[Leaplist] Different netbooks ... CPU and GPU limitations (or
not) ...
Kevin Korb
kmk at sanitarium.net
Wed Jan 13 17:46:13 EST 2010
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
I agree completely about Hyper-Threading. It sucked horribly on the P4
back when it came out. I had actually thought that Intel abandoned it
until I got my EeePC and it had 2 "logical" CPUs. Having the system try
to multi-thread across 2 CPUs that are really the same CPU is worse than a
waste of time.
Picture 2 way traffic on a 1 lane road and you get the idea what HT is.
It isn't quite that bad but it is close.
I can't say I have any experience with it on systems with lots of CPUs
though. I think a dual P4 was the highest system I ever saw HT on and it
ran better with HT turned off.
On Wed, 13 Jan 2010 14:37:46 -0800 (PST)
"Bryan J. Smith" <b.j.smith at ieee.org> wrote:
> Hyperthreading on a single or dual-core Atom is like putting a
> supercharger on a Yugo. Intel is marketing non-sense again. The whole
> idea was to have hyperthreading across multiple Atom cores. Now it's
> being marketed like the Pentium 4 was, totally useless.
>
> Hyperthreading purposely kills performance of the Atom, everything
> will run much, much slower. The original idea for hyperthreading was
> to distribute, say, five or even six (5-6) threads, across, say, quad (4)
> of 2-issue Atom cores. That would work well. Not faster in immediate
> performance, but more work done over time, despite inefficiencies.
>
> Hyperthreading on a single or dual-core Atom is a joke, and will almost
> always result in worse performance. It's really designed so a batch of
> Atom X cores (where X is at least 4) can run more like X+X/2 threads --
> 50% more.
>
> Atom's are 2-issue, in-line, "I get it I do it" non-speculative,
> non-predicting, non-anything. They are simple in design for a reason --
> lots of small cores running lots of different threads from a pool.
>
> -- Bryan
>
> P.S. Software RAID-5? You're going to tie-up the entire interconnect
> so you can't do jack with MythTV. Understand one of the original targets
> for a quad-core Atom was to _be_ the intelligence of a dedicated
> RAID-5/6 controller -- not the end-system.
>
> Maybe this is the best way to put it ... ;)
>
> "The Atom is a 2-issue micro_controller_ -- _not_ a micro_processor."
>
>
> P.P.S. The Marvell cards are actually designed with _software_ support
> that is _not_ open source. The Broadcom-Marvell approach is to give
> you a basic, "part-hardware" software intelligence for RAID and _charge_
> for the "part-software" management components in a varying "software
> stack." It's neither 100% hardware nor 100% GPL in software.
>
>
> ----- Original Message ----
> From: Richard F. Ostrow Jr. <rich at warfaresdl.com>
>
> I, too, have two atom-based systems - an Atom 320 (1.6 GHz w/
> hyperthreading) that I was trying out as a mythtv backend, and an Atom
> 330 that I'm using in my car (1.6 GHz dual core w/ hyperthreading). The
> 320 actually had 4 SATA connectors that I ended up putting 4 1TB hard
> drives into a software RAID-5 on. However, I'm afraid that I terribly
> underestimated the CPU requirements of mythtv on the backend side... the
> commercial scanning (which is really the only CPU-dependent task it has)
> is now over a month behind on the recordings, working around the clock
> 24/7. Add to that the problem that the thing crashes mythtv for some
> reason without explanation (a problem I did not have with my old athlon
> 64), and sometimes records video rather sloppily (jerky audio and
> video)... and I'm about done with this system for that purpose. I was
> hoping to reduce my power requirements, but it simply isn't going to
> work.
>
> On the other hand, the Atom 330 in my car seems to be adequately up to
> the task... it has no issues running my car pc software, doesn't fail to
> POST with my touchscreen (as my last mini-itx form-factor machine (A Via
> C7)did), and theoretically can run actively for quite a while on the car
> battery before it would need to shut itself down to avoid running the
> battery down too far. I had originally purchased a PCI GeForce 5600 for
> it for my limited graphics processing needs, but find that the intel
> chipset can actually handle what I'm throwing at it (to my surprise). My
> only gripe with the thing is that it has a RAID daughterboard from
> Marvell that falsely claims it has linux support (it was originally
> going to be my mythtv backend server). The car has no need for such a
> thing, so I removed it.
>
> In both cases, they have a very wide temperature and humidity tolerance
> (critical for my car, and important for my mythtv backend, which I've
> located in my garage).
>
> I should also note that neither of these are "netbooks" - I absolutely
> refuse to pick up such a machine.
>
> On Wed, January 13, 2010 11:23 am, Kevin Korb wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> >
> > I have 2 Atom based systems. They have their uses but you are correct
> > about them being weak.
