Better To Wait For The Omap 3630?


Status
Not open for further replies.
'mali' said:
Just a question, what is the topmost chip on the SoC? Is it the CPU or is it the PoP RAM?
the latter.

my two cents (though i believe i have already thrown them around at least once):

keep your systems at stock, people! many brilliant engineers (both hw and sw) have worked hard to bring determinism to computational technologies based largely on statistical mechanics. do not sway the luck away from you. yes, you can experiment all you want to see what-if, but when expecting reliability and actual work from your pandora, stay cool ™

i expect that in a couple of years pandora team may release a board based on a 36xx, running cooler and trivially reaching higher clocks than our current omap. then you will be able to naturally bump up your pandoras, at the expense of a logic board replacement. until then, though, pandora sw devs will largely write for a 3530@600 and will expect some conformance in the install base.

btw, that c2d analogy was pointless. pandora has neither the TD freedom of the desktops, nor the C2D lithography advantages. and if you believe heat does not shorten an IC lifespan (and heat is always present whenever electrons travel, unless you're a kryptonian) you need to wake up. a modern IC is not immortal even at its nominal clock, it is just a matter of meeting satisfactory lifespans (usually ~5 years for consumer-grade parts) and reliability quotas (errors per number of ops).
 
Last edited by a moderator:
Hopefully Haiku will have a good arm port by 2010 it will be totally sick on a multicore ARM its like BeOs is like greased lighting on my dual 300mhz pentum I can do stuff with it that i could only dream of on Linux ie browse the web and even play 3 .AVIs at the same time :-||



hopefully they will integrate the same level of hardware optimiations as Be did (the even had DSPs in the original BeBox)
 
'craigix' said:
I'm not buying a Pandora until we see the OMAP5.
Well, if he's holding off, so am I!

Seriously though, if these guys are crazy enough to do this again, we should be talking about omap 4xxx stuff and skip right over the 36xx. Only question is whether or not it will be on 32nm tech by then.
 
Last edited by a moderator:
'Laurent' said:
'MWeston' said:
IVA3

I wonder if that'll use the same C64x+ DSP...

@mali : the pipelined VFP indeed is one strongpoint of A9, but there are some others :p
I'm personally more excited about out of order execution/speculative execution/register renaming, removing a lot of the burden on the compiler/code generator for scheduling and relieving register pressure for avoiding false dependencies. And smaller pipeline, meaning lower mispredict penalty (and with interpreters in emulators every emulated instruction is probably a mispredict). It may also have 4-way NEON, not totally sure on this one, I bet you could clarify that part.
 
Last edited by a moderator:
'craigix' said:
I'm not buying a Pandora until we see the OMAP5.
Isn't OMAP5 available already? And no where near as powerful as OMAP3?

But seriously, I'm not buying a pandora until it has an NVIDIA Quadro4 980 XGL either. That will be the best, I don't understand why you don't do that. And Windows Vista too, why are you using Linux? Linux sucks! Everyone knows that. Why no Vista!!??

;)

Unrelated: Agghhh, why all the apostrophes? And backslashs now too
 
Last edited by a moderator:
'lulzfish' said:
I'm waiting for it to go x86.
Haha, yes. You're starting to sound like someone else now...
 
Last edited by a moderator:
I have made this argument several times before, but there really wont be any hardware past what the P has that hits any major milestones quite some time.

What can it not do right now? Assuming Doom 3 manages to work at a playable degree, that leaves quite a lot of breathing room for arm ported games; I wouldn't expect to see people complaining about X game being unable to run unless there is some sort of ARM revolution and all gaming companies suddenly decide to adopt within the next two years. In other words, any milestones that exist on this front are not very important and do not merit a new hardware revision to meet.

For utility purposes, I don't see any real restrictions. Sure it would be nice to have every single app you could ever use loaded in the RAM while compiling and with firefox open with six tabs, but most netbooks aren't put under such expectations. And for good reason; its a little unreasonable to expect desktop-level performance out of such netbooks, and as desktops evolve, so will the definition of desktop-level productivity
performance. So really, aiming for such a thing will prove fruitless; desktop-level performance is not a reasonable milestone to aim for.

Emulation seems to be the only front that has a visible milestone, albeit a distant one. N64 emulation seems to be a matter of effort, so eventually somebody will tackle it. While it could be argued that dreamcast emulation is the next milestone, I would say that it is not a major enough milestone to warrant new hardware. Really, the next milestone is PS2 emulation. Yeah, its pretty damn far away. That's the point. I doubt we will see any major hardware revisions until it can at least hit that kind of emulation capability.

