PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)

Status
Not open for further replies.
If that was disclosed, I must have missed it.
[strike]I think it wasn't. Must have heard it somewhere (can't remember where exactly anymore).[/strike]
Edit:
Actually, it is here. 144 to 160 cycles memory latency. Could have changed slightly with the clock changes, though.
 
Last edited by a moderator:
I was looking at the memory subystem leaks. I had forgotten about the CPU article's table.

That's 90-100ns, which is a bit worse than Llano, for reference as to where on the cringe-worthiness scale it should belong.

Trinty is around 60-70ns, which is a 10 or so higher than what Intel's desktop chips (or AMD's non-APUs) can do.
 
Some amount of interference would happen.
The GPU doesn't seem like it would be as severely affected. The numbers for the CPU block's bandwidth (<20 GB/s) are a very small fraction of the whole, so the CPU is physically unable to block the GPU from most of the bandwidth. The cache subsystem is designed to reduce the amount of time the CPU would get anywhere near that.
The ratio of GPU to CPU bandwidth is such that the bus should manage decently, even if the CPUs were actively trying to interfere with the GPU. The GPU is able to tolerate a significant amount of latency, so that doesn't seem like a likely problem.

The CPU might be a bigger loser here, depending on how effectively the memory controllers can balance the latency requirements of the CPUs versus the sheer volume of GPU accesses.
This wouldn't be unique to Orbis.
APUs in general have tended to have noticeably worse memory latency, with the earliest ones having pretty terrible numbers.
Certain numbers of an unspecified competitor I can't use in a versus comparison to Orbis show it's not a GDDR5 or single-pool problem.

I'm aware its a similar potential problem for both consoles, but I'm concerned it might be particularly an issue for PS4. I've seen the hynix timings for 6ghz gddr5 that have been floating around and while some of the timings are marginally higher, some where double in real ns time.
Certain things I've seen have lead me to wonder if the shared memory is causing some stalling issues and possibly impairing cpu performance to a degree. I'm hesitant to talk about them openly, maybe only in a pm, though because if some people read this they will go bonkers celebrating the info and spreading it. haha
 
Let me rephrase that since I cannot yet edit posts: hilighting the editted parts

I'm aware its a similar potential problem for both consoles, but I'm concerned it might be particularly an issue for PS4. I've seen the hynix timings for 6ghz gddr5 that have been floating around and while some of the timings are marginally higher (20%) in real ns time, some where double in real ns time. To what extent those timings are reflective of the memory controller and what extent they are a characteristic of the technolgy I'm curious to know.


I recall seeing benchmarks of system memory latency on current AMD desktop chips and unlike Intel chips AMD doesn't high latency as well intel. I wonder if that carry over to desktop cpu on the AMD APUs?

Certain things I've seen have lead me to wonder if the combination of potential increased gddr5 memory latency + jaguar architcture not hiding latency well + shared memory causing stalling issues and possibly impairing cpu performance to a degree. I'm hesitant to talk about them openly, maybe only in a pm, though because if some people read this they will go bonkers celebrating the info and spreading it. haha
 
Which ones are doubled compared to DDR3?

http://forums.anandtech.com/showpost.php?p=35067042&postcount=1552

Until you attach the memory to that CPU we don't know the latency numbers. I thought I could measure it with SiSoft sandra, but it is measuring CPU to GPU memory latency.

So instead I went digging through the datasheets from Hynix and such and found that at 6.0Gbps frequency the following (with data bus inversion off, but it likely needs to be on and then its normally +1 these figures):

GDDR5 timings as provided by Hynix datasheet:
CAS latency= 10.6ns
tRCD = 12ns
tRP = 12ns
tRAS = 28 ns
tRC = 40ns


DDR3 timings for some Corsair 2133 RAM 11-11-11-28
CAS 10.3ns
tRCD 10.3ns
tRP 10.3ns
tRAS 26.2ns


So its up to about 16% higher latency on the raw access timing in measures people are used to with DDR3. The challenge is that the command rate is at half the frequency, so I doubt it translates like this at all.

In converted timings its impressively high because the base clock speed achieved is actually 1500 Mhz so the GDDR5 uses something like 16-18-18-42.

That is enough of a change in timing to have an impact. Some with an AMD CPU llano/jaguar could tell us roughly how much impact it will have. I'll ask in the CPU forum.
 
Well, Jaguar seemed bad before in relation to prior gen cpus and desktop cpus but now after having seen low power cpus like Apple A7 ARMv8 cpu and Intel Silvermont with similar or better performance it seems way worse. I hope this gen isn´t very penalized for such weak cpus choice. There is no point in having an architecture capable of saving cycles moving data between cpu, gpu and memory if all these saving is wasted afterwards in a weak cpu processing.
If Naughty Dog was also a PC developer I would jump ship to PC this gen with no doubt.
 
