Yes, it's farcical. I've never seen sampling confused with tape out.Exophase said:
That said, it's not unusual to start sampling only days after first samples back from fab. I've seen it done after 2 days.
Tegra 3 officially announced; in tablets by August, smartphones by Christmas
- Thread starter Mike11
- Start date
That said, it's not unusual to start sampling only days after first samples back from fab. I've seen it done after 2 days.
The 5x performance increase is for the entire SoC and that's including 2 more A9 CPU cores at a higher frequency compared to T2
The 50% more units (12 vs. 8) most likely should go to the pixel shader side. If Anand isn't wrong about the 4+4 in T2 then it'll go from 1 Vec4 PS to 2 Vec4 PS ALUs.
Helloooo it's still on TSMC 40nm.
Excuse me I'm just reading what Anand stated which obviously comes from the lion's mouth. If we're going to play the wild theories game before you're going into anything that will further increase die area/chip complexity you may want to again remember that it's on 40nm. Neither higher frequencies nor more die area are for free under the same manufacturing process.
Tegra 3 SoC = 5x Tegra 2 SoC. The 4*A9's might be clocked up to 1.5GHz in T3, put that against 2*A9@1GHz and a more modest increase from the GPU side.
How much faster did NV claim in the past T2 would be against T1 again?
He's also stated 4 MADDs from the pixel shader units or the vertex shader units respectively. That's 8 FLOPs/clock from each side. There should be one more additional FLOP from the programmable blending unit in the PS ALU, but it's only available if the core isn't blending so it's rather silly to say 9+8, and what for anyway? You're not going to do anything GPGPU with merely FP20 precision either.
I'm sorry, i wasn't aware of any Tegra real numbers.. I assumed that the 8 cores where similar to the desktop architecture.
So it's 1.92 Gigaflops.
http://www.engadget.com/2011/01/24/nvidia-tegra-3-equipped-with-1-5ghz-quad-core-madness-teased-b/
The performance chart is related to the entire SoC, not just the GPU part as i was thinking in the beginning. The 5x performance comes from 3x times the graphic power and 2.5x twice the cpu power.
So Kal-El has 4 Cortex A9-core and it's 40nm (right?)
Wayne may have 4 Cortex A15-cores and it's on 28nm.
From the time frame, i think Logan and Stark are on 20nm.. and probably based on ARMv8 architecture.
Finally someone that can decipher marketing hieroglyphs
labwarrior
Newcomer
Are these the same cores as in Tegra 2 though ?
Not impossible; yet wouldn't the relevant pre-release marketing slides trumpet with it?
Okay, but nVidia claimed they were shipping samples in December 2010, not 12 days before February 16. So I think something doesn't add up here.
Same with the claim that competitors will only be sampling quad cores in 2012. I'm pretty sure the quad i.MX6 is going to be sampling long before then. Other vendors like TI and ST-Ericsson are, not entirely surprisingly, not as interested in quad core.
Last edited by a moderator:
A past leaked slide indicated that the smartphone AP30 SoC would have either a dual or quad core A9 option. If that should still be valid it would make sense for the smart-phone segment since quad core there could still be debatable today for its actual benefits. On contrary tablets can go to much higher resolutions than smart-phones and are a totally different chapter.
As for TI or ST (if you should be referring to OMAP5/NOVA9600) they're to arrive a LOT later than Tegra3. Besides there's also a difference between quad A9 and dual A15. On top of that OMAP5 strikes me as a MP4 GPU.
http://www.anandtech.com/show/4181/...-a9s-coming-to-smartphonestablets-this-year/4
These are all speculative opinions of course (especially the quote above); however I'm still not convinced that a higher concentration on the CPU side of things vs. a higher concentration on the GPU side of things could win in the long run. With 3D GUIs and the advent of some GPGPU functions in future SoC software I'd expect from the worldwide graphics leading IHV an exact opposite concentration.
It'll most likely come eventually, since I have severe doubts that if NV's roadmap has any merit that Logan will reach such a performance increase mostly from the CPU side. Rather the contrary and yes I might as well be wrong again here, but I don't see a gazillion of CPU cores at ultra-wild frequencies to make sense in a embedded SoC.
Man from Atlantis
Veteran
so what gfx performance are you expecting with tegra 3? around Geforce FX, 6 or better ?
Up to 3x times graphics performance compared to Tegra2. There's no reasonable comparison with desktop GPUs. Tegra2 manages to get roughly 45fps in Q3a in 1024*600 without any AA or AF. http://www.firingsquad.com/hardware/tioverclocking/page5.asp
Well yes, it's farcical. It's also a way to get announcements out earlier.
Other than for CPUs and GPUs (ever there often just rumors), the press hardly ever talks about when a chip has taped out.metafor said:
So how do you know if people are making these false claims?
Let's just say I'm often surprised when I wake up and read a press release and think "wait, we've got silicon back already?"
1.5x the cores, and 2x frequency? (if it's on 28nm it could be possible)
I can't find any info on that. Arun in the other thread said that he was pretty sure about being 28nmLPG.
You must be working on very cool chips then. Good for you!metafor said:
That coremark is highly suspect:
-> GCC 3.4.4 May 18, 2005
-> GCC 4.4.1 July 22, 2009
C2D was released in 2006
O2 - optimizations that decrease size without affecting speed and increase size without decreasing speed
O3 - O2 optimizations plus (essentially) anything that will increase speed
and here's the confirmation
http://techreport.com/discussions.x/20451
And here's source of T7200 score from slide - benchmark scores page. So it's not internal testing nvidia's result for T7200
Who cares if nVidia submitted the score or not? They still knew better. They COULD have ran a new fair score for T7200 and didn't. If they really didn't have access to a T7200 they could have at least ran theirs on the old compiler with the same flags. Or they could have just not made the comparison at all.
nVidia using a stock score is just a tactic to try to make it look legitimate. Besides, that score was submitted pretty recently and we don't know for sure that this company wasn't doing it on nVidia's behalf. In fact, why else would this company have submitted a score years obsolete just last November?
I wonder what else we can find on this Gerrit Slomma.
https://bugzilla.redhat.com/show_bug.cgi?id=491554 < reported a bug for an nVidia chipset
http://www.primegrid.com/forum_thread.php?id=1571 < released a CUDA prime solver
Okay, that probably doesn't say much on its own, except that he probably isn't anti-nVidia.
So what about ITS-Slomma, what do they do?
http://translate.google.com/translate?js=n&prev=_t&hl=en&ie=UTF-8&layout=2&eotf=1&sl=auto&tl=en&u=http%3A%2F%2Fwww.itsslomma.de%2Fabout.jsp
They're a small IT firm, I can't imagine why they would officially release a CoreMark processor score except on someone's behalf.
If nVidia really did ask for another entity to release this score for them then that makes things even worse.
Maybe someone should straight up ask him why he submitted the score.
Last edited by a moderator:
