For what purpose does a PS4 need 8 weak cores?

[Enigmoid]

By June 25, 2013, which is cutting it crazy close to the release date of PS4 for testing and manufacturing, NV released GTX760 with 161W power usage, or 68% more power than an HD7850 2GB, but only 45% faster.
http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_760/25.html

However, that launch was too close to realistically use in PS4 due to testing+manufacturing and logistics for Nov 2013 time-frame. To make launch date, you would have had to choose GTX660/660Ti/670/680, but all of them have higher power usage than 100W, so cannot be directly comparable to a 7850/7970M. Your only option at this point is the 680M with its 100W TDP.

Problem is 680M has a voodoo power rating of 143 vs. 7850 with 141. You gain nothing really while the former probably costs $300+ from NV. The 680MX had 122W TDP, so that wouldn't work.

That brings us to 770M, 775M and 780M. 770M < 680M, so that's a fail. 775M is basically identical to a 680M in performance, so that's not gonna work either. Finally 780M would have worked and provided 30% more performance than 680M/7970M/PS4's GPU. 780M had a retail MSRP of $750 USD. With NV's profit margin of about 55-56%, NV would have had to sell that GPU for $330 USD at cost to get just 30% more performance from PS4's GPU on launch date.

Considering in 4-5 years we'll have GPUs 5-10X faster than PS4 and an extra 30% would hardly make a difference for PS4 longevity in 2018-2019, it's fair to say it would have been the dumbest decision in the world to put a 780M inside a PS4/XB1 for a mere 30-40% faster GPU performance and for Sony to take a $400+ USD loss on the GPU alone because no way would NV have sold that GPU with $0 profit.

Therefore, Sony's choice for an 1152 SP GCN was hands down the best option possible.

You realize that the mobile GPUs you are talking about are binned dies of the desktop chips? There is no reason not to use the much cheaper desktop components (which are pushed more for performance rather than efficiency as the mobile gpus are). There is no $330/750 780m that was ever on the market (the demand would be much too high for the binned supply). The 680m/780m could easily be simulated; simply put a TDP lock on GK104 at 100W and adjust clocks.

A 7970m performs similarily to the console GPUs but IS NOT COMPARABLE TO THE CONSOLE CHIPS. 7970m gets binned, console APU either passes or fails the bin, there are no die harvests; that's why the consoles run comparatively low clocks. The console APU functions as a poorly binned desktop chip (extra voltage to keep the poorer chips going) therefore the is no rhyme or reason in comparing it to a highly binned and expensive mobile chip.

Your usage of mobile chips is completely absurd.

 

[R0H1T]

Now very likely the real reason they didn't go intel was that they wanted an APU and also intel doesn't like doing die shrinks of its existing architectures.

The i3 is an APU, with its IGP, & no way 2core + HT would've been any better than 8core Jaguar especially when it had to share a constrained TDP with another (Nvidia?) GPU.
This is what the (improved) Jaguar is capable of ~
http://www.notebookcheck.net/HP-Pavilion-13-a000ng-x360-Convertible-Review.127351.0.html
http://www.notebookcheck.net/HP-Pavilion-13-a093na-x360-Convertible-Review-Update.130928.0.html

 

[john5220]

since we don't look at business and money on the whole from personal feelings etc

Would it not be better for intel to have sold very cheap i3 for the PS4 even if they made say $1 billion at the end of the life cycle. Thats a tiny amount of money for intel over 5 years BUT

BUT

its $1 billion more. It only benefits them so the question is why not stoop to the cheapness of AMD for the sake of business? even Nvidia could provide very cheap GTX 660, as small as the profits would be its still profits not so?

What are they going to lose out on?

 

[R0H1T]

since we don't look at business and money on the whole from personal feelings etc

Would it not be better for intel to have sold very cheap i3 for the PS4 even if they made say $1 billion at the end of the life cycle. Thats a tiny amount of money for intel over 5 years BUT

BUT

its $1 billion more. It only benefits them so the question is why not stoop to the cheapness of AMD for the sake of business? even Nvidia could provide very cheap GTX 660, as small as the profits would be its still profits not so?

What are they going to lose out on?

You think MS/Sony didn;t think of that? The most obvious deduction from the two threads on this topic is that no chipmaker would've delivered the level of performance (CPU+GPU) on an SoC that AMD made for the consoles. Even taking price out the equation, neither Intel nor Nvidia could ever get around their weak points (GPU & CPU respectively) & any other compromise would've blown the TDP budget of these consoles, this was never a reality for anyone besides AMD.