Last edited by a moderator:
Well, Jaguar seemed bad before in relation to prior gen cpus and desktop cpus but now after having seen low power cpus like Apple A7 ARMv8 cpu and Intel Silvermont with similar or better performance it seems way worse. I hope this gen isn´t very penalized for such weak cpus choice. There is no point in having an architecture capable of saving cycles moving data between cpu, gpu and memory if all these saving is wasted afterwards in a weak cpu processing.
If Naughty Dog was also a PC developer I would jump ship to PC this gen with no doubt.
Well I don't see what makes Jaguar worse. Those new Atom are faster, thought they come from Intel and use a more advanced process. Pretty much like anything Intel it is not an option. The biggest win vs Jaguar is foremost power consumption, console are power constrained though they are not operating within the pretty drastic constrains you have when powering phones, tablets, or tiny net-book. I don't think that Intel highest sku beat the faster jaguar (in reviews they compare Bay trail with a 1.5Ghz part) though that comes at significant premium in power consumption.

I think the same applies to the A7, it is not a given that it beats Bay trail to begin with as the state of benchmarking on iOS or Android is not that great, I think we got a better, more legit pictures, when Atom and Jaguar platform are compared on windows. Anyway it is a pretty awesome architecture, that is for sure. IPC should be in the same ballpark as Jaguar (+/-), significanlty above baytrail, but when it is all said an done, the main win is power consumption (again).
Pretty impressive, that is only Apple second CPU, they delivered on time (2 architecture in 2 years), etc. but we don't know how high it can be clocked, to which extend the design can be clocked, die size( for the cpu not the SoC), etc.
What I like with that chip is that Apple is not running after marketing point, like the number or core or the clock speed. Pretty much like Intel they focused on good cpu cores and it seems the memory subsystem and cache hierarchy, a good sign for their upcoming designs, the down size is that people are clueless and it may not translate into win for the market they are addressing, not that people are dumb but they are following hype (and Apple is going down in this regard) and are clueless wrt silicon powering their devices.

Anyway, it makes around no difference even though Jaguar should raise one's electricity bill a tad vs hypothetical competing products.
Those CPU architectures are not an option for Sony or MSFT. If Apple ever do something else with that CPU or its heirs it will be for it-self, and Intel, well they were most likely too expensive and I think that whereas they have the man power to deal with a custom project they are facing a changing market and worthy challengers, I'm not sure a console win and some volume for the fabs is that relevant to them.

I don't think that there was an actor (outside of Intel) that were in the situation to provide the combo of CPU and GPU perfs AMD delivered. It might come at a premium in power consumption but it is that relevant.
 
Last edited by a moderator:
It would help in some of the benchmarks if Jaguar's unexplained lack of turbo functionality were remedied. IPC comparisons are uncertain at this point because we know Silvermont does have some pretty good turbo, while A7 is an unknown.

I wish the console variants had any indication of this being changed, but it might be something too unpredictable for their desire for a consistent platform.
 
It would help in some of the benchmarks if Jaguar's unexplained lack of turbo functionality were remedied. IPC comparisons are uncertain at this point because we know Silvermont does have some pretty good turbo, while A7 is an unknown.

I wish the console variants had any indication of this being changed, but it might be something too unpredictable for their desire for a consistent platform.

By Anandtech trial of benchmarking A7 it seems like an astounding ahievement as it gets almost Silvermont performance and wattage but at a 28nm process!. I give Silvermont the win over Jaguar by being in 22nm. At 28nm and without turbo it surely would be worse at the same wattage. But A7 is another history. Sony should have bought an ARM design and put 16 of those chips inside.

And well, GPU is another thing, i would like to talk about Rogue and its efficiency but we have still not enough data, but i would bet that at the same wattage that a 7850 I am sure a big enough Rogue chip would have more than words with it.

Funny thing is consoles have finally gone x86 (don´t counting OG Xbox) when ARM more efficient ISA will reign in the future.

My conclussion: new gen will not last 5 years until next one. Or Apple will dominate console gaming selling an official controller and a hdmi connection. If let´s say Iphone 7 is powerful enough ( they already are at 1 billion transsitors, is not crazy to thing A9 would be in the same ballpark as PS4 APU ) i am sure Rockstar and the likes will be willing to make versions of GTA 7 in IOS. The advantage smartphone and tablet makers have is that new processes are used not only for reducing heat and cost but to improve performance while console manufacturers not.
 
Last edited by a moderator:
By Anandtech trial of benchmarking A7 it seems like an astounding ahievement as it gets almost Silvermont performance and wattage but at a 28nm process!. I give Silvermont the win over Jaguar by being in 22nm. At 28nm and without turbo it surely would be worse at the same wattage. But A7 is another history. Sony should have bought an ARM design and put 16 of those chips inside.
I think there are significant differences in design that make Silvermont a better arch from power perspective. Turbo is here and is significant to the design, you can't remove (even more add it) at whim. I would bet that Silvermont would be more power efficient than Jaguar even if Intel were to produce it on 28nm process. There are multiple reasons behind my claim, from the decoder, to the FP/SIMD units (in order), pipeline length (Atom is design to clock well it is not out of magic that it can turbo really high and I think quite often), etc. Jaguar is a more complex CPU core overall.

The A7 is pretty impressive, it outperforms A15, that is quite something but it is not available to Sony (or msft for that matter), that is the first shipping CPU compliant with the ARm v8 ISA.
A57 would have been possibly a good choice, but it is too late.
As a side note Apple just proved that you are way better off with fewer and better CPU cores (backed by what seems to be a good memory subsystem) than with more lesser CPU cores (and the impact on the memory sub system: more clients, most likely more latencies, more power, etc.)
I'm not sure 16 CPU cores is the best idea.

On the GPU side, well Rogue GPU and late Adreno GPU are great but it is tough to make a comparison with desktop parts, they are supposed to scale well, I guess "supposed" is the proper wording, it is a tough bet to do vs Nvidia or AMD proven high performances, high power, GPUs.

Imo the only manufacturer that could design something significantly better than AMD, investing ~ the same amount of silicon and within the same thermal characteristics, was Intel.
Intel could have design something that may have run around both Orbis and Durango with regard to the CPU perfs and deliver ~the same GPU performances and top of the line GPGPU performances (people forget but Intel GPU are impressive in both perfs per watt and perfs per mm^2 when dealing with compute workload). The issue is that Intel lack the intensive to design something custom and to sell it at an affordable price.

As a side I would not bury Intel that early, baytrail is great, I'm still waiting for proper measurement wrt to power consumption and the die size. THe 14nm version comes next year, so does Haswell shrink. They are facing tough competition, like it did not happened in a long while, but they are doing the right thing and are still the best at their job be CPU desing or process.
I'm not sure they are going to win, though I would not declare that they are to lose, it is too early imo.
EDIT
I missed that part of your post:
My conclussion: new gen will not last 5 years until next one. Or Apple will dominate console gaming selling an official controller and a hdmi connection. If let´s say Iphone 7 is powerful enough ( they already are at 1 billion transsitors, is not crazy to thing A9 would be in the same ballpark as PS4 APU ) i am sure Rockstar and the likes will be willing to make versions of GTA 7 in IOS. The advantage smartphone and tablet makers have is that new processes are used not only for reducing heat and cost but to improve performance while console manufacturers not.
I agree at this point I would be surprised if nobody enter the market (apple, google / whoever it is), the entry bar in perfs is low and with 20nm/14nm ahead and new breakthrough in memory technology it should be doable to compete at a sane price (/without subsidizing hardware).
For Apple and the A9, technology ain't that great ;) the lowest hanging fruits are the easiest one to harvest, they are reaching the point where you pay more and more (silicon and power) to grab less and less extra performances. Though a chip intended for a home console does have to be designed and operated with the constrains of a tablet/phone SoC.
 
Last edited by a moderator:
As I said in another thread, remember Anands benchmarks are looking at the highest performance Silvermont with dual channel 128b memory interface vs the single channel 64b interface on Kabini which will affect a lot of benchmarks. There is a vast gulf in terms of bandwidth between Kabini and the consoles.
 
As I said in another thread, remember Anands benchmarks are looking at the highest performance Silvermont with dual channel 128b memory interface vs the single channel 64b interface on Kabini which will affect a lot of benchmarks. There is a vast gulf in terms of bandwidth between Kabini and the consoles.

Well, from the beginning was clear, when Anand hadn´t even benchmarked Jaguar, that he didn´t like it. I don´t know why ;).
 
And well, GPU is another thing, i would like to talk about Rogue and its efficiency but we have still not enough data, but i would bet that at the same wattage that a 7850 I am sure a big enough Rogue chip would have more than words with it.

Funny thing is consoles have finally gone x86 (don´t counting OG Xbox) when ARM more efficient ISA will reign in the future.

My conclussion: new gen will not last 5 years until next one. Or Apple will dominate console gaming selling an official controller and a hdmi connection. If let´s say Iphone 7 is powerful enough ( they already are at 1 billion transsitors, is not crazy to thing A9 would be in the same ballpark as PS4 APU ) i am sure Rockstar and the likes will be willing to make versions of GTA 7 in IOS. The advantage smartphone and tablet makers have is that new processes are used not only for reducing heat and cost but to improve performance while console manufacturers not.

