Analyzing the PS4 / Xbox 720 GPU

The specs for Sony’s successor to the PlayStation brand, the PS4 will feature parts from AMD. Advanced Micro Devices are supplying both the GPU (graphics processing unit) and the CPU for the new system.

This article gives some insight on why certain things are as they are, what to expect and how it all works.

[yewles1]

PS4's Liverpool GPU is confirmed to have 512KB L2 Cache and has 1.848TFLOPS performance. Also, the PS4's CU's are unified for both graphics and GP compute.

[aceitman]

a good read .

[GameNameFame]

and obviously 720 with far less flops of 1.2

[DatNJDom81]

Confirmed that xbox will hold PS4 back next gen once again when it comes to multiplats.

m$: Ok guys, heres a "BASKET". Make sure that the games dont look better on PS4.

Developers 1: But, we can push this just a little more. Our fans deserve the best possible outcome for the game were making, right?

Developer 2: Shut up and take the "BASKET"!

pc fanboyz: "This game will look better on my nvidiazzzz 7.1Tflop masterbation gfx card! Zieg Heil!"

[Gaming101]

A lot of technical speculation that doesn't have all the facts right, like the teraflops on PS4 etc. The NextBox specs are still just rumoured based on what we know so far, so this would be more useful once we know confirmed final specs.

Really we're just splitting hairs here. Devs will be able to come up with whatever will work, however the PS4 won't be held back, the NextBox will just be pared down versions of PS4 games. The only thing holding back would be space limitations, and evidently you install all games to the hard drive of the Nextbox, which has people nuts since it disables used games. It just means you'll be standing around installing 6 DVD discs since Msoft won't support Bluray even if it kills them.

[NewMonday]

from the recent M.Cerny interview the APIs are still not optimized for HSA like programming, when that happens expect a much higher level of performance from the APU.

[ProjectVulcan]

Various little inaccuracies in the report. All you need to know I guess is that the core is maybe 15 percent slower than a 7970M because of cut down CU units and clocks and less than half the performance of a full blown desktop standard 7970 I. e not the gigahertz edition which is even faster.

Memory bandwidth about the same as the 7970M because I have seen 20gb/s or so claimed to be used by the CPU which is roughly what you would expect.

But we known this for some time now...

[hesido]

[Because the ram is shared] "the amount of [bandwith] that the GPU will have ‘available’ will likely be considerably lower than this."

This is not accurate. It won't be "considerably" lower, as CPU bandwidth requirements will be MUCH lower compared to the GPU, so the GPU will have the lion's share of the bandwidth.

Plus, it makes me think they didn't exactly grasp what "unified" ram means. The unified ram will allow 0 copy operations between CPU and GPU, any one of their outputs will be accessible by the eachother without having to move data around, which is a MASSIVE gain and will open up a lot of possibilities further down the PS4 life cycle. As openCL and HSA libraries mature, PS4 will continue to offer performance gains that currently few can imagine.

[nunley33]

@gaming101 If rumors are to be believed, the 720 will have blu-ray. The 360 struggled alot with many games especially in the 2nd half with dvd 9 so how bad would it be with much bigger 720 games? It'll have blu-ray for sure, even if it's like the WII-U with no blu-ray movies supported.

[TheXonySbox]

confirmed; next gen consoles holding back PC gaming yet again.

[MysticStrummer]

Has the 720 GPU been confirmed? If not, can we hold off on these articles until it has been?

Having said that, lol @ DatNDom81's script.

@TheXonySbox - Get your PC brethren to pay full price for more games and that might change. Consoles don't hold back PC gaming. PC gamers hold back PC gaming with piracy and buying new games at incredible discounts. Besides that, people who pimp out a gaming PC overestimate how many other people actually do that. Most PC gaming is done on stock PCs or PCs with minimal upgrades. Devs aren't going to put in too much extra effort so that relatively few people get marginally better graphics.