 

[Enigmoid]

The i3 is an APU, with its IGP, & no way 2core + HT would've been any better than 8core Jaguar especially when it had to share a constrained TDP with another (Nvidia?) GPU.
This is what the (improved) Jaguar is capable of ~
http://www.notebookcheck.net/HP-Pavilion-13-a000ng-x360-Convertible-Review.127351.0.html
http://www.notebookcheck.net/HP-Pavilion-13-a093na-x360-Convertible-Review-Update.130928.0.html

Highest binned Beema part against mid/low HW-U. Beema throttles to 1.4 ghz during max load test if that is where you are taking power numbers.

2C + HT SB/IVB would have been comparable in performance. 22 nm would use a similar amount of power (54W for CPU + IGP but mobile 35W IVB chips have no problems staying under 35W for CPU only at 3 ghz +).

 

[R0H1T]

Highest binned Beema part against mid/low HW-U. Beema throttles to 1.4 ghz during max load test if that is where you are taking power numbers.

2C + HT SB/IVB would have been comparable in performance. 22 nm would use a similar amount of power (54W for CPU + IGP but mobile 35W IVB chips have no problems staying under 35W for CPU only at 3 ghz +).

And you think that i3 wasn't binned ? Also ~

Temperature

During the stress test (Prime95 and Furmark run for at least one hour), the CPU clocks at 1.4 GHz and the graphics core runs at full speed (800 MHz) in mains operation. On battery, the CPU clocks at 1.2 and 1.4 GHz and the graphics core clocks at 400 to 500 MHz. The Pavilion does not get too hot. The temperatures remain significantly below 40 °C in all quadrants. The same is also true for the sister model.
Power Consumption

While idle, the power consumption of the Pavilion is below 10 Watt - just like the sister model's. But, they perform differently during the stress test. The Pavilion's consumption of 23.8 Watt is lower than the sister model's (29.7 Watt).

This for the beema variant & again probably because of this ~

Previously once the silicon temperature hit 60C, AMD would cap max CPU/GPU frequency. However what really matters isn&#8217;t if the silicon is running warm but rather if the chassis is running too warm. With Beema and Mullins, AMD increases the silicon temperature limit to around 100C (still within physical limits) but instead relies on the surface temperature of the device to determine when to throttle back the CPU/GPU.

From ~ www.anandtech.com/show/7974/amd-beema-mullins-architecture-a10-micro-6700t-performance-preview/2

With the consoles AMD had a lot of legroom & as such throttling would be the least of their worries especially after a die shrink.

 

[TurtleCrusher]

An i3 can't even stream twitch with any intensive game going using a PC. Going with a dual core is a horrible idea. With twitch taking up most of a core on an i3, add OS, what do you have left?

i3? You're out of your mind thinking that's a good idea.

 

[el etro]

john5220, the PS4 CPU allowed the console to reach a unique trifecta: Price/Performance versus DieSize/TotalChipDieSize versus PowerConsumption/chipTDP.
So the jaguar CPU was unique to the PS4 building.

 

[RussianSensation]

People forget that the RSX,even though technically the same G70 chip as in the 7800GTX,was actually slower in reality. It had only a 128 bit memory controller(instead of a 256 bit memory controller) and half the ROPs of the G70 found in the 7800GTX.

From the RSX Wiki page:

:thumbsup: Excellent points. I've been saying for a while that the GPU in PS3 was junk on day 1 and nowhere near 7950GT/7900GTX. Around PS3's launch 8800GTX completely owned the PS3's/Xbox 360's GPU and as I said 20 months out GTX280 was leagues apart, 5.75-6X faster than PS3's GPU. Essentially RSX became shockingly outdated with paltry memory bandwidth and major VRAM bottleneck less than 2 years after launch relative to the best PC cards such as HD4870 and 280. However, to this date, there is a myth that PS3/360 were some uber powerful consoles for years after launch when in fact they became outdated quicker relative to PC GPUs that followed, and most PC gamers refuse to acknowledge this data.

To this date no game ever released for PS3/360 matched 2007 Crysis 1 on the PC. Yet, CDPR already claims that Witcher 3 on PS4 will equal Witcher 2 PC graphics. That's pretty good because the generation is just getting started.

