This whole thing is hilarious, it's obviously misleading for NV to compare it to that T7200 score, but it's at least as misleading in the other direction that they're running Kal-El at 1GHz while the top SKU runs at 1.5GHz (Rayfield said so himself in a comment he posted on the NVIDIA blog). I'd tend not to take conspiracy(-like) theories too seriously because of that. It's probably just a very foolish mistake.
Tegra 3 officially announced; in tablets by August, smartphones by Christmas
- Thread starter Mike11
- Start date
It isn't a conspiracy if it is blatantly posted on the slide anyway. That means it is cited. It is misleading for those that are not aware of what it means, but technical presentations are full of things that could be misleading for people who have no idea what is going on. If it is cited correctly that is fine. IT could be underhanded to try and influence people that don't know about it and those that don't want to waste time looking it up, but it is at least as humorous to see the frantic arm waving and crying out in the technological desert by the prophets trying to protect us from marketing.
So it's okay for you to blatantly lie about something (their claim that the benchmark demonstrates that Tegra 3 is > Core 2 2GHz level) so long as you display in fine print that you're lying?
The "conspiracy theory" as it was put isn't so much the details of the legitimacy of the comparison, but whether or not nVidia had any connection with the posting.
I agree that putting the details of the score's conditions in the slide helps, and I'm sure they put it in there to cover their asses. But they must have hoped no one would notice, or that somehow no one would care. How they could believe something this silly is completely beyond me.
It's actually not a blatant lie because it spells out right there on the slide what is actually being compared.
The slide mainly serves to demonstrate what kind of performance class the device is in. Sure, the comparison is flawed, but these over the top indignant hissy fits about a detail don't detract from the fact that it's damn impressive. A slightly slower Core 2 would still lose to the Tegra 3 even when using the same compiler version and flags. It's just not that important, is it?
Please understand, I'm not saying that slide is a lie. I'm saying all of their further claims that Tegra 3 performs better than that Core 2 made from that point onwards are lies.
It wouldn't take a slightly slower Core 2, you'd have to go a good 50+% slower. It's pretty important. Personally I don't think it's that impressive, it's exactly where you'd expect a Tegra 2 to be if you doubled the core count. Of course, this is on a benchmark that has perfect core scaling and is not impacted in the slightest by the memory subsystem or even L2 cache performance, not to mention entirely integer.. not exactly representative of an awful lot to begin with.
It wouldn't take a slightly slower Core 2, you'd have to go a good 50+% slower. It's pretty important. Personally I don't think it's that impressive, it's exactly where you'd expect a Tegra 2 to be if you doubled the core count. Of course, this is on a benchmark that has perfect core scaling and is not impacted in the slightest by the memory subsystem or even L2 cache performance, not to mention entirely integer.. not exactly representative of an awful lot to begin with.
If somebody interested in X86 vs ARM comparisons, then here's a few of them:
7-zip and memlat tests
square root execution comparison(pure synthetic)
wav to ogg conversion(oggenc)
povray 3.6.1(multithreaded rendering disabled, it's only for one core and HT probably didn't work too)
Anand's sunspider atom result(looks like win + IE)
Anand's XOOM result(sunspider testing browser's javascript engine, so it's heavily dependent on browser itself)
It's clear that on same frequency Atom often is slower than A9, at least in cases when code is math bound. It would be interesting to see more of such ARM vs X86 comparisons
7-zip and memlat tests
square root execution comparison(pure synthetic)
wav to ogg conversion(oggenc)
povray 3.6.1(multithreaded rendering disabled, it's only for one core and HT probably didn't work too)
Anand's sunspider atom result(looks like win + IE)
Anand's XOOM result(sunspider testing browser's javascript engine, so it's heavily dependent on browser itself)
It's clear that on same frequency Atom often is slower than A9, at least in cases when code is math bound. It would be interesting to see more of such ARM vs X86 comparisons
C'mon, It doesn't prove anything. There were not so recently submitted results of AMD Athlon X2 5000+ 2.6GHz, Intel Pentium M 1.30 GHz 1.30 GHz, Texas Instruments OMAP3430 600MHz, ARMv7 Processor rev 3 (v7l) 600Mhz, AMD K6-2 500MHz, etc. Are they on Nvidia's behalf as well?:smile:
Last edited by a moderator:
As well it should be clear that A9 is better than Atom clock per clock.. what does this have to do with Core 2?
ARMv7 is not ARM7, it's the ISA high end ARMs (Cortex-A8, Cortex-A9) are using today.
I said more than just when they were submitted. The first two were submitted by BSN, a news site, so it's not hard to imagine why - comparing things is their job. OMAP3430 isn't actually especially old/still in use, likewise "ARMv7", and are probably used in conjunction with research or simulation done on those platforms. I dunno about the K6-2, but when you start seeing any of these used in slides by companies then yeah, there'd be reason to be suspicious.