> >
> > The first is my 10" EeePC with an Atom N270 (single core 1.6GHz). I
> > only use it for IRC, email, web browsing, ssh, Audacious, and the
> > occasional classic arcade style game. For those purposes it works
> > decently. The crappy SSD that Asus ships is actually more limiting
> > than the CPU is. But
> > the CPU is still strong enough that things like disk encryption and
> > OpenVPN tunneling do not make a noticeable difference in performance
> > (probably because the SSD is only slightly faster than a thumb drive).
> > With my usage which generally involves both heavy wi-fi usage,
> > encryption,
> > and music playing I usually get right at 4 hours of battery time.
> >
> > The second is an Atom N230 (single core 1.6GHz) that I use as my
> > firewall/router. It is in a phone book size box that has 2 PCI slots
> > so I
> > had no trouble getting 3 NICs into it. I don't use the video at all
> > so I could care less about the performance there. I like the low power
> > consumption (about 1/3 of the power is used by the 2 PCI NICs I added).
> > It is faster than the old AMD K6-2 500MHz system it replaced while
> > using less than half the power so I am happy.
> >
> > Frankly I can't think of anything else I would use an atom for. If I
> > could get one that could handle a crapload of storage I might consider
> > it for a file server (sort of a DIY NAS) but none of them I have seen
> > would be capable of that.
> >
> > I would like to try a newer Atom Z530 based system though. Those are
> > dual
> > core still 1.6GHz but they have some newer features like hardware
> > virtualization. It is probably still wimpy but it may be well suited
> > to the low powered netbook market. Were any of the ones you tested
> > from the Z series?
> >
> > On Wed, 13 Jan 2010 08:08:46 -0800 (PST)
> > "Bryan J. Smith" <b.j.smith at ieee.org> wrote:
> >> Okay, I've now had the chance to mess with various types of netbooks:
> >> - Intel single/dual-core 45nm Atoms with the 65nm 965GSC
> >> - Intel single/dual-core, ultra-low voltage 45nm Pentium with same
> >> - AMD, single-core, low voltage 65nm Athlon 64 with 55nm M690E
> >> - AMD, single/dual-core, low voltage 65nm Neo [x2] with 55nm M780G
> >>
> >> Let's get one thing straight, the 2-issue, in-line, non-speculative
> >> Atom
> >> sucks. You could quad core it and then you'll get what an aged AMD Neo
> >> does. Anything with an Atom is going to be CPU limited. Ironically,
> >> the
> >> 965GSC makes a great partner for the Atom here. The nice thing is
> >> that you do get 6 hours out of the sucker, and that should improve
> >> with Intel
> >> outsourcing to TSMC and their 40nm process for the GPUs. I understand
> >> the reasons for Atom, but as they have existed so far, they make poor
> >> processor for a general purpose OS.
> >>
> >> nVidia has introduced it's Ion chipset for Atom. But adding a GPU to
> >> a processor that struggles with basic operations is not going to help
> >> much.
> >> As I said, the old 965GSC design is a perfect match. ;)
> >>
> >> Now, ironically, I've had the chance to mess with some of the nexgen,
> >> 45nm, ultra-low voltage Pentium processors, including dual-core.
> >> Those actually have a bit of kick to them. GHz for GHz, they will
> >> slump against AMD, but Intel has the power-GHz superior fab
> >> technology, so they come out ahead in both performance and battery
> >> life. Almost makes you want to forget Atom, as they can give you 4
> >> hours. The pricing is the only problem at this point.
> >>
> >> I haven't checked if nVidia has a solution for these ULV Pentium
> >> designs.
> >> If they use the same interconnect, they should. I need to investigate
> >> further, but there are few netbooks in this area so far.
> >>
> >> AMD has a problem. They are still selling 2+ year-old technology, and
> >> we're still waiting for 45nm products to hit the market. You have a
> >> choice of the low-voltage 65nm Athlon 64 or the optimized, but still
> >> aged 65nm Neo [x2] single/dual-core.
> >>
> >> The Neo x2 is capable of keeping up with feeding the newer Radeon
> >> 2100-3200 GPU cores (M690x and M740x/780x), definitely bests the
> >> ULV Pentium dual-cores GHz for GHz, but at an extreme power penalty.
> >> So the power and cost savings with Neo [x2] over Turion [x2] is
> >> minimal,
> >> and yesteryear's refurbished models are the only affordable under
> >> $500. Same problem as the ULV Pentium on the pricing front (worse
> >> battery than
> >> it though).
> >>
> >> Which brings us to the low-voltage Athlon 64. Here you have a 1.2GHz
> >> Athlon 64, underclocked and power drops below 25W. But then it can't
> >> feed even the old 2100 GPU core (shrunk to 55nm in the M690E). I
> >> know, I just bought one. It's great for normal usage, kicks the
> >> living crap out of a dual-core Atom (ignore the artificial
> >> benchmarks, I've used both myself on Linux and Windows), and then you
> >> have the GPU for off-load of things the 965GSC just can't do. And
> >> that's the funny part. You can jack up filtering, textures, etc...
> >> on that Radeon 2100 GPU without hitting the CPU at all (unlike the
> >> feature-set in the 965GSC) -- "totally free." But the GPU still
> >> can't drive/off-load the actual CPU operations, so if the CPU is hit
> >> hard (doing absolutely nothing to do with any 3D/rendering/playback),
> >> it doesn't matter.
> >>
> >> AMD does this to get the CPU power consumption down, and battery life
> >> to 4-5 hours. The chipset is already 10-15W at 55nm, and the new revs
> >> will go to sub-10W at TSMC's 40nm (although Intel's new use of TSMC as
> >> a
> >> foundry might get interesting on speed-to-market).
> >>
> >> Now it's clear AMD took a major chunk of the market with the
> >> low-voltage
> >> Athlon+M690E combo. Because _all_ of the Acer-Gateway and whitebox
> >> models I've seen lately are sporting a ULV Pentium Single/Dual-Core
> >> with
> >> an Intel chipset like the X3000/4000 series. It's funny because that
> >> now
> >> reverses the problem (and even kills the battery a little bit more
> >> than AMD's aged solution). The only AMD combos I can find are old
> >> models**.
> >>
> >> AMD has a great GPU, but is either CPU-limited or power-limited (by
> >> CPU).
> >>
> >> Intel has a great CPU (ULV Pentium), but is always ships a GPU clunker
> >> (in _real_ features, not those artificial "dumb framebuffer"
> >> benchmarks), let alone in Linux (where the Intel driver totally lacks
> >> IP
> >> features of their Windows counterparts -- ATI and nVidia offer a
> >> closed source driver for such).
> >>
> >> So, in a nutshell, there's no happy medium right now. Maybe Intel
> >> will stop fighting with nVidia and work with them for a combo to the
> >> ULV Pentium. Then again, pricing is still an issue -- and I hope
> >> AMD's new 45nm finally delivers what it should. 2GHz 45nm single-core
> >> performance
> >> of 4-5 hours with a 40nm M780G/T (Radeon 3100) or successor. But
> >> right now it seems there's still a lot of old stock out there.
> >>
> >> -- Bryan
> >>
> >> **P.S. With that said, I still just bought the older LV Athlon +
> >> M690E refurb combo for $270 from Woot. The 1.2GHz Athlon 64 is
> >> capable of meeting requirements for a 2GHz Penitum 4, and -- again,
> >> it's funny -- to jack up _all_ of the DirectX 9 / OpenGL 2.x features
> >> for shading and textures with the Radeon X1270 (2100) integrated GPU
> >> and it makes _no_difference_ in performance (because the GPU can
> >> handle them). It's that CPU doing background stuff in a title that
> >> seems to hit any performance, not any rendering (the GPU is more than
> >> capable). At least
> >> memory and disk are beyond the artificial XP netbook $5 pricing,
> >> although it ships Vista, I'm running Linux on it anyway. ;)
> >>
> >
> >
> > - --
> > ~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~
> > Kevin Korb Phone: (407) 252-6853
> > Systems Administrator Internet:
> > FutureQuest, Inc. Kevin at FutureQuest.net (work)
> > Orlando, Florida kmk at sanitarium.net (personal)
> > Web page: http://www.sanitarium.net/
> > PGP public key available on web site.
> > ~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~
> > -----BEGIN PGP SIGNATURE-----
> > Version: GnuPG v2.0.11 (GNU/Linux)
> >
> > iEYEARECAAYFAktN84YACgkQVKC1jlbQAQeVwACeKzVA1kXrpARwAIXMTwU3i5fq
> > zyYAnRvkzqzl5QTBhkrhTCSm8R3eVXrr
> > =NVym
> > -----END PGP SIGNATURE-----
> >
> > --
> > This message has been scanned for viruses and
> > dangerous content by MailScanner, and is
> > believed to be clean.
> >
> > _______________________________________________
> > Leaplist mailing list
> > Leaplist at leap-cf.org
> > http://lists.leap-cf.org/mailman/listinfo/leaplist
> >
>
>
- --
~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~
Kevin Korb Phone: (407) 252-6853
Systems Administrator Internet:
FutureQuest, Inc. Kevin at FutureQuest.net (work)
Orlando, Florida kmk at sanitarium.net (personal)
Web page: http://www.sanitarium.net/
PGP public key available on web site.
~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~'`^`'~*-,._.,-*~
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.11 (GNU/Linux)
iEYEARECAAYFAktOTTUACgkQVKC1jlbQAQcXjQCg5tQioPJMp8OaHQ2JsC68/RvO
fRcAoOw4l2ro4sGadhhRta4KZkHRjKAV
=w+ok
-----END PGP SIGNATURE-----
--
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
More information about the Leaplist
mailing list