Intel wouldn't sell it below $100 without the GPU? Intel is currently selling the cheapest i3s below $100 with the igp ($99 in MC). One of the higher end i3s.

You are not seeing the big picture. Core i3 + Intel motherboard would still use more power than an 8-core Jaguar part. Secondly, Intel motherboards/chipsets cost significantly more as a total package. Finally, a Core i3 would have been an epic fail for a console since they would have allocated 1 core for OS/background tasks leaving just 1/2 of the Core i3 for gaming. You think PS4 using 2 cores for background/OS would have been not resulted in a major issue for a Core i3? We've already seen the Original Xbox using Intel Pentium chip and it didn't do squat of a difference against GameCube. In fact the less powerful PS2 easily outsold both.

A 7970m performs similarily to the console GPUs but IS NOT COMPARABLE TO THE CONSOLE CHIPS. 7970m gets binned, console APU either passes or fails the bin, there are no die harvests; that's why the consoles run comparatively low clocks. The console APU functions as a poorly binned desktop chip (extra voltage to keep the poorer chips going) therefore the is no rhyme or reason in comparing it to a highly binned and expensive mobile chip.

Your usage of mobile chips is completely absurd.

It's not arburd. As in the other thread you are not understanding the main point I am making. The best performance/watt is seen in the mobile space. Therefore, 680M/680MX/775M/780M and 7970M represent the best possible GPU performance in an energy constrained space that you could have had at that time. Using those chips as a reference point tells us the absolutely best reasonable 100W GPU performance that PS4 could have had.

As I already said, unless Sony or AMD wanted to take a hit on yields and release a fully unlocked 7970M with 1280SPs, the only way to tangibly improve performance was to use a chip like 780M (whether desktop or mobile has no relevance to the upper-performance boundary since 780M tells us the best NV could have done at that time). Under the best case scenario, we would have had a PS4 with 30-40% faster GPU but considering 780M sold for $750 in retail, such an optimized desktop/mobile NV chip would have been completely unfeasible to use in a PS4.

Furthermore, given that PS4's GPU is at least 30-40 faster than XB1's but we aren't seeing PS4's games looking much better, even if PS4 had a GPU 30-40% more powerful, 90% of the developrs would have never prioritized for that level of performance leaving both XB1 and Wii U on the sidelines. The risk of making 1 of 3 consoles too powerful is that most developers will never spend the time and $ extracting maximum performane out of it. Therefore, using a much more expensive 780M style GPU in PS4 would have been a big waste of time and resulted in massive losses for Sony or no profits NV.

Right now PS4 is powerful enough and is cheap enough that it hit the perfect sweet-spot for a next gen consoles. 18.5 million sales reflect that. PS4 with 780M style GPU and an i5 priced at $599 would have completely bombed against this holiday's XB1 going for $330-350. ^_^

An i3 can't even stream twitch with any intensive game going using a PC. Going with a dual core is a horrible idea. With twitch taking up most of a core on an i3, add OS, what do you have left?

i3? You're out of your mind thinking that's a good idea.

Well apparently there are A LOT of PC gamers on our forum that think a G3258 OC is better than FX8000/9000 series for gaming and think dual core is still awesome for modern gaming. When "unknown" sites show that modern titles really run much better on FX8000/i5, these tend to be ignored since it's "not the popular opinion".

The other point brought up is that MS's cost savings of fusing the CPU and GPU together later in Xbox 360's life was noticed by ALL of the major console makers. How would MS/Sony realize future die shrink power saving benefits together with reduced cost savings had they gone with an Intel+AMD or Intel+NV solution? Considering MS's experience with having real world facts on the yields, costs and benefits of going APU with the 360's later SKUs, I am sure they carefully considered the alternative of going i3+NV/AMD stand-alone GPU.

Since XB was meant to be a media device that promotes multi-tasking such as Skype + video steaming + gaming, a Core i3 would have been a massive fail for such a console.

It's shocking how many people are disappointed with the current consoles considering their prices and that we have games like Infamous Second Son, Driveclub and Ryse Son of Rome providing amazing graphics on a $350-400 console. Sure, with the cost of PC games and flexibility of mods, I still like PC gaming, but for the millions of gamers out there, PS4 is a FAR better designed/balanced gaming console than PS3/Xbox 360 ever was. The major issues with PS4 are poor media playback capability and lack of BC support.