[Muerte2494]

gamers say that consoles are holding PC games back. It's kinda hard to hold PC back when games don't get released on PC until 6 months after consoles have it (Mainstream games). Anyone remember Super Street Fighter? Did Gears 2 or 3 even make it to PC? I'm mentioning xbox360 games because the architectures were similar.

[PLASTICA-MAN]

The article is just biased, just downplaying the PS4. If it was objective it would mention this :

PS4 GPU is based on GCN 2.0 architecture. PS4 GCN has 8 ACE's, each capable of running 8 CL's each. Tahiti is 2 per ACE, 2 ACE's.

http://www.neogaf.com/forum...

And it was confirmed by Mark Cerny these days:

"This idea has 8 pipes and each pipe(?) has 8 computation queues. Each queue can execute things such as physics computation middle ware, and other prioprietarily designed workflows. This, while simultaneously handling graphics processing."

http://www.neogaf.com/forum...

That's a damn amount of processing power and units that the article forgot to mention. This alone will make the PS4 outshine.

[darthv72]

all i see on here is:

blah blah blah, blah blah blah blah.

No its , blah blah blah. NOT blah blah blah!

you all just need to just be patient for the final release. Then it will get torn down and analyzed piece by piece by every tech site in the known internet.

The current ps3 and 360 offer good games. The 720 and PS4 will no doubt offer good games but for that to happen, we need to let these guys work out the details of what this does and where does that go.

Is that too much to ask?

[ATi_Elite]

can someone please inform me when did SONY confirm a HD7970m for the PS4?

Oh wait they didn't so this article is FAIL!

the ONLY gpu specs we got were 1.84tFlops of compute performance which puts it nicely at a HD7790 level (1.79tflops) when u add the addition 4 compute units that the PS4 version will have now you may get 1.84tflops

(HD7970 is 3.79 tflops)
(HD7970m is 2176 gflops yes gflops)

PS4 is NOT getting a HD7970 or HD7970m way to make stuff up stupid website!

The HD7790 is low watt and has watt tune software that manages watt consumption and is most likely what is going into the PS4.

Now Granted Sony has NOT confirmed anything but going by factual SPECS this seems more realistic.

I'm so amazed how these websites just make crap up to get hits from consolers who just don't check facts as they are too busy glorifying any article that promotes their console as a hero machine.

This website just made up this HD7970m CRAP!

Story = WTF
Like website = NO

[iGAM3R-VIII]

Well even though the desktop version seems only slightly better, it means that the console will be easier to push to their limit which means better quality games

[zmack]

Well the desktop version isn't exactly "slightly" better. Both AMD and Nvidia tend to tone down their mobile GPUs a bit compared to their counterparts and name them accordingly for marketing purposes. Basically, they want to make the consumer see a 7970 or 680 desktop card and then when said consumer checks out a laptop they will see a 7970 and a 680, but the catch is the added m at the end. However, Nvidia did release a 680mx variant and it has the full cuda core count as the desktop version. So, yeah, both companies love to play on names like this, but their mobile cards are still pretty decent.

Steam processors can make a big difference in performance. The 7970m has 1280, which is a bit less than the 2048 found on the 7970. So, I'm sure you have seen a lot of people compare the PS4 GPU to a 7870 desktop version, well that's because it has 1280 stream processors just like the 7970m.

http://www.newegg.com/Produ...

http://www.notebookcheck.ne...

So, let's compare the 7870 to a 7970 in benchmarks:

http://www.guru3d.com/artic...

http://www.anandtech.com/sh...

There can be a 10-20fps difference when comparing the two cards. So, the 7970 can be a bit more powerful depending on the game than the 7870. There's also a GHZ edition of the 7970 desktop GPU with even higher clocks as well. Furthermore, the GPU in the PS4 is running at lower clock speeds than the 7870, so there is a bit more of an fps gap than what was previously stated (7870 = 1GHz and 7970m = 850Mhz).

However, the great news is that a 7970m is still an awesome GPU and it can run a lot games at some nice settings. So, Sony still really picked out some nice hardware for their system.

