-
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.
Kotaku: "Only 90% of Xbox One GPU usable for games"- PS4 GPU up to 66% more powerful?
- Thread starter Gemüsepizza
- Start date
Possible? Yes. But it would have been a huge waste of money because it wouldn't add anything to the system. 3 GBs of reserved RAM is a lot but 12 GBs would still be excessive. The problem is the type of RAM, not the amount. And the RAM type also means they're stuck with a comparatively crappy GPU since anything better would be starved for bandwidth. Hell, the GPU they have now will be bandwidth starved unless the ESRAM is utilized really well.
Windom Earle
Member
Great post, couldn't agree more!
We really dont have exact PC equivalent of PS4 gpu. It sits right in the middle between 7850 [16CU] and 7870 [20CU].
DeFiBkIlLeR
Banned
Whilst the numbers are pretty meaningless in context of console optimisation, the relative performance difference of the two GPU's is very real.
Whichever way you cut it, the PS4 GPU is always going to be able to do more.
MS have decided their GPU is 'good enough', going down the Nintendo path, Sony have decided we need 'as much as we can give you'.
I know which one I prefer.
velociraptor
Junior Member
So we may well see 30fps vs 60fps games then.
It's closer to the 7850 than 7870.
velociraptor
Junior Member
Is the PS4 GPU more powerful than the 7850?
Melchiah
Member
Yes, a little.
Lagspike_exe
Member
They've lied before.
Goddamn, X360 is beating even the PS4. The true RAPID SPEED RAM.
nelsonroyale
Member
The only standard feature that the PS4 has over either is bandwidth I think.
Melchiah
Member
I would have loved to see their reaction, if someone had confronted them with that slide in the panel.
Its not only that. PS4 GPU has 2 more Compute Units than 7850, and it has different architecture for direct compute [8 schedulers with 8 pipelines each, standard is 2-2], streamlined communication path between GPU-CPU.
They've lied before.
Lol is this graphic really posted to MS? My God 360 seems next generation here.
nelsonroyale
Member
True, but I was indicating that the bandwidth is also higher compared to the 7870 as well.
Again, no. PS4 has 176GB/s ram bandwith that must be shared between GPU and CPU. [7850 130GB/s, 7870 175 GB/s]
Sony choose their bandwiths to perfectly eliminate bottlenecks. PS4 GPU will use ~150 GB/s, which will leave great amount for octacore jaguar [~30 GB/s].
fatgamecat
Member
I'd like to know the visual or real world performance difference when it comes to multiplat titles. Guess we'll be waiting for eurogamer's famous comparisons.
DeFiBkIlLeR
Banned
Well, the difference will be down to a developers willingness to exploit the advantages on offer to them with the PS4..some won't.
Actually according to wiki both 7870 & 7850 have access to 153.6GBPS. Also the CPU in ps4 is said to use <20GBPS most of the time.
The 7770 shown does not have this same hardware architecture, some low level bits are different,and there is no ESRAM...
In other words, it's a bit misleading..
Sebbi on B3D said it's more likely that resolution would drop rather then framerates based on looking at it from their architectures.
We all know the PS4 has the potential for a sizeable lead,but the school yard graphs and simple numbers comparisons is sad.
Melchiah
Member
Yes.
EDIT: Read "to" as "by". I dunno if anyone has rubbed it to their noses.
http://mobile.pcauthority.com.au//Article.aspx?CIID=23155&type=Feature&page=4
They sure like to add numbers together.
Windom Earle
Member
Why is it sad? Given that the PS4 gpu will also be faster than the one in the graph, this should be a decent indication of performance differences.
Gemüsepizza
Member
It's GCN. And it needs eSRAM because it has DDR3 RAM instead of GDDR5 RAM.
What's "more likely" is up to developers. They decide if the use the increased hardware power for higher resolutions or for higher framerates. In both cases the GPU has to render more pixels.
77xx and 78xx are not created from same architecture "bits".
Xbone GPU is 7850 without 4 CUs, and PS4 GPU is 7850 with two additional CUs. Presence of ESEAM changes almost nothing for Xbone, it just saves it from slow DDR3 main pool speed.
JimiNutz
Banned
Will it really be that difficult or time consuming to harness the extra power?
If they are similar systems anyway but one is just a bit more powerful won't it be easy for them to make the PS4 one look slightly better - especially if these games will be coming to PC too where they will look better anyway....
Windom Earle
Member
Has this been confirmed?
Melchiah
Member
That's not the worst one:
Holy Hell! I had forgotten that one. Their bullshit knows no limits.
Yes, all old VGleaks/Digital Foundry rumors were 100% spot on [except crazy GDDR5 ammount]. Both Xbone and PS4 use standard GCN architecture from 78xx series, with Sony having few specific changes inside GPU to make it better for gaming.
Given the triangle set up that seems like the only logical option.
Windom Earle
Member
Makes sense, this should indeed put them a bit closer together than indicated in the benchmark then. (Which has more like 90% difference than 50% for some games)
DeFiBkIlLeR
Banned
One of the Xbones lead designers let slip (I'm guessing he wasn't supposed to) that it has 768 stream processors.
..current AMD designs have 64 SP's per CU.
I'm not a techie so can someone explain why they went with 102 GB/s ESRAM and not the EDRAM of x360 which seems to have huge bandwith?
Expense.
Kung Fu Grip
Banned
Didn't the rumors put XB1 more in line with the 77 series? First I'm hearing it's similar to the 78 series like the PS4.
From what I remember the Xbone has a down tuned 7790, While PS4 GPU is a down tuned 7870. This is first time I'm hearing they're both from the 78** series.
Gemüsepizza;59553621 said:
Sebbbi gives some explanation, as a developer, what's involved going from 30 to 60, and which bits of the hardware need double the performance, it's not exactly as obvious as you think.
Here's the thread for anyone interested
http://forum.beyond3d.com/showthread.php?t=63566&page=13
Someone mentioned the 7770 only does 1 triangle per clock and the xb1 can do two.. Not sure if true, but these are both custom parts...
Anyhoo, I'll no doubt buy both, but I'm preparing myself not to have this almost 100% difference that people are suggesting we'll be seeing.
Punkster_NY
Member
hahahahaha
I know you are also dismissing the cloud bullshit, but even your example is not feasible to be offloaded to cloud: What about the first 5 minutes of play, until you can download the first pre-rendered data? Some data will have to be on the disc / already downloaded in this case, calling for the fact that all the sequential data could also be put on disc.
Having said that, if you are really going for static lighting (static lighting being a next gen feature
), you may as well use precomputed radiance transfer, which supports dynamic lighting. It would be rendered only once, put on the disc, and it allows multiple dynamic lights on static geometry, different time of day, all those bollocks!TOAO_Cyrus
Member
Technically this and the bandwidth figures are true. Too bad the CELL's SPE's more then make up for its single general purpose core and the vast majority of that bandwidth can only be used to access a small frame buffer.
AgentP
Thinks mods influence posters politics. Promoted to QAnon Editor.
I think the real bandwidth between the GPU and eDRAM was 32GB/s. The 256GB/s number is the internal bandwidth that assumes compression.
http://beyond3d.com/showthread.php?t=60901
Melchiah
Member
Technically true isn't really the same as applicable in all circumstances.