blazorthon :
You seem to have misunderstood much of what I said.
I said that Tegra 4 doesn't beat Snapdragon 800 in my last post, not that Snapdragon 800 beats Tegra 4. By my math, their CPUs should be fairly similar in performance.
I never said that the process has no demand. Quite the opposite, in fact. I said that supply is far more limited than demand and that is true.
It could be more expensive, again, because of that process technology. It is known for poor yields and is in very high demand while having little supply. Using older processes is often cheaper at least until the new one matures despite the new one being able to make smaller, less power-hungry versions of any given chip even before it matures.
I never said anything about anyone having a die-size advantage over anyone, so I really don't know why you spent so much of your post about it.
I don't know why you're going into graphics because I repetitively said that I didn't know about how they compared in it and made no claims about it.
I also went out of my way to call T3 not junk, but rather just not particularly impressive in many ways.
You said:
"Also, it doesn't matter what form or process node we look at T3 in, it still wasn't really impressive against Krait competition and a few others. Yes, the die-shrinks and such improved it a lot, but they still didn't completely fix it and besides that, it would have probably been more expensive than you seem to think to use the latest die-shrink of it. It's still a large SoC and it's on a process technology that is in much more limited supply than there is demand for it."
Tegra 3+ is 28nm. I thought all top phones were 32nm (some samsung/apple) and 28nm (all others now). Demand is at an all-time high for 28nm yes, but I'm not aware of a shortage in supply now. Underestimated product sales maybe, like nexus 10/4 sold out in hours, but not a supply issue here (google miscalculation I'd say for both, but all in stock now). TSMC fixed yield issues AFAIK on 28nm quite a while back and these are far less complicated than a Kepler etc & is cheaper to produce and deal with (heat/watts/battery etc) than T3 regular 40nm. I don't understand your "large soc" size comment or your expensive vs. what I seem to think remark, with 28nm being mature & far smaller than a 40nm T3. I don't see many phones that use 28nm that aren't in stock. IF your statement is true shouldn't a LOT of stuff be out of stock on 28nm socs? With your quote, how can I NOT talk about die size/cost of the soc? You called it large, and in many ways said it's expensive (and now commenting on yields too), thus causing a die size/cost/yield discussion here
http://www.electronicsweekly.com/blogs/david-manners-semiconductor-blog/2012/08/tsmc-gets-28nm-yield-up-over-8.html
http://www.pcmag.com/article2/0,2817,2411985,00.asp
That's clear back in Aug 2012, Yield up over 80%, & pcmag from Nov 2012 (UMC still has issues though, but heck, only 15% of their sales are even 40nm & 5% 28nm - they suck?). There are others that have written yield issues are over at TSMC. Digitimes notes TSMC lacked capacity in 1H2012 only not 2h2012. I'd say it was mature in Aug 2012, likely better in Nov2012 as shown with orders returning (and we're 3 months later now past that pcmag article).
The 40nm T3 is 80mm, same as T4, already not a large soc in either case. The 28nm T3+ is smaller than both (as I noted before) and easier to integrate as such. I don't think I misunderstood what you said when you called T3/T3+ a "LARGE SOC" either. The Qcom S4 chip is bigger than T3+ (maybe not T3 40nm but even that isn't proven), T3+ is NOT a large soc, it's smaller than all chips mentioned by me and even old T3 is only 80mm thus quite small compared to exynos 5250 (nexus10), A6, A6x and I'm guessing S4 pro also (you have any data showing S4 pro is SMALLER?). I can’t find any top chip less than 80mm so how is tegra 3@28nm a large soc? You mention cost/expense for the 28nm chip & die size comes into cost figures and directly relates to how much you make when hawking your chips. T4 is cutting cost by cutting die size (just like T3 did), which with the modem should finally allow enough unit volume to make money...LOL. Small dies enables you to sell your chip cheaper (which will be even cheaper when they ship T4 in July to OEMs) with likely better yields months from now (~90's instead of 80's?? shield soaks up all the first runs), making the phone margin either bigger or just allowing a cheaper phone to be produced and sold lower. I didn't consider the older process being cheaper here since I knew 28nm is mature @TSMC as shown which is where all T3/T3+ chips come from AFAIK, but you are correct it can be more expensive when on a new process (but not here/now for T3 vs t3+ @TSMC). I believe NV delayed 28nm to make sure they got good enough yields/volume for nexus 7 etc. At the time they shipped T3 to google, TSMC was having 28nm issues for 6/11 OEM's (ouch). Qcom etc went 28nm for quad (which delayed them) and missed nexus7, as design work started for nexus 7 Jan2012 with mass production in may'12. Maybe you weren't aware of the yield changes in the last 8 months at TSMC
Responding to Aicom you said:
"S4 Pro is a faster CPU IIRC. IDK about how the graphics compares and won't comment about it.