I just can't imagine why this consulting company uploaded this score, except on nVidia's behalf. It isn't necessarily the case at all, it just strikes me as the most likely explanation given limited information. You just happened to say that this score has nothing to do with nVidia, so can you prove that instead?
This is just a typical marketing trick. NVidia can claim they where being open as they included detals of test conditions and at the same time NVidia know that the non technical decision makers/execs will just look at the headline numbers and be wowed while being oblivious to the "trick" that's being played on them.
Marketing marketing marketing, data for dummies, smoke and mirrors, executive flash bulbs, out right lies etc etc call it what ever you like it's just the usual corperate massaging of reality
Marketing marketing marketing, data for dummies, smoke and mirrors, executive flash bulbs, out right lies etc etc call it what ever you like it's just the usual corperate massaging of reality
According to the Roadmap, Wayne is 2 times as fast as Kal El, so any guesses as to which cpu they're gonna go with. To get 2X performance using the same Cortex A9 cores they're gonna have to run them at 2 Ghz, which might be too high for 28nm, at least initially. Cortex A15 at 1.5 Ghz sounds like a much better fit. But will a Cortex A15 design be ready that early? (ie early 2012). TI which is the lead licensee for Cortex A15 has announced OMAP 5 availability for H2 2012. Should be the same timeframe for Wayne as well so im guessing it is possible, especially given Nvidia's track record with Tegra 2 and 3 so far.
Wayne, assuming that's referring to Tegra 4 and not Tegra 3.5, will probably not make it into end product until the latter half of 2012.
A15 could be doable. I'm not sure whether nVidia originally targeted quad A15 or not, but they'll have to back off their aggressive CPU:GPU performance ratio soon if they want to remain competitve in phone SoCs. They'll need to consider substituting either dual A15s or quad A9s and a nice clock.
A15 could be doable. I'm not sure whether nVidia originally targeted quad A15 or not, but they'll have to back off their aggressive CPU:GPU performance ratio soon if they want to remain competitve in phone SoCs. They'll need to consider substituting either dual A15s or quad A9s and a nice clock.
Thats interesting considering that zacate is 75mm2 on 40nm .
I wonder what would perform better
Power efficiency reduces area efficiency to a certain extent, so it's obviously not that simple - and Zacate requires an external southbridge. Anyway, Kal-El might be faster on the CPU front for quad-threaded workloads, but it should be slower for single/dual-threaded ones, and the GPU is significantly slower as well.
Yea H2 2012 is the earliest we can expect Wayne. It should follow a similar schedule as Kal El.
I think this has been discussed in one of the threads, whether dual A15's would be more efficient compared to quad A9's. Im not sure what the final outcome was
You're right Arun, it would be ~3X at max (twice the number of cores X 1.5 times the clock i suppose). So yea if Wayne is actually going to be less than twice as fast as Kal El, they've probably stuck to the same quad A9 config. I think they're missing a trick if they arent developing a dual A15 config as well though. I dont see smartphones adopting quad cores as quick as tablets.
Zacate for sure. Bobcat is significantly faster clock for clock than Atom. And we know that clock for clock, Cortex A9 is about the same or a bit faster than Atom. Also Zacate GPU performance is an order of magnitude faster than any SoC shipping today. You cant really compare the die size though. Zacate is far from being a complete SoC.
80 SP's. Its basically a Cedar GPU
Ok then; NGP isn't shipping yet and despite it having "only" 64"SPs" it can show some of its muscle against such a GPU (as long as there aren't any DX11 applications).
D
Deleted member 13524
Guest
Not with Zacate. That one has its GPU clocked at 500MHz.
But Ontario (the 4-9W version) has its "Cedar" clocked at 280MHz, so it should actually be comparable-ish.
Nonetheless, the NGP release should be just a couple of months away from the release of the next-gen low-power APUs on 28nm.
According to wikipedia, 4-9W Ontario will be replaced with Wichita, which has 1 or 2 "updated" Bobcats and a GPU with the performance of a "current Zacate GPU", so GPU performance will be approx. doubled from current Ontario.
The 18W Zacate will be replaced with Krishna, with 2 or 4 "updated Bobcats" and a GPU with twice the performance of the current Cedar @ 500MHz.
My guess is that AMD will use the same new GPU in both low power chips (Wishita and Krishna), and again will just clock it differently between them.
It's also a good guess that the new GPU has 32 vec4 units (128sp), while keeping the same clocks as the current 40nm solutions.
AMD is trying to push GPGPU hard with the APUs, it's pretty logical to think they'll use the new computing-friendlier architecture introduced in Cayman.
In other words, if the NGP ever gets to "compete" with AMD's low power offerings in pure performance, it won't be for long.
Cortex A15 + Rogue solutions should be a whole other story, though.
2012 is the year when x86 and ARM solutions will really clash together in apples-to-apples comparisons.