Almost impossible in 2 years to happan, because:

1. An iPhone has limited power consumption due to heat problem. So you can only depend on new process and better chip design to increase GPU performance. It' s very hard to imagine that iPhone can have increase 15~20x to 1Tflops in just 2 years.

2. Currently the largest RAM for mobile devices is 3GB, while PS4/x1 uses 5GB for "one single game". I don't think any single app can use 5GB RAM in iOS in 2~3 years.


And let's talk about GPU efficiency. Mobile GPUs often use low power process with low voltage, so they can have very good power efficiency compared with PC GPU. In fact a 200 mm2 PowerVR @800 MHz can't have such good efficiency like their mobile counterpart.
 
Almost impossible in 2 years to happan, because:

1. An iPhone has limited power consumption due to heat problem. So you can only depend on new process and better chip design to increase GPU performance. It' s very hard to imagine that iPhone can have increase 15~20x to 1Tflops in just 2 years.

2. Currently the largest RAM for mobile devices is 3GB, while PS4/x1 uses 5GB for "one single game". I don't think any single app can use 5GB RAM in iOS in 2~3 years.



And let's talk about GPU efficiency. Mobile GPUs often use low power process with low voltage, so they can have very good power efficiency compared with PC GPU. In fact a
200 mm2 PowerVR @800 MHz can't have such good efficiency like their mobile counterpart.

Well, for IOS market 2TB is = 400Gflops. You only have to see the advertising about Infinity Blade 3 about being already next gen, and with what? 76Gflops?. Not only Apple is interested in selling this idea, Epic and other developers too. They want the IOS and Android market share.

And i am not sure flops are totally equal between AMD and Nvidia architectures and a true TBDR architecture not totally based in brute force. My kyro 2 in paper was much weaker than Geforce 2 but it outclassed it in many games, above all in the AA department.

You have to be radically better in graphics for people to deny what they will try to sell you. Perceptions are key in marketing.

And after all this talk Tim Cook anyday could sit down with his collegues and decide to launch a proper real good Apple TV with lets say 4 A9s. That they haven´t done it yet is a favour MS and Sony gets of nowhere. Maybe GTA V sales and/or Vita TV make them change their mind.
 
Last edited by a moderator:
Well, for IOS market 2TB is = 400Gflops. You only have to see the advertising about Infinity Blade 3 about being already next gen, and with what? 76Gflops?. Not only Apple is interested in selling this idea, Epic and other developers too. They want the IOS and Android market share.

You have to be radically better in graphics for people to deny what they will try to sell you. Perceptions are key in marketing.

I am trying to clarify why next gen. GTA is very hard to port to iOS system. I am not trying to discuss how to advertise a 400 Gflops console to compete with PS4 with 1.8 T. Besides, a game with 1.8 Tflops will look quite different with 400 Gflops on a large HDTV.
 
I am trying to clarify why next gen. GTA is very hard to port to iOS system. I am not trying to discuss how to advertise a 400 Gflops console to compete with PS4 with 1.8 T. Besides, a game with 1.8 Tflops will look quite different with 400 Gflops on a large HDTV.

As hard or less than porting it from power pc ISA to pc like happened with GTAIV. And for sure they would sell more in IOS than in PC. Remember GTA 3 was ported from last gen processors to IOS already.
Ipad games have to be downscaled to be seen in a fullHD tv. If you get a good looking version in the Ipad screen in the TV they could look even better.
If people like Oculus Rift are going also the smartphone market is because they foresee enough power to make it happen and take advantage of such huge market.
 
Last edited by a moderator:
As hard or less than porting it from power pc ISA to pc like happened with GTAIV. And for sure they would sell more in IOS than in PC. Remember GTA 3 was ported from last gen processors to IOS already.
Ipad games have to be downscaled to be seen in a fullHD tv. If you get a good looking version in the Ipad screen in the TV they could look even better.
If people like Oculus Rift are going also the smartphone market is because they foresee enough power to make it happen and take advantage of such huge market.
I have to say it's much more reasonable to port a PS3/360 game to iOS system after 2 years, not a PS4 gen. game because the performance of GPU and the amount of available RAM is more suitable for porting current gen. (only 500 MB required @ 720p resolution for a typical 360 game).
 
I have to say it's much more reasonable to port a PS3/360 game to iOS system after 2 years, not a PS4 gen. game because the performance of GPU and the amount of available RAM is more suitable for porting current gen. (only 500 MB required @ 720p resolution for a typical 360 game).

Well, all of what i am saying about cpu and gpu evolution must also take into account a radical memory revolution, that it seems is coming. All depends of course of the future mobile SOCs getting enough bandwitdh and memory amount.
 
Status
Not open for further replies.
Back
Top