Nvidia, like I said, is getting better, but they're still going to be a little behind."
& more "but they seem like they'l still have a little room to make up, at least in CPU performance, to be the best"
& another post:
"Uhh, no... T4 isn't supposed to be out for like six months, yet it's already not as fast as some of Qualcomm's latest. Nvidia is improving, but as usual, they're staying a little behind in technology."
& one more & this is the latest comment:
"My bad, it's the newer top Snapdragon 800 version that Tegra 4 doesn't beat in CPU performance, thanks for the clarification. Regardless, I seem to have given Nvidia too little credit with Tegra 4."
Not quite sure how NV can be behind (or tied) when neither S800 nor T4 is out so that statement can't really be proven either way yet. But OK...I don't see you clearly saying they're tied here anyway (until just now in the post I'm replying to), and even then based on what data? Everything I have found shows A15 should win everything in CPU.
http://www.anandtech.com/show/6425/google-nexus-4-and-nexus-10-review
Even the Tegra3 does OK and this is the OLD one in OneX on 40nm not 200mhz more 28nm T3+ (I think the one they benched runs 1.5ghz). Tegra3 is beating Nexus4 in sunspider, browsermark, Octane, & Kracken. Tegra3 beat Nexus 4 in ALL 4 CPU benchmarks and this chip is not DIE SHRUNK to 28nm running another 200mhz more (this says even T3 is rather good and grey will be great at 2.3ghz with A9's). Vs Qcom it seems fairly impressive in cpu (gpu got beat quickly). Exynos in Nexus 10 shows the power of the A15 DUAL over Qcom's QUAD tech in Nexus 4. Nexus 4 was near the bottom of every cpu benchmark so how could this be anywhere near TIED with a T4 A15 quad (or any A15 quad?)? What math is being used to extrapolate a tie here? Clearly some lackluster perf is due to unoptimized software as anand noted for Qcom but the dual A15's still smack around all snapdragons in cpu here. This is a november 2012 article here, so if you're still not getting your software optimized by now and you lose in everything you couldn't claim snapdragons are on par in cpu even vs. dual here and certainly not vs quad A15. Even so the dual vs. dual contest shows A15 vs Qcom that IS optimized and still loses by large margins. This is why I expect all A15's to win vs. Snapdragons when quad vs. quad comes later this year (Apple A7 or whatever they call it if A15 based, octa or T4 etc vs. S800). In some cases the dual A15 was doubling the dual snapdragon. You could optimize and consider 100% improvement for 4 cores over their own 2 core and still lose some benchmarks vs. the dual A15 (totally discounting quad A15 here). Kracken and Octane both show this. The A15 dual is over double Snapdragon dual. So even a quad would barely catch the dual in both if software was 100% efficient, which we see it isn't. I'm not quite sure where you're looking to say any A15 would lose core to core vs. Snapdragons, they do not appear even close to A15 in cpu. If you have links to benchmarks please provide them.
I said cpu's wouldn't matter as they'll all be pretty much the same, @~A15 quality - which is why I mainly discussed GPU differences (as I believe I have a very good idea of how T4/A330 etc will perform in gpu using what we know already) but after looking at all the data it's seems I was very generous saying that about CPU. Sorry if that seemed like a wild tangent to you but I thought making my case clear on gpu's was more important than any comment I thought I'd make on cpu's. Discussing cpu's was kind of pointless to me before
Maybe I was just giving you/others GPU data so you could comment on the benchmarks shown, or for the person who would have liked to see a comment on GPU here
I usually provide ample links to data to prove the point I'm making, generally speaking too many...LOL - I'm not a fan of being called vague (and want people to look at the data themselves, rather than take my word for it, then comment). As noted the gpu's will be what makes you stand out this year but it won't look good if Snapdragons lose every cpu benchmark. QCOM & others agree on gpu being the differentiators (hence my comments) and they're moving towards gaming/gpu being the centerpiece (wisely):
http://money.msn.com/business-news/article.aspx?feed=OBR&date=20130228&id=16180571
Hired 35 game devs (news from 28th), to show their chips prowess (NV has been doing this since tegrazone started). Having said that, I don't see how NV (or others on A15) can be behind krait 300/400 when krait is slower than A15 as aicom noted in the link below and the benchmarks of nexus 4/10 show this clearly unless you have proof of the reverse? Blamed on software or not, nexus 4 got killed in everything on cpu.
https://en.wikipedia.org/wiki/Krait_(CPU)
A15 dual 5250 is faster than any krait shown here and this is BEFORE the 28nm shrink that looks like an extra few hundred mhz to 2ghz vs. 1.7 max now for Samsung's A15's, making this worse than now vs. S600/S800 later. So if you're ahead in both cpu/gpu how can you be behind or tied at all now or later this year? I'm confused. Even if cpu's are considered a wash (it's not a wash as shown) it should be dominant over A330 in gpu (making qcom the loser here). All A15 based chips should be in front of similar clocked Krait quads. I would agree it's perf is closer to A15 than A9 (but not until optimized better, as T3 is A9 and won all the cpu benchmarks at anand vs. S4pro), but it's below A15 unless I'm missing something in the data
😉 I'm sure we'll be able to find a winning krait case, but by and large I doubt it. Again, if you have anything at all to prove krait faster than A15 please post it. I'll read it. I haven't found anything showing A15 beaten or tied by krait yet.
From Anand:
"Overall, the Nexus 10 results show us some real promise for what we can expect from ARM Cortex A15 based SoCs. The potential upside to this new architecture is huge."
He ended cpu talk with that. I saw nothing in there saying krait was good. Actually quite the opposite and totally sucking on software using quad so far (he noted most software seems to just ignore their extra 2 cores). No S600's in there yet but they're not going to double perf on cpu IMHO or Qcom is really under-promoting their tech. I agree with his statements based on the benchmarks. I wish he'd have included the ONE X+. It would be closer to T4i. But you can add 800mhz over One X here to guestimate. Grey should be pretty potent. OneX has a Tegra 1.5 vs. Grey (T4i) at 2.3ghz & still 60gpu vs. 12 (T604-ish? or a little better?) and very cheap to make with lower complexity, with fewer unneeded features for most resulting in smaller dies & cost savings (power and materials).
I think the current data shows Krait may have an advantage in power use vs A15, but not perf, but I can't prove the power comment until chips are released and benched in devices. I think the power is why big.little got announced (A15 power sucks somewhat & Arm would know this early on) and it remains to be seen how this plays out on phones. One more note, A6x/A6 is 123mm/96mm, so it's impressive engineering to get apple 6x perf in 80mm if T4 does this. A shrink of apple's A6x chip to 28nm from 32nm would not cut off 43mm (IMHO) and apple's no slouch in the engineering dept (100 PA Semi engineers & 100 TI engineers gained recently when they shutdown mobile etc - they keep buying tons of talent).
The gpu is basically doubled on A6x which adds the mm to A6x and Tegra 4's is only 10mm on gpu. After these two posts (and more analysis of data doing them & other posts elsewhere) I think Grey may be very compelling with 4 A9's upclocked ~800mhz vs T3 40nm (600mhz faster than T3+) running 60gpu cores with dual channel. I already showed 1.7ghz A9 T3+ winning some benches vs. quad snapdragon and doing a number on exynos 5 in a lot of benchmarks on cpu previously. I really thought Grey was totally low end here but it may be around A330 gpu with possibly better battery with all this and integrated i500 on die. While not directly related to this discussion it's fairly pertinent to the article here (doubling down on efficiency is the title after all...LOL). But I question it getting into xmas crap (doubt it), more like Q1 devices.
