PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)
- Thread starter Love_In_Rio
- Start date
- Status
I don't think any developer that has had to care about memory bandwidth in the slightest in their career would have taken the peak theoretical number as anything but a theoretical peak.
Within the field, giving the peak wouldn't be misleading because they've been paying attention to DRAM bus utilization since DRAM was invented.
D
Deleted member 11852
Guest
Exactly, so the likelihood is, given math is kind of a thing in programming, that Sony gave 'average' usage scenarios.
Xbox has the same issue with memory contention:
Significant performance degradation from DRAM contention.
Prioritize DRAM for CPU.
Current record holder for ESRAM as of June 2014 is 141GB/s, despite the engineers in the Eurogamer interview wording it to sound like average ESRAM bandwidth is 140-150GB/s
Page 9
http://www.slideshare.net/DevCentralAMD/inside-xbox-one-by-martin-fuller
Significant performance degradation from DRAM contention.
Prioritize DRAM for CPU.
Current record holder for ESRAM as of June 2014 is 141GB/s, despite the engineers in the Eurogamer interview wording it to sound like average ESRAM bandwidth is 140-150GB/s
Page 9
http://www.slideshare.net/DevCentralAMD/inside-xbox-one-by-martin-fuller
Last edited by a moderator:
Xbox has the same issue with memory contention:
Significant performance degradation from DRAM contention.
Prioritize DRAM for CPU.
Current record holder for ESRAM as of June 2014 is 141GB/s, despite the engineers in the Eurogamer interview wording it to sound like average ESRAM bandwidth is 140-150GB/s
Page 9
http://www.slideshare.net/DevCentralAMD/inside-xbox-one-by-martin-fuller
Also, water is wet.
They didn't word it like that at all. And if they had (they didn't) it wouldn't have made any sense anyway.
The 141 GB/s was from a real title, before the release of the June SDK. The figure now would likely be above the 140~150 GB/s peak that the architects had seen from the hardware last year.
Neither MS nor Sony engineers seem to be bullshitting about the hardware, it's a shame that console warriors are trying to infer that they're operating like, well, console warriors ...
My mistake I simply quickly skimmed the article and saw them state 140-150GB/s over and over.
You are accusing me of being a console warrior? I sense a little console warrior in you....
No they didn't.
One guy said it twice in detailed response to one question specifically about the esram. And he was being truthful.
You've misrepresented what the Xbox architects said, two posts in a row, both times in a distinctly negative way, in a frikkin PS4 thread. It's not hard not to do this, and it doesn't help the thread in the slightest.
Not intentionally. I kept quickly glancing at the interview and thought while they were talking about 204 being the raw peak spec, saw 140-150 a bunch of times, and mistakenly thought they were suggesting that the 140-150 was some average that most developers would attain in most scenarios. They aren't saying that. You are totally right the Xbox engineers are not bullshitting and being professional.
Last edited by a moderator:
No it's not!
Water is the element that create wetness, but it's not wet itself.
This is true, except that water is not an element.
Fair enough. I guess it's an easy enough mistake to make when ploughing through detailed stuff. I've certainly done it enough. I feel like a bit of an asshole now.
It's a fair point to make that the Xbox suffers similarly with CPU access too. A good reason to keep render buffers out of main ram. No doubt Sony were well aware of this when they plumped for this design, hence the fat bus that can edge out a 7850 or even a stock 7870.
So yeah, I think Sony/AMD understood this BW contention issue and planned around it. Certainly interesting, but not something to worry about.
Technically you'd be right, but both words predate the discovery of elementary particles, so their definition should reflect their roots. "Water" and "Wet" in this context are the classical element, and the sensible quality, respectively.This is true, except that water is not an element.
Except if this 141GB/s number (which was not an average measured during a whole frame cycle, but a peak number so probably measured during a very short time) was only measured during one short fully allocated GPU time slice. I am not sure If I a am clear enough: we know the 10% kinect reservation was a time slice reservation.
For instance if they did a 1ms kinect reservation every 10ms, giving Kinect 10% of the GPU and esram bandwidth (those numbers are theorical), then they could have measured 141GB/s peak bandwidth during one specific tasks processed only during the 9ms of the 100% GPU available during the game 90% slice.
In that hypothetical case, the june SDK would obviously improve any total average bandwidth but not any peak bandwidth measured during only the 9ms hypothetical time slice.
I am not saying it's the case, but as long as we don't have any more details, that's a possibility.
Define "outdated."
It was designed to appropriately pair with the APU; the hope is that it winds up being able to feed the CPU and GPU reasonably well in any reasonable use case. That is, although more size and speed would be nice, the hope is that it never becomes a disproportionate bottleneck of its platform.
For overall size, 8GB of GPU-accessible space (albeit shared across both processors) is still massive, though the larger GPUs are getting there.
As far as bandwidth is concerned, larger GPUs can of course make use of more data. Some are already being fed with over 300GB/s.
- Status
