PowerVR Series 6 now official
- Thread starter roninja
- Start date
No. There were quite a large number of engineers who worked on CLX.
I've missed that d00d; where is he trolling nowadays?
The Series6 performance claim of that Pocketgamer article, where Pottsey appears to have correctly sourced the 100x comparison as being related to MBX rather, is now spreading across websites in all its mistaken and/or misinterpreted glory.
And for better or worse, IMG and ST Ericsson have made GFLOPs with too little context into the new polygons per second marketing psuedo-metric.
And for better or worse, IMG and ST Ericsson have made GFLOPs with too little context into the new polygons per second marketing psuedo-metric.
http://blogs.arm.com/multimedia/400-triangles-per-second-performance-metric-or-chocolate-teapot/
I guess a blog entry for GFLOPs could be expected soon.
I won't say that the author doesn't have a point with many awkward theoretical peak numbers that circulate in the embedded space lately, but on the other hand it wasn't and isn't that much different in the desktop space either.
A couple of things though:
1. IP licensing companies such as ARM do not and cannot control how their partners will market their products. In one hand we have Ed Plowman's objection against effective fill-rates and on the other hand Samsung listing the Mali400MP fill-rate at 3.2GPixels/s. As I said before I can understand Samsung since they listed in the S5PC11x 1.0GPixels/s fill-rate. If IP licensing companies will manage in the long-run to get their partners to consult them before marketing any piece of IP for real-time capabilities I'd personally welcome it without a 2nd thought. At least that way we won't see 96M Tris/s for stuff like Samsung Wave.
2. I've seen in various occasions ARM employees disregarding public synthetic benchmarks like Kishonti's GL Benchmark. No idea if there's some legitimate objection behind it or if those were singled out cases. However if there are such objections companies could drive game ISVs to include time-demos in their games and try to drive websites that deal with embedded GPU benchmarks to use those also. Even if, I'd still sense objections even there, since we're seeing it over and over again in the desktop space for years now. Another alternative would be several IHVs to try and reach an agreement to create together a suit of synthetic benchmarks, but guess what if if they'd try it would be damn hard for them to reach common ground.
3. An alternative to triangles or polygons per second would be a geometry complexity rate of how many triangles per frame can be processed at 30 or 60Hz. ARM isn't revealing anything about Mali T604 performance metrics apart from:
I assume here from the wording that a single T604 is up to 5x times as fast as a single Mali400. If my interpretation is correct, then a T604 MP4 should be up to 20x times faster than a single Mali400.
Where and how those equally meaningless performance ratings are in any way different compared what any other of their competitors are claiming is beyond me.
Honest question: which GPU has trouble rendering Angry Birds? I haven't done an extensive search on the web, but this one was the first that popped up: http://www.pcworld.com/article/211152/angry_birds_devs_angry_at_android_fragmentation.html
Oh and to avoid misunderstandings I don't take the above blog entry necessarily as a "blow" against IMG in particular; at least I don't read it that way. Graphics IP providers could IMHO work together if they wanted to try and level performance ratings a bit or at least drive the press to measure real game performance increasingly instead of only concentrating on theoretical peak numbers or purely synthetic benchmarks. And yes Q3a is getting benchmarked occasionally, but it's also fairly outdated for the bulk of OGL_ES2.0 GPUs out there.
Have to agree with much of what Tom has written, the mobile space in particular has been plagued with an obsession with polygon throughput even though it is a completely useless metric by itself. A better metric here would be how well you manage to run vertex programs and I think Mali 400MP4/Orion GLBench2.0 results are a good example of what can happen when you don't scale this along with rasteriser throughput.
The same can be said of GFlops, although as a rough guide peak giga flops isn't a bad measure of how well shader bound code will run assuming that the achitecture is capable of hitting a reasonable proportion of that peak in general shader code.
You could also add shader instruction throughput to this list.
Ultmately application performance depends on the acheivable low level metrics by varying proportions, so no one metric is really useful by itself...
John.
The same can be said of GFlops, although as a rough guide peak giga flops isn't a bad measure of how well shader bound code will run assuming that the achitecture is capable of hitting a reasonable proportion of that peak in general shader code.
You could also add shader instruction throughput to this list.
Ultmately application performance depends on the acheivable low level metrics by varying proportions, so no one metric is really useful by itself...
John.
It's a call to all of you involved one way or another with GPUs in the embedded space; it would be a damn shame if things would end up as ridiculous as in the desktop space. I realize it won't be easy, but on the other hand OGL_ES (especially given how "young" it is as an API) is far less a mess than OpenGL used to be years ago.
When representatives from IHVs join the same table for an API draft for instance at the Khronos group, it would be an idea to propose something that would work for everyone or at least the majority.
My humble 2 cents.
When representatives from IHVs join the same table for an API draft for instance at the Khronos group, it would be an idea to propose something that would work for everyone or at least the majority.
My humble 2 cents.
Second announced licensee for Series/Rogue is MEDIATEK:
http://www.imgtec.com/corporate/newsdetail.asp?NewsID=625
http://imgtec.com/News/Release/index.asp?NewsID=628
http://www.imgtec.com/corporate/newsdetail.asp?NewsID=626
http://www.imgtec.com/corporate/newsdetail.asp?NewsID=625
Further SGX/VXD licenses (I'm too bored to open another thread for those): Fujitsu Europe and LG Electronics:
http://imgtec.com/News/Release/index.asp?NewsID=628
http://www.imgtec.com/corporate/newsdetail.asp?NewsID=626
Total n00b question here but wouldn't those performance statistics put it in the same league as an Xbox 360 with DX10 support? In a phone???
Probably. But mem bw isn't going to be anywhere near console level.
That gpu has more bogo-flops than my laptop's gpu.And my laptop has a nv gt 330M. That CPU comes awfully close to my laptop's cpu, which is an i5 clocked @2.26GHz.
I think my laptop just got pwnd by a freaking tablet.
I'm new to this space but I didn't think an A15 came even remotely close to Nehalem on a core for core / clock fo clock basis.
And the i5 has twice as many cores. So unless I'm missing something the a dual A15 at 2.5 Ghz wouldn't be anywhere near an 2.26Ghz i5.
which is why i said bogo flops.
I never expected Mediatek to be amongst the Series6 licensees; albeit it remains to be seen when they'll introduce a SoC containing it. I have the weird feeling that IMG is slowly announcing the small fish for Rogue first and leave the bigger surprises for later on.
It's fairly common knowledge that PS3 has trouble with AB....no joke.
D
Deleted member 13524
Guest
LOL... terrible programming?
PR from IMG today regarding DirectX support, contains confirmation that Series6 will support DX11.
"Cores in the POWERVR Series6 family, which is codenamed ‘Rogue’, support from DirectX 10 up to DirectX 11"
http://www.imgtec.com/News/Release/index.asp?NewsID=631
"Cores in the POWERVR Series6 family, which is codenamed ‘Rogue’, support from DirectX 10 up to DirectX 11"
http://www.imgtec.com/News/Release/index.asp?NewsID=631
With Series5 already ending at DX10.1 for the 545, that comes hardly as a surprise.