I mean we will have Witcher 3 and Uncharted 4 on PS4. To max out Witcher 3 on the PC will require a GPU that costs more than PS4 alone. The diminishing returns of highest level of graphics beyond medium/high are extremely costly, requiring 3-4X the graphical power of PS4's GPU for less than commensurate increase in graphical quality. The average gamer would never pay $600+ for a PS4 that could run Witcher 3 at Ultra vs. a $350-400 XB1/PS4 that could run it at a Medium/High. That's why the goal of consoles is not to shove a 4790K with a 780Ti in there and sell it for $400. Last generation of consoles has shown that the historical model of loss leader strategy (printer+ink/razor+razor blades) for consoles is no longer sustainable. It's simply too risky and it takes too long to re-coup the losses on hardware, which unnecessarily extends the normal life-cycle of consoles, thus actually hurting the industry long-term. Game publishers/developers would also like for the adoption of next gen consoles to happen as quickly as possible in order to justify the large expenditures associated with game development today. This is why this generation we are seeing $299-399 consoles.

 

[MeldarthX]

RS - they also seem to forget that MS got burnt by Intel with the first Xbox along with Nvidia......*though that spat was both MS and Nvidia's fault* MS has pretty much written off Nvidia for their consoles; specially since they know they can get custom designs through AMD and are willing to work with MS on them.

Second Nvidia burned Sony last round with graphics chip they gave them... they took the 7800 gt off the shelf and then cut the bus in half; crippling it *along with the fact it was also part of the bumpgate chips and Nvidia knew this all along*. 90% of the ylods on the PS3s were; yep you guess it graphics because of the poor solder used by Nvidia. *I've reflowed a few thousand where I used to work and reballed several hundred*

Sony did approach ATI/AMD for a chip last time but AMD said they couldn't give them a custom one like they wanted in the time frame......*thus the approach to Nvidia who gave them that*

Nvidia's pretty much burned their bridges on the consoles.....

 

[Tuna-Fish]

It's not arburd. As in the other thread you are not understanding the main point I am making. The best performance/watt is seen in the mobile space. Therefore, 680M/680MX/775M/780M and 7970M represent the best possible GPU performance in an energy constrained space that you could have had at that time. Using those chips as a reference point tells us the absolutely best reasonable 100W GPU performance that PS4 could have had.

No, they don't. 7970M is binned for low power. This means that as they come out of the production line, all of the Pitcairn dies are tested for power use. The least power-consuming of those tested become 7970Ms, the most power-consuming ones become 7870s. As there is significant variability in power use, binning allows sale of less-consuming mobile models without having to waste half the dies, as they can be sold as the desktop models.

In contrast, consoles one sell a single model. None of the chips they make can be used for anything else useful. This means that unless they are willing to trash half of their production output, they have to design for the upper end of the power consumption of the manufactured chips. This means they absolutely cannot match the performance/power of mobile chips.

 

[Enigmoid]

You are not seeing the big picture. Core i3 + Intel motherboard would still use more power than an 8-core Jaguar part. Secondly, Intel motherboards/chipsets cost significantly more as a total package. Finally, a Core i3 would have been an epic fail for a console since they would have allocated 1 core for OS/background tasks leaving just 1/2 of the Core i3 for gaming. You think PS4 using 2 cores for background/OS would have been not resulted in a major issue for a Core i3? We've already seen the Original Xbox using Intel Pentium chip and it didn't do squat of a difference against GameCube. In fact the less powerful PS2 easily outsold both.

You can't take the desktop motherboard prices. Any console motherboard is going to be stripped down solely to what is needed (no need for 12 USB, 4 PCI-E slots, etc.) The actual motherboard would be slightly more expensive (PCH) but all in all quite similar. Its easier to cool two smaller sources of heat rather than one large heat source though it may be more expensive to. Why would they need an entire IVB core + HT for 'background' tasks? Very likely they simply would have redesigned their software stack to avoid this type of thing.

It's not arburd. As in the other thread you are not understanding the main point I am making. The best performance/watt is seen in the mobile space. Therefore, 680M/680MX/775M/780M and 7970M represent the best possible GPU performance in an energy constrained space that you could have had at that time. Using those chips as a reference point tells us the absolutely best reasonable 100W GPU performance that PS4 could have had.

