GPU for some graphics tasks.

Given enough memory, Xenon + Xenos can't deliver fully what Cell could do. They are too specialized. At least we don't find them in a few supercomputers or clusters like Cell did. It takes later iterations of GPU to be able to match Cell performance.
 

AmFreak

Member
Can anyone comment on why the PowerXCell 8i might of never been considered? The thing seemed like much befier and better designed Cell, and surely a modern updated revision of that today would be quite capable, no?

The 8i has no advantage over the first one, despite being the much higher double fp performance. And there is a reason that the original Cell didn't have high double fp - the standard in games is single, double is seldom used. So putting an 8i in the ps4 is basically the same as putting a normal cell in.
 
As soon as you halve the G80 bandwith, you would get rid of a lot of die size.

I told you, G72 and G80 had the same TDP.

About the second part of your post, you are just too enthusiastic about cell capabilities.
 
Then go the crazy way. Add a full CellBE able to retro PS3 and to be used as a custom co-processor for PS4. Such as Z80 in Megadrive.

Free physics or things like that.

Expensive, though.

PowerXCell 8i is just a server upgrade for Cell.

Cause maybe the 1PPU4SPU will be much cheaper and more capable at the same time. I would think the PPU and SPU's would be fitted with much more memory and would overall be improved.
 

patsu

Member
I always thought it would just be there for mainly BC, and maybe as an audio processor. Surely a 1PPU4SPU with more memory, and combind with the rest of the hardware could accomplish being fully BC with PS3 games. Might require some emulation. Is that not possible?

Yes, it is technically possible. They can even use it to handle security, process Kinect and other natural input devices. However, it may not make economic sense. They can use the budget to beef up their new architecture for example.

In other words, it may be more efficient and effective to reuse the Cell principles (Data locality, pipelining, many cores, etc.) in the new environment.

As for backward compatibility, as long as it makes business sense, Sony will (have to) figure a way out. e.g., port them like today's HD collection, sell an additional B/C module, or stream it in the future in markets where the network is robust enough.

Yes it would increase motherboard complexity. but so did having GS+EE on the launch PS3 motherboard.

Can anyone comment on why the PowerXCell 8i might of never been considered? The thing seemed like much beefier and better designed Cell, and surely a modern updated revision of that today would be quite capable, no?

PowerXCell 8i, Super Companion Chip are designed for a different market.
 

patsu

Member
As soon as you halve the G80 bandwith, you would get rid of a lot of die size.

I told you, G72 and G80 had the same TDP.

About the second part of your post, you are just too enthusiastic about cell capabilities.

Ha ha... it's in the AirForce clusters, the Virginia Tech clusters and the BlueGene/L supercomputer for a few years. The enthusiasts are the supercomputing scientists, not me. I'm just the messenger. If I remember correctly, some banks and oil company also used Cell systems for modelling.

Forget about G80. They will have to solve more than just bandwidth issues to fit it in, both technical and business.

In today's world, they have other options anyway. Why get so hung up over just that GPU.
 
AMD has a SOC design where other processors can be added to the design.
Exactly, AMD has a HSA SOC design where other multi-CPU package designs that use a Xbar switch can plug into the AMD SOC. The 1PPU4SPU package is designed to plug into a multi-CPU architecture. You can put two in a AMD SOC with two other CPU packages all sharing a 4 way Xbar switch. Think of the 1PPU4SPU module as 1 CPU package for a multi-processor system. No additional motherboard complexity.

AMD Jaguar CPU packages (multi-processor) have 4 Jaguar CPUs with cache in that one package. Two of them are rumored for the PS4 which would then have 8 Jaguar CPUs but still have room in the SOC using a 4 way cross-bar switch for 2 more CPU packages which could be 1PPU4SPU (multi-processor) packages. That gives the SOC 18 CPU compute units (not the same as GPU compute units but I wonder if the rumor of 18 compute units in the PS4 got mangled and they meant CPU compute units).

The point of the Method and apparatus for achieving multiple processing configurations using a Multi-processor System Architecture is to provide SPUs that can be used in other Multi-processor systems or combined into a larger Multi-processor system.

This was a Dec 2010 published patent. The timing of the patent is at a point with enough lead time for it to be in next generation and if work on Sony next generation started 9/2010 and the patent was filed 9/2010 this indicates it's possibly going to be in next generation. Doesn't guarantee it but it can't be ruled out with statements that IBM dropped Cell in 2009 and it's dead. Why does the patent say a PPU redesigned (by IBM) from the ground up to match the F04 of the SPUs.
 
Status
Not open for further replies.
Top Bottom