OOPS that was some wasted time...We have later data from Feb 25th (not a lot), with Snapdragon 600's now. So we extrapolate a bit more now:
http://www.anandtech.com/show/6792/lg-optimus-g-pro-hands-on-performance-preview-snapdragon-600-tested-at-mwc-2013
Snapdragon 600 QUAD losing to Nexus 10 which is only a dual A15 on 32nm. They scored 16309 vs. 11146 respectively (lower is better), a serious butt kicking by Dual A15 CPU correct? This would get worse vs. Octa/T4 quad's @28nm right? The supposedly "speed enhanced" A320 of S600 barely catches the Mali T604 in Nexus 10 vs LG Optimus G PRO in egypt ON screen (26 vs. 25.9 margin of error?). In offscreen 1080p Mali604 blows it away (33fps Nexus10 vs. 26 for A320 on S600 even a 50% improvement for 330 in this case would be beat here by T604 already). I do think they may be quicker in shipping units but we'll see. We need more benchmarks but I see nothing that special in 600 even vs. dual A15 cpu's so far (conservative above I think totally discounting quad A15 here, so how much better does S800 do vs. even dual A15s?) and the gpu is just a bit lacking no matter how you slice it, it seems. Still seems software optimizations need work. All 600's are supposedly "speed enhanced" A320's AFAIK (but this one didn't bench like one IMHO, should match T604 here). I expect A330 of S800 to end up ~T604 now (winning more than losing) at the same res which the T4 should handily beat, T658/Rogue6 should too (A330 will be against these 3 not T604 though). Octa got kicked out of Galaxy S4 though, so either samsung chips have issues in phones, or all A15's will
We don't know power/heat etc of the gpu in octa vs T4 which can have a huge impact on usability in phones, or even a tablet. But T4 is 80mm vs. 5250 96mm so should be less power IMHO with 10mm gpu vs. larger T604 & up. Octa uses PowerVR SGX544MP3 @ 533 MHz supposedly, not sure it's not a T658? T678? or something else (or all of them)...who knows...Depends on who you believe...LOL (Multiple versions seems likely). But T4 has a smaller gpu than everything else AFAIK so can't be too bad on power. Obviously I'm off your post in a lot of this blazorthon (my tangents?
) so if anyone has comments on the data here, please jump in. This is just me analyzing all the data out loud so to speak. Is samsung's 28nm process showing exactly what Blazorthon says above basically where they have a mature 32nm but sucking on 28nm for awhile (which has no bearing on TSMC 28nm but still...)? Hmmm...Questions questions and more ????'s.
It will be a sad state of affairs though if nobody can get A15 quad in a superphone. The freezer Nexus 4 at anandtech shows interesting throttling happening a lot on the regular non freezer one. Note it's not overclocked because of this, just that when it doesn't throttle it scores 39.6 vs. 25.6 in glbench. That's a heck of a lot of throttling for snapdragon gpu. It seems they're clocking (for spec claims?) way higher than you ever run on the gpu or this phone wouldn't be so fast in a freezer right? I'm wondering if all are lying, and wish they'd toss a few other phones in the freezer. I'm guessing snapdragon wouldn't be running flatout until they dropped a hundred mhz at least on the gpu. This only happened on gpu & cpu didn't show an advantage for freezer Nexus4 (so the adreno 320 is overheating a lot and downclocking?).
https://en.wikipedia.org/wiki/Exynos
Regarding my Octa comment before on overheating: Contradictions all over on octa even in the same article unless they have 4 versions I guess: says it uses 544mp3, and lower in here it says T678 or 554mpX?...I'm confused...What the heck is it using for a gpu? Is samsung going to be late because they can't figure out what works yet in a phone? Even the cpu's info here conflicts with itself...LOL. Is it a 1.8q+1.2q, or 2.0+2.0 for octa (with single/dual/quad A7's?)?...Are there 4 versions of octa? Jeez get this crap straight samsung...whatever
I think I've read too much this last month
😉 Comments welcome here if anyone has better spec data
I hope toms does an article like this one on OCTA. A deep dive so to speak like they did here on Tegra4 (S800 also perhaps, since we basically know a good bit on S600 now and it's not the flagship for the year).
For stock people because I'm too lazy to make another post today
If apple throws it's hat heavily behind games (much like Qcom seems to be doing now with 35 devs in house) I will likely sell my NV stock xmas or Q1 unless it looks like it will be a while before apple optimizations kicks in (or READ even more then as I get more paranoid...LOL) or at least spread more evenly across the 3. Currently I don't like apple without games being a major focus going forward and market share keeps dropping across macosx/ios
I like QCOM until everyone ships a modem (all bets off after that). They're shipping ~38% of the modems currently (~86% of LTE if memory serves, sorry read that a few weeks ago, can't find the link again...LOL), but that's a battle by end of the year I think (or Q1'14). Is everyone entering the modem market this year?:
http://low-powerdesign.com/strauss/
Yes way off our discussion now - this last bit should have been at seekingalpha/fool.com etc but whatever
Facebook
Twitter
Instagram