[JsonHenry]

Can someone explain to me how the next xbox's "move engines" work and how that will change the performance of the console?

[Cueil]

the purpose of the move engine is to basically increase the productivity of the bandwidth available. One of the things that you have to understand is that the 1.2 tflops of the next Xbox will be realized. Most GPUs are 40-60 percent their theoretical peaks. Microsoft uses a varied of methods to achieve near 100 percent theoretical limitation. And remember the PS2 could do more flops then the original Xbox... I think we all understand how much more powerful the Xbox was then the PS2... it's more then flops that's just the new "Bits" of the past two generations.

(was that two dumbed down? I'm really tired, but in reality we wont know exactly how much effect they'll have till the specs are out and people get their hands on it outside of the NDAs)

[NewMonday]

from what I understand they are for going around bottlenecks. the PS4 GPU has a cache bypass that dose something like it.

[JsonHenry]

No, that explained it well enough for me. I have a novice's understand of PC hardware and the like (been building my own for 15 years) just I was not sure how the move engine was being used. I guess just making the system more efficient would help more than adding raw throughput since the flops of performance thing is generally way over-rated in terms of what is "capable" vs. what is "actual". Thank you for your reply. +1 bubble

[THEDON82z1]

@Cueil = TRU what you said,but microsoft problem will be the same one PS3 HAD.It will take heavy optimization(time/resources=MO NEY) AND UNLESS microsoft builds up some REALLY good 1st party studios to take advantage OF IT LIKE SONY did with ps3,then it will really just become another road block in developers way when trying to get there game out in a timely mannor.As I said in another post, the slight advantage they had by desighning this system was killed when Sony double its gddr5 ram.You also have to remember that even when Sony pump all that money into 1st party R/D it still took alot time for developers to really harness it.The result was multiplates on PS3 struggled(in the beggining)untill they built it on ps3 then ported to xbox. Thats what I see is coming to the next xbox when everyone starts asking about 1080p/60fps as a standard.The only diffrence will be that I believe developers will for the most part stick with ps4/pc as lead development.

[Cueil]

What people don't seem to get is that programing for the X720 is not going to be any different then what they were doing with the 360 only instead porting games over to and from PS4 will be easier since they share the same hardware programming language and both use Microsoft Visual Studio if I'm not mistaken... a great move by Sony if they were not already doing so. Microsoft isn't becoming some kind of complex piece of hardware it's simply creating efficiency and that seems to be the direction Microsoft and AMD have been moving towards. I'm not going to pretend to know the real specs of the Xbox, but my guess is that the system is being build so that programmers can access it's power at or near theoretical limits.

[THEDON82z1]

@Cueil -I hope you are right because if the rumors are true and 3 out of 8 gigs of ram is for OS,they are good to need every trick they can get once next gen games start to muture around 2014/15.I dont want to see my PS4 (MULTIPLATFORM) GAMES held back in anyway.

[Skynetone]

while the xbox is busy doing calculations, the move engines are free to add even more calculations

its like a turbo boost

what does it mean in the real world, i guess well have to wait for halo 5

[DeadlyFire]

Move engines do boost, but still not to the same level. It will be slower. 32 MB ESRAM with move engines waiting in line with data at 100 GB/s can only boast so much with data while 8 GB can run at 176 GB/s on PS4.

[Cueil]

I understand that the numbers look pretty, but bandwidth is a lot more complex then some of you seem to understand. I'm by no means an expert, but there are all kinds issues with just throwing up arbitrary numbers without understanding all the parts and technology behind those numbers. Do we know how much memory and cycles the OS is going to take up or how much is going to be reserved for the social features of each console? That stuff takes up bandwidth.

[Mkai28]

Xbox 720 has four move engines itself which allows for fast direct memory access to take place.

Their true purpose is to take workloads off of the rest of the system while yielding positive results at low cost.AKA "Secret Sauce"

The four move engines have abilities such as: from main RAM or from ESRAM, to main RAM or to ESRAM, from linear or tiled memory format, to linear or tiled memory format, from a sub-rectangle of a texture, to a sub-rectangle of a texture, from a sub-box of a 3D texture and to a sub-box of a 3D texture.