As I already said, unless Sony or AMD wanted to take a hit on yields and release a fully unlocked 7970M with 1280SPs, the only way to tangibly improve performance was to use a chip like 780M (whether desktop or mobile has no relevance to the upper-performance boundary since 780M tells us the best NV could have done at that time). Under the best case scenario, we would have had a PS4 with 30-40% faster GPU but considering 780M sold for $750 in retail, such an optimized desktop/mobile NV chip would have been completely unfeasible to use in a PS4.

Furthermore, given that PS4's GPU is at least 30-40 faster than XB1's but we aren't seeing PS4's games looking much better, even if PS4 had a GPU 30-40% more powerful, 90% of the developrs would have never prioritized for that level of performance leaving both XB1 and Wii U on the sidelines. The risk of making 1 of 3 consoles too powerful is that most developers will never spend the time and $ extracting maximum performane out of it. Therefore, using a much more expensive 780M style GPU in PS4 would have been a big waste of time and resulted in massive losses for Sony or no profits NV.

Right now PS4 is powerful enough and is cheap enough that it hit the perfect sweet-spot for a next gen consoles. 18.5 million sales reflect that. PS4 with 780M style GPU and an i5 priced at $599 would have completely bombed against this holiday's XB1 going for $330-350. ^_^

The absurd thing is you are taking mobile prices out of it. 780M performance level is fine but 780m is a binned GK104 with reduced clock speeds and thus pretty much the same thing as a downclocked 770 (with 10% higher power at those clocks). PS4 generally gets higher resolution, higher framerate, or slightly better effects. One of the major reasons why the PS4 is outselling the XBONE is because of this.

http://ca.ign.com/wikis/xbox-one/PS4_vs._Xbox_One_Native_Resolutions_and_Framerates

PS4 gets higher fps of resolution in more than half the available titles.

I mean we will have Witcher 3 and Uncharted 4 on PS4. To max out Witcher 3 on the PC will require a GPU that costs more than PS4 alone. The diminishing returns of highest level of graphics beyond medium/high are extremely costly, requiring 3-4X the graphical power of PS4's GPU for less than commensurate increase in graphical quality. The average gamer would never pay $600+ for a PS4 that could run Witcher 3 at Ultra vs. a $350-400 XB1/PS4 that could run it at a Medium/High.

Price is king and I can understand that. However much of the diminishing returns in games is due to designing for the lowest common denominator then tacking effects on. Crysis 3 on low (+ ultra textures) looks better than most games and runs very well, higher settings brings less IQ for more performance hit precisely because that is they way the game was written. Bioshock Infinite would also be a good example. Some complex lighting effects/ DOF when they have things as crappy looking as a rose texture painted on a flat surface to simulate an actual rose, or the terrible baskets of apples looking like the game was made 10 years ago.

Fixing these kinds of things rather than expensive lighting and shadows dramatically increases IQ and immersion.

However these are expensive and time consuming and done to the lowest common denominator. Moving the baseline forward is the only way to get better models and facial animations. This is why I don't like the consoles too much. The weak CPU is going to end up causing a lot of games with tacked on expensive and low IQ features while not addressing the more core gameplay (physics, AI, animations, immersion, etc).

 

[tweakboy]

MS and Sony lost money from day 1 on hardware. Not sure why people hold on to that myth.

Wow nice 980 ,,,,, but CPU is a i5 ? Talk about bottleneck. Your CPU will leave you behind..

 

[zlatan]

I have wondered this, for example most games today only use 1 or 2 cores at most. Games that claim to need 4 cores actually run better on 2 faster cores, so so much for quad core requirements.

Only the PC games. But this is mainly an API+graphics driver limitation. The D3D11 API is only allow serialized writes to the command buffer, so even if the GPU commands are generated is parallel, there are some unoptimizable serial points and these totally kills the throughput on PC.

Therefore how exactly are devs planning to use 8 cores for gaming on consoles? is it not a bad design and a total waste of time, money and resources to have put in 8 weak cores in a PS4?

why not just 3 or 4 cores?

It is much easier to utilize 6 cores (2 cores are used by the OS) in the PS4, than writing a PC port that may fully utilize 2 cores. D3D11 is a serious issue for the PC. PS4 use a much more robust API called GNM. It is the most efficient API ever created. Writing commands to the command buffers is explicitly parallel. Even most of the commands are very simple. These may take up only 3-4 clock. Doing the same job in D3D11 is more than a million times slower. D3D11 doing so much unnecessary workloads like checking the resources in the GPU memory. Even if I know that all of my data is consistent in there (because I designed the engine to be consistent) I have to wait this very slow procedure.
Parallelizing the code is not a hard job when you have enough data, and in a game you have. The hard part is dealing with an API with unpredictable behaviours. It will make most of the CPU time unaccessible.