The one point I will concede, however, is high def output. Currently the hardware cannot support it, and (if my observations are correct) usb output appears to be limited to 800x600 resolution. So long as the meat and potatoes of the system remains the same, I can easily see some sort of Pandora HD coming in the future.
 
Last edited by a moderator:
The comparison to OMAP4 made earlier was from OMAP36xx, not 35xx/34xx.

In that context, they share the 45nm advantage and some of the increased clock speed, yet OMAP4's upgraded SGX540 core gives it twice the shader pipelines and TMUs for probably around a 2x+ performance increase (not 4x since the 36xx's clock speed was also increased somewhat from the 35xx/34xx families.) Availability of the 36xx is scheduled to only be a quarter or two ahead of OMAP4.
 
Last edited by a moderator:
'Exophase' said:
I'm personally more excited about out of order execution/speculative execution/register renaming, removing a lot of the burden on the compiler/code generator for scheduling and relieving register pressure for avoiding false dependencies. And smaller pipeline, meaning lower mispredict penalty (and with interpreters in emulators every emulated instruction is probably a mispredict).

Yes, Cortex-A9 has all of that. And some other nice improvements :)
QUOTE
It may also have 4-way NEON, not totally sure on this one, I bet you could clarify that part.

I could but then I'd have to kill you or, worse, lose my job :p
 
Last edited by a moderator:
As for overclocking and variation.. Nokia bought speed-picked omap2 chips from TI, in order to guarantee that they could run at 400MHz (which IIRC is slightly above normal specs for that particular cpu). I doubt we're getting speed-picked chips for our Pandoras, which means that even though some could run at 900MHz it's unlikely that _all_ will be able to.
 
Last edited by a moderator:
'Laurent' said:
'Exophase' said:
I am personally more excited about out of order execution/speculative execution/register renaming, removing a lot of the burden on the compiler/code generator for scheduling and relieving register pressure for avoiding false dependencies. And smaller pipeline, meaning lower mispredict penalty (and with interpreters in emulators every emulated instruction is probably a mispredict).

Yes, Cortex-A9 has all of that. And some other nice improvements :)
QUOTE
It may also have 4-way NEON, not totally sure on this one, I bet you could clarify that part.

I could but then I would have to kill you or, worse, lose my job :p

wait, laurent has access to an a9? o_O

*reaches for pliers and other useful truth extraction tools..*

on a tangent, i am planning on arranging a spectacular showdown between my pandora and my (yet to order) efika (moto^H^H^H freescale mpc5100, e300@400MHz - basically vanilla 603e - no simd). tickets for the match available soon.
 
Last edited by a moderator:
'darkblu' said:
wait, laurent has access to an a9? o_O

*reaches for pliers and other useful truth extraction tools..*

on a tangent, i am planning on arranging a spectacular showdown between my pandora and my (yet to order) efika (moto^H^H^H freescale mpc5100, e300@400MHz - basically vanilla 603e - no simd). tickets for the match available soon.
Yes, Laurent has access to some very privileged information.. he's pretty water tight but it's often apparent that he knows these things :>

Is your Efika board not really MPC5200B instead? If so, I think Cortex-A8 at 600MHz will win in most cases (unless you need double precision or IEEE754 compliant floating point). 256KB of L2 cache is a huge advantage.
 
Last edited by a moderator:
'Exophase' said:
Is your Efika board not really MPC5200B instead? If so, I think Cortex-A8 at 600MHz will win in most cases (unless you need double precision or IEEE754 compliant floating point). 256KB of L2 cache is a huge advantage.
of course it is an mpc5200b - i have been reading lately about the mbx-equipped 5100 so that was a freudian slip on my part. the 5100 also has bigger L1 caches, so it may have clouded my expectations of the 5200. truth be told, tough, genesi are rumored to have a 5121-based in the works, so we will see (as i mentioned i am still to order the thing). anyway, i am constantly on the search for good ppc implementations, so if you know of anything worth checking let me know.

btw, my original idea was to get my software rasterizer running on both; it may be utterly unoptimised, but at the same time is a good complier- and fpu-litmus (single precision, naturally).
 
Last edited by a moderator:
Here's what is recommended for BeagleBoard CPU frequency. You'll notice the increase of voltage needed for the various frequencies.

CODE
http://elinux.org/BeagleBoard#Clocking


I hope that'll stop some people from doing unfounded claims, yes I'm a dreamer :)
 
Last edited by a moderator:
Status
Not open for further replies.
Back
Top