Each of the four move engines can read and create 256 bits of date per GPU clock cycle, which equals out to be a peak throughput of 25.6 GB/s two ways.

All of the engines use a single memory path, resulting in the best throughput for all of the engines that would be the same for only one of the engines.

They share their bandwidth with different components of the GPU, like video encode and decode, the command processor and the display output. The other source typically only consume a small amount of the bandwidth.

The great thing about the move engines is they can operate at the same time as computation is taking place. When the GPU is doing computations, the engines operations are still available. While the GPU is working on bandwidth, move engine operations can still be available so long as they use different pathways.

[profgerbik]

PS4! PS4! PS4!

[WeAreLegion]

YEEEAAAAAAAAAAAHHHHHH

[DragonKnight]

I think it's great that redgamingtech.com knows what's in the Xbox 720 when no one else does.

[TheKayle1]

ok is the time...look this

http://i.imgur.com/g61Dtdy.... (amd 7970)

compared to the xbox gpu

http://i.imgur.com/sqkJUmO....

i think the problem in vgleaks and the others siste is that they calculate and SC ..as a CU for a total of 12 CU :)

http://www.vgleaks.com/dura...

.....while a sc (sharer cores) is a group of 4 cu (for a total of 32)..the document is ok ..the interpretation is wrong :)
.......for this reason in the latest durango devs kit they use a amd 7970 as bkillian said on GAF

http://www.neogaf.com/forum...

"The current devkits have 7970's inside them...."

obv bkillian isnt in favour of the xbox (lol) so he think we will receive a downgrade (x2LOL) and this sound pretty stupid...

now why use a 7970 (where devs use the kit to point a performance target and then downgrade it? :D:D:D:D)

...old rumors was pointing at gtx680 performance and if we go deep we will see the tech is pretty the same

nvidia use 8 SMX (= group of 4 cu) in the gtx680
amd use 8 CU ARRAY (group of 4 cu) in the 7970
(this r just names! for groups of cu!)

so why xbox gave different name at the cu if they was just cu calling it Shared Cores ....(as amd and nvidia did) ..jsut coz arent JUST cu...r groups of cu

8smx= 8*192=1536 cuda cores (gtx680)
8 Cu arrays = 8*256=2048 ALU (amd 7970)

but we have 12 CU ARRAYS on the vgleak leaked document and u know what ? :O

on google the next 8970 desktop XTX is rumored 3072 ALU :O

and if u look at the xbox

12 SC r 12 AMD CU ARRAY = 12*4 = 12*256= 3072 ALU :O

3072 alu is + or - x3 time faster than a ps4....

now this is pretty hard to accept also for bkillian ("WTF WHY THEY USING A 7970!"they thinkin)..so ppl point at a downgrade..or other things...

im saying from months....just wait....the show...coz we will have a lots of laughs to do there :)

now why UE4 have a new tech demo?! the infiltrator one?
and why ps4 showed a old tech demo (elemental) that we know alredy how much it was bad compared to the pc version

well coz the infiltrator was running on gtx 680 :O that is in the same ball park of 7970 :O and .....i bet my balls that we will see this tech demo at the xbox show ;)

[TIER1xWOLFPACKx]

lololololol

[DeadlyFire]

Both are aiming for performance comparable to 3 Tflops with PS4/X720. They have two different ways. Only when we see them in motion will we know which is better. PS4 has a better starting point so its natural it will be better.

You are aware that 12*64 = 768 right? Just wondering

[strickers]

Digital Foundry( at Eurogamer) said the Xbox method was a band aid and cannot match ps4 bandwidth/speed. Move engines were described as more about media compression and decompression . Rumoured PVR stuff most likely . I'm not an expert so just saying what experts said.

[DeadlyFire]

I agree its not a perfect way to catch up to PS4 at all. The degree of difference will be noticed by all.