Why not let Intel build a Pentium G dual core or an i3 for the PS4? would it not have been better than 8 x 1.6 GHZ AMD cores?

Based on the workload. Personally I prefer more slower cores than less faster ones, if the throughput performance is equal. My only problem with the PS4 (and XOne) CPU design is the L2 latency when one CPU module accessing the other module L2 cache. But I know that this is slow so I'm try to avoid these scenarios.

How come AMD managed to beat intel and nvidia when it came to getting the contract for both Microsoft and Sony?

They brought an SoC for Microsoft and Sony with shared memory before the end of 2013. This simple.

 

[Erenhardt]

Hey zlatan, great post.

Could you explain the ps4 cpu resources a bit more? The 2 cores that are dedicated to background tasks (system etc) are on the same "module" leaving 4+2 available cores for game devs, or is it that on each module there is one core locked to background tasks leaving 2 modules with 3 cores each?

Also, if console CPUs have slow cache performance - just like amd cpu line, does it mean desktop amd cpus can benefit if the optimizations from console are brought to PC through dx12?

 

[zlatan]

Could you explain the ps4 cpu resources a bit more? The 2 cores that are dedicated to background tasks (system etc) are on the same "module" leaving 4+2 available cores for game devs, or is it that on each module there is one core locked to background tasks leaving 2 modules with 3 cores each?

Some SDK information is under NDA, so I can't give you a direct answer. But I can say I would choose the 4+2 config, and I'm happy with the PS4 configuration, so you can guess how it works.

Also, if console CPUs have slow cache performance - just like amd cpu line, does it mean desktop amd cpus can benefit if the optimizations from console are brought to PC through dx12?

The L2 cache latency inside the module is good. It's 26 cycles. The only problem is the cross module access. Reading something from the other module will take 190 cycles. These scenarios should be avoided.
These are very special optimizations designed for the consoles. It won't help them. But, in general, the low-level access in the PC is a good thing for AMD. It won't change the game in the CPU front, but it will "introduce" the "more core is better" era in the PC.

I think this low-level access concept will help AMD in the GPU market. They doesn't care if the program will use Mantle or D3D12 or both. It doesn't matter. What gives them an advantage is the concept of the low-level access. For now with the D3D11 the graphics driver is handle a lot of thing automatically. Filtering the states, compiling the shaders, allocating buffers under the D3D API, and so many other things. The vendors optimise these for the application and this is what they call an application profile. For example they know what instruction supported by the hardware and knowing these they can replace the original shaders with faster ones, or they can optimise the buffer allocations for each architecture. This is why you need new drivers for the new applications.
With low-level access this will be very different, because the graphics driver will be very thin. It won't be much more than a shader complier. In theory the vendors can replace the original shaders, but nothing more. For example they can't optimise for buffer allocations, because the application will explicitly handle this. The driver will see raw data, and it won't know anything about the data layout. This is good for AMD, because every developer know now how to optimise an engine for GCN architecture. I know that register pressure is a key part to get fast performance, the sampling speed parameters are public, so I can choose the right format and filter type, and there are some well known bitshifting tricks to get some extra performance, these are very useful. I even think some extreme optimization when playing Quake Live with my son in the holidays. Fast inverse square root could be a good way to speed up some effects, but this is theory now. In the other hand, If somebody ask me how to optimise for Kepler or Maxwell ... I say ... I don't have a clue. And this is not because I'm lazy. These architectures don't have good documentations, only some CUDA papers. This is why AMD want this low-level thing so badly.

 

[krumme]

Thanx very interesting perspectives. But what would you personally prefer to happen when gcn2 arives then?

 

[Shivansps]

zlatan, the mayor problem that we users are not noticing the difference, where is the result of all that work? console games so far had great problems to even hit 1080p and a few problems to maintain 30 fps, im not gona even ask 60 because that seems a bit too much for that gpu, we also see, because we are no blind, the use of dirty tricks, like playing with rendering distance and shadow quality at background.

Them there comes Star Citizen teaching a leason to eveyone about what can be done with the D3D11 API. And there is also "The Vanishing of Ethan Carter" that did just open a school about how to do a good game at 1/3 of what a AAA game needs.