[sourav93]

What did I just read? 0.o

[TheKayle1]

now ;) another hit that vgleaks (i really dont understand how they did this mistake!)..is

7970 have 6 MC @ 64bit =384 bits

http://i.imgur.com/sqkJUmO....

the xbox have move engine that r

4 ME @ 256 bits = 1024 bits
http://i.imgur.com/g61Dtdy....
;)

the leaks on vgleaks (and on what everyone is basing on) is correct...the interpretation is wrong....but was easy...

ps before disagree pls explain in what i could be wrong..seen that for u this is speculation i would hope u do speculation based on facts....so pls explain

[sourav93]

I think before people explain why you might be wrong, first be clear in what you're saying. It's hard to follow what you have written.

[TheKayle1]

sorry i know my english is a shit..but pls try to follow me coz is important to understand why ppl r understimating and not understanding the durango gpu

this is the vgleaks leaked ms durango diagram
http://www.vgleaks.com/dura...

now read under the "Compute" paragraph :
"Each of the 12 Durango SCs has its own L1 cache,LSM(Local Shared Memory), and scheduler, and four simd units" .

read 4 (ok we r talking about FOUR )simds...now

this is the orbis leaked docu ( that we know is real)
http://www.vgleaks.com/worl...

go in the UPDATE part and u see:

"Each CU contains dedicated:

- ALU (32 64-bit operations per cycle)"

Now all is said there! :) come back on the durango leaks and u will read:

" A SIMD executes a vector instruction on 64 threads at once in lockstep "

64 threads at once!!!.....:) in th ps4 we have 32 threads!!!

and we have FOUR (4) simd in every SC

12 Shader cores each on 64 threads = 768 ..on durango against
14 CU each 32 threads = 448 If you add the other 4CU it's = 557

as u see ....that is why rumors about DOUBLE PRECISION came out..just coz durango do double the threads of the ps4

again on the 7970 we have 32 CU divided in groups and amd smartly call them 8 CU ARRAY...same thing happen in GTX 680 they call 8SMX a group of 32 CU.

same things is in the durango ..12 SC (shader cores) arent 12 cu...but 48 cu....4 x sc...

as u can see how i showed in the diagram in top post

in fact as thuway (neogaf trusted insider)..saying in the last durango devs kit they got a 7970's inside ( i dont know why he use the S....maybe coz r two to emulate? booh)...but the performance target is that...not 1.2tf..now i ended my bubble so if u interested send me pm

i hope u can understand this

@deadlyfire we know that xbox is aiming at that perf....specs talk for themself.....ps4 specs dont..but rumors saying they overclocking something and "want code to the metal"....

normal cu..usually lose 30% 40% of their compute power..in realworld perfomance...

[Munky]

Putting aside all the technical stuff (that 99% of the ppl on N4G don't understand anyway) in this article and articles similar to this, the fact is that a lot of the devs are saying that the PS4 is slightly more powerful the the Nextbox, period.

[JeffGUNZ]

Who is saying that? No one, I repeat, no one has said ANYTHING about the next xbox. You can't base your conclusion on rumors and speculations from poor gaming websites and "journalists" foaming at the mouth for hits. I am not saying the nextbox will be the same spec or better, rather I am waiting until an official reveal or statement is released from MS.

[Munky]

I think you are missing my point. The fact they are saying it is slightly powerful speaks volumes, it is not the massive gap ppl are saying or hoping it's going to be. In the end the two consoles should be roughly equivalent in terms of power. But like you said, who the f*ck really knows until MS officially reveals the specs of the Nextbox.

[MysticStrummer]

At least one dev has said something about it. He said about the PS4 that Sony is clearly in it to win, and about 720 that MS has a different attitude in comparison to PS4. Sure it's not a direct reference, but it speaks volumes. Put that together with all the MediaBox and KinectBox rumors, and a picture definitely emerges. Only time will tell how accurate it is, but the more "bad" rumors pile up, the more likely at least some of them are to be true. I put bad in quotes because, well, look how many people bought a Wii. "Bad" to PS3/360 gamers may still spell a win for MS next gen. We'll see.