So its not the first time that i heard how good is to have multiple cores and low level apis that consoles have, but its also not the first time that i see zero results, and thats what matters, the results, we all know about the problem with the D3D11 API, but even with that big problem, it does not seems to matter at all, so dont be suprised when people ask for console hardware to be more in-line with what we use here.

 

[zlatan]

Thanx very interesting perspectives. But what would you personally prefer to happen when gcn2 arives then?

Obviously that will be a drawback with D3D12 and Mantle. Every architecture will need some special optimization, and if a newer one comes than it may not as fast as the earliers. Tonga and BF4 is a great example. I don't know what is the problem, because I don't able to see the source code, but I'm pretty sure it's a buffer allocation problem, because the newest GCN IP is different in this perspective. Unoptimized buffer allocations can cause this kind of performance issues. But this is my own opinion, and it is based on my experience with Tonga.
All of these will be fixable, but not in the driver. The application will need a patch with optimized support for the newer architectures.

 

[zlatan]

Dear Shivansps, and everybody! It is important to understand that the real next-gen engines will be different, than what you see now. These are now in early alpha stage, but (we)/I'm working on it. The GNM API and the power of PS4 can lead some very new rendering options. Many CGI tech from movies (like Lord of the Rings) is possible now in real-time. We have the batch performance for it with GNM, and the PS4 use a 1,8 TFLOPS GPU, with the ability to parallelize graphics and compute workloads which was not possible before. This is a huge deal, because many people think that the GPUs are data parallel processors, which is true when we speak about the hardware itself, but knowing what possible with the GPU, and programing it is very different. The current PC APIs will allow a pipelined processing without any workload parallelization (the drivers try to do some, but it is very minimal). And this is bad for today's GPUs because these processors are designed for heterogeneous compute. Some workloads will use mainly the ROP engines, and with this most of the ALU capacity will be free, and we can't use it.
Making new VGAs for the PC with 4-5-6 TFLOPS is not next-gen, it is just raw power that we can't access with the current public APIs. My friend always joke with this. I remember when AMD launched the Hawaii GPU. Our doorman came up and said "throw away that PS4 guys PCMasterRace has more than 5 TFLOPS now". My colleague immediately respond "that's a new record Frank, you now have 4 unusable TFLOPS". The whole studio is laughing on that, but it is kinda sad situation when you can joke with this. Okay with Mantle it is possible to parallelize the workloads on PC, but what about the others?

All in all, I want to say that the new real-time engines will be closer to Pixar's Renderman. One interesting option is to shade sample points and not triangles. It is possible on a PS4, and it will give the ability to create movie quality effects. I'm expect a lot of innovation here in the coming years. "PCMasterRace" should pay attention to Oxide Games. I saw they doing this kind of innovation for PC-only. Kudos to them.

 

[Erenhardt]

We should make a dedicated thread to interview this guy! ^

Are you working on ps4 exclusive title?

 

[MeldarthX]

Zlatan thanks for the feedback which again has backed up what I've said. Most that are saying PS4 is already tapped don't work in the industry or actually work with the hardware.

Question; as I've not had time to talk with my friends work on the Xbox1 and PS4 lately; which of the next gen engines are you looking forward to working with; and will you be doing with anything with morphous when it comes out?

 

[Shivansps]

"In the coming years" seems a little too late.

In insight i hear it before, i dont want to keep hearing, i want less talk, less promises and more results, now, no when the consoles are 5 year old, if takes that long to get results its not worth it.

Meanwhile i keep seeing people doing crazy and increible amazing things with a hell of a old game engine, did you checked what egosoft did with X3:AP? that game is running on a 90s game engine that is ST!!!!! it does not use more than 1 core, im yet to see something that is 1% of that complexity running on a console.

Them i look what the folks of the community HLP managed to do with the original 1998 Freespace game engine that is also ST, "Diaspora: the shattered armistice" is running on that thing improved by amateurs!

Them i look at the PS4, problems to hit 1080p, problems to mantain 30fps, "30 fps is more cinematic", quality is less than in pc with the addon on dirty tricks, and i keep hearing how good it is to have a low end api and lots of tablet cpu cores, sorry but no, i want to see results, talking about how good it is does not cut it for me anymore.

Im sorry if i sound agressive, but im tired of this already.