[JeffGUNZ]

@ MysticStrummer

What developer said that and do you have a link? I am under the understanding NO developer can release or even acknowledge the next xbox until MS announces it. Is there a link you can produce or did you read this from an article that was a rumor/leak/speculation?

[JeffGUNZ]

Ah, thanks for the clarification. I agree with you.

[Munky]

no worries man. peaz

[Tzuno]

E3 has the answers.

[LostDjinn]

Look at all the comments here. How many of them are misinformation? Seriously, how many don't even make sense?

It looks like this BS FUD thing is going to be with us for some time yet.

It's sad really.

[Machiavellian]

I am actually waiting for someone to debunk TheKayle1 post because after going through it, my thoughts on the Nextbox has changed. So before anyone start to talk about 1.2TFlops and all that, can someone break down what TheKayle1 stated because that makes a big difference in the interpretation of VGLeaks chart. I will repost what TheKayle1 stated for reference

sorry i know my english is a shit..but pls try to follow me coz is important to understand why ppl r understimating and not understanding the durango gpu

this is the vgleaks leaked ms durango diagram
http://www.vgleaks.com/dura...

now read under the "Compute" paragraph :
"Each of the 12 Durango SCs has its own L1 cache,LSM(Local Shared Memory), and scheduler, and four simd units" .

read 4 (ok we r talking about FOUR )simds...now

this is the orbis leaked docu ( that we know is real)
http://www.vgleaks.com/worl...

go in the UPDATE part and u see:

"Each CU contains dedicated:

- ALU (32 64-bit operations per cycle)"

Now all is said there! :) come back on the durango leaks and u will read:

" A SIMD executes a vector instruction on 64 threads at once in lockstep "

64 threads at once!!!.....:) in th ps4 we have 32 threads!!!

and we have FOUR (4) simd in every SC

12 Shader cores each on 64 threads = 768 ..on durango against
14 CU each 32 threads = 448 If you add the other 4CU it's = 557

as u see ....that is why rumors about DOUBLE PRECISION came out..just coz durango do double the threads of the ps4

again on the 7970 we have 32 CU divided in groups and amd smartly call them 8 CU ARRAY...same thing happen in GTX 680 they call 8SMX a group of 32 CU.

same things is in the durango ..12 SC (shader cores) arent 12 cu...but 48 cu....4 x sc...

as u can see how i showed in the diagram in top post

in fact as thuway (neogaf trusted insider)..saying in the last durango devs kit they got a 7970's inside ( i dont know why he use the S....maybe coz r two to emulate? booh)...but the performance target is that...not 1.2tf..now i ended my bubble so if u interested send me pm

i hope u can understand this

@deadlyfire we know that xbox is aiming at that perf....specs talk for themself.....ps4 specs dont..but rumors saying they overclocking something and "want code to the metal"....

normal cu..usually lose 30% 40% of their comput

[Mkai28]

I am glad you did your research on this, Double precision Is a type of floating-point number that has more precision (that is, more digits to the right of the decimal point) than a single-precision number. The term double precision is something of a misnomer because the precision is not really double. The word double derives from the fact that a double-precision number uses twice as many bits as a regular floating-point number. For example, if a single-precision number requires 32 bits, its double-precision counterpart will be 64 bits long.
The extra bits increase not only the precision but also the range of magnitudes that can be represented. The exact amount by which the precision and range of magnitudes are increased depends on what format the program is using to represent floating-point values. Most computers use a standard format known as the IEEE floating-point format.

So lets just say, the Next Xbox would to have just 1.79TFlops. Being double precision aka " two PCs taped together" will have a clock speed of 3.58TFlops. Sony's PS4 having a single precision clocked at 1.84TFlops. Even if it was at the first rumored Durango specs to have 1.23 TFlops, being doubled is clocked at 2.34TFlops . Getting the picture ...If true, then this machine will RISE TO HEIGHTS where other console can't even fathom.