 

[soccerballtux]

It's GPU was bleeding edge at the time, and it's CPU, while lacking in general IPC was an SIMD monster that in some ways is comparable to modern OoO x86 CPUs. But you have to remember that it was many devs first experience with multicore and those coming from the PC had a difficult time with the narrow cores.

cell didn't have a GPU. PS3 did, but cell didn't. and it wasn't bleeding edge, it was an NVidia 7800GTX with less bandwidth

Nope. Another myth of last gen consoles. Their CPUs were slow for games and underpowered on Day 1. Even a Core 2 Duo E6600 @ 3.6Ghz was faster than either CPU in PS3/XB360.

The entire Xbox 360 CPU was not even as fast as a single core in the Core i7 920, a CPU from 2008.

at any rate I was talking about comparing Cell to the Xenon. Xenon was much better for game development because it allowed lazier programming so long as you wrote for multicore well. Bad branch prediction? No OoOE? No big deal, since it's hyperthreaded the other thread can jump on and do stuff.

Digital Foundry: How would you characterise the combination of Xenos and Xenon compared to the traditional x86/GPU combo on PC? Surely on the face of it, Xbox 360 is lacking a lot of power compared to today's entry-level "enthusiast" PC hardware?

Oles Shishkovstov (Metro 2033 developer): You can calculate it like this: each 360 CPU core is approximately a quarter of the same-frequency Nehalem (i7) core. Add in approximately 1.5 times better performance because of the second, shared thread for 360 and around 1.3 times for Nehalem, multiply by three cores and you get around 70 to 85 per cent of a single modern CPU core on generic (but multi-threaded) code."

^ That's pathetic -- the entire Xbox 360 CPU was only as powerful as a single core i7 Nehalem at 1.86-2.26Ghz. Like I said a Core 2 Duo E6600 @ 3.4Ghz would have mopped the floor with it in games.

But people like to talk about how the GPUs in PS3/360 were more future-proof. This isn't true since the rate of GPU progress at the time was much faster than it is now.

"The 360 GPU is a different beast. Compared to today's high-end hardware it is 5-10 times slower depending on what you do"
http://www.eurogamer.net/articles/digitalfoundry-tech-interview-metro-2033?page=4

^ Just 4 years after PS3 launched, we had GPUs 10-11X more powerful.

PS3 ~ 7800GT, but for I'll even give it 7900GT ~ 16.1 VP
November 2010, 4 years after PS3 launched --> GTX580 ~ 180 VP
-----
GTX580 was 11X more powerful than RSX in PS3 in just 4 years after PS3's launch.

http://forums.anandtech.com/showthread.php?t=2298406


The idea that PS3/360 were some super powerful consoles is a complete myth. 8800GTX was 3.3X faster than the RSX in PS3 and they launched days apart. By June 2008, just 20 months years after PS3 launched, GTX280 launched with 5.75X the performance of PS3's GPU. We are 13 months out from PS4's launch and a GTX980 is about 2.5X faster than R9 265 ~ HD7850 ~ PS4's GPU.
http://www.computerbase.de/2014-09/...vidia/6/#diagramm-rating-1920-1080-4xaa-16xaf

All the facts are there but PC gamers refuse to do the legwork or they don't want to admit that PS3/XBox 360 were also weak underpowered consoles and became outdated just as fast if not faster shortly after launch.

we had a thread a while ago that looked at the CPU performance of the Xenon and Cell's PPE-- each was on the order of a similarly clocked Atom. So, the Xenon was a tri-core hyperthreaded 3.2ghz Atom.

I would very much have liked a dual core HT 3.2ghz Atom in a laptop.

It was more than enough for game logic.

 

[SPBHM]

Nope. Another myth of last gen consoles. Their CPUs were slow for games and underpowered on Day 1. Even a Core 2 Duo E6600 @ 3.6Ghz was faster than either CPU in PS3/XB360.

The entire Xbox 360 CPU was not even as fast as a single core in the Core i7 920, a CPU from 2008.

Xbox 360 is older than Core 2 Duo, high end PC was dual k8 at 2.4GHz +- when the xbox 360 was released, actually most PC gamers had single core K8 and Netburst at the time... the Xbox 360 CPU was somewhat impressive when released (late 2005/early 2006)

same for the PS3 a year later, the Cell SPEs had some respectable numbers, considering it happened before GPGPU was really a thing.

the PS4/XO CPU was underpowered (compared to mid range PCs) since the start, unlike the previous gen.