Is AMD doing well again?

On Fri, 28 Aug 2009 19:43:24 +0000, Dan Lenski wrote:

eah, since writing that I found out about the Nvidia Ion chipset, which
is pretty new. There are a bunch of good deals on Zotac boards with
this chipset, apparently (http://forums.slickdeals.net/showthread.php?
t=1504403). The problem is that NVidia has the worst open-source driver
support of the major GPU makers. And with 2D/3D graphics improving so
rapidly for ATI and Intel GPUs, I'm reluctant to go with NVidia.

That's strange. I've been running Linux since 2000 and Nvidia is the only
one I've found with good drivers. Granted, they aren't open source, but
they are free. OTOH, I tried an ATI board and their drivers sucked.
Finally sold the MB.



--
Want the ultimate in free OTA SD/HDTV Recorder? http://mythtv.org
My Tivo Experience http://wesnewell.no-ip.com/tivo.htm
Tivo HD/S3 compared http://wesnewell.no-ip.com/mythtivo.htm
AMD cpu help http://wesnewell.no-ip.com/cpu.php

On Sat, 29 Aug 2009 05:52:51 +0000, Wes Newell wrote:

On Fri, 28 Aug 2009 19:43:24 +0000, Dan Lenski wrote:

Yeah, since writing that I found out about the Nvidia Ion chipset, which
is pretty new. There are a bunch of good deals on Zotac boards with
this chipset, apparently (http://forums.slickdeals.net/showthread.php?
t=1504403). The problem is that NVidia has the worst open-source
driver support of the major GPU makers. And with 2D/3D graphics
improving so rapidly for ATI and Intel GPUs, I'm reluctant to go with
NVidia.

That's strange. I've been running Linux since 2000 and Nvidia is the
only one I've found with good drivers. Granted, they aren't open source,
but they are free. OTOH, I tried an ATI board and their drivers sucked.
Finally sold the MB.

Yeah, until a couple years ago, I would've agreed completely... NVidia's
drivers were fairly good but closed-source, and ATI's drivers were awful
and slow *and* closed-source. I used NVidia gear until about a couple
years ago, and was getting fed up with poor software suspend support.

Today, ATI's closed-source drivers have improved somewhat, and NVidia's
were pretty frustrating for software suspend last time I played with them.

But more importantly... AMD/ATI is now cooperative and releases
documentation, and there is very rapid progress in the open source
drivers. My ATI HD3200 integrated graphics went from /zero/ open source
support (not even VESA) a year ago, to fully-accelerated 2D support and
kernel modesetting and experimental 3D acceleration today. My ATI RV370
PCIe card now has pretty much full 3D acceleration support with the open
source drivers! Meanwhile, the Nouveau open-source drivers for NVidia are
progressing slowwwwly, since they are purely reverse-engineered, lacking
any support from NVidia.

So, I'm pretty gung-ho about ATI graphics these days. I think AMD is
reaping the benefits of open-ness in terms of much better Linux support at
lower cost. If NVidia decides to do the Right Thing, I'll drop my qualms
about using their products.

Dan

Dan Lenski writes:
So, I'm pretty gung-ho about ATI graphics these days. I think AMD is
reaping the benefits of open-ness in terms of much better Linux support at
lower cost. If NVidia decides to do the Right Thing, I'll drop my qualms
about using their products.

I agree. The state of ATI FOSS drivers depends a lot on what GPU you
have -- older chips tend to be well supported, newer ones are still
being worked on -- but most importantly, they're on the right path, and
have official support and documentation from AMD/ATI.

Nvidia, on the other hand, is still cramming its fingers in its ears and
chanting LALALA as loudly as it can... (there are certainly Nvidia
engineers who support FOSS, but management is still strictly in the
stone ages) Not Very Promising.

[and that Jen-Hsun Huang guy... man, what a blowhard! :]

-Miles

--
Kilt, n. A costume sometimes worn by Scotchmen [sic] in America and Americans
in Scotland.

On Aug 29, 3:42*am, Miles Bader wrote:
Dan Lenski writes:
So, I'm pretty gung-ho about ATI graphics these days. *I think AMD is
reaping the benefits of open-ness in terms of much better Linux support at
lower cost. *If NVidia decides to do the Right Thing, I'll drop my qualms
about using their products.

I agree. *The state of ATI FOSS drivers depends a lot on what GPU you
have -- older chips tend to be well supported, newer ones are still
being worked on -- but most importantly, they're on the right path, and
have official support and documentation from AMD/ATI.

Right. As I said, my RV370 card is totally supported, whereas support
for my newer HD3200/RS780 IGP has improved immensely (using it now!).
I assume that now that AMD/ATI are releasing docs, the lag time for
newer chips to be supported by the FLOSS drivers will decrease a lot.

I guess a real test will be whether or not ATI promptly releases docs
for the upcoming R800-series GPUs!

Nvidia, on the other hand, is still cramming its fingers in its ears and
chanting LALALA as loudly as it can... *(there are certainly Nvidia
engineers who support FOSS, but management is still strictly in the
stone ages) *Not Very Promising.

What is their rationale for *not* supporting FLOSS at this point??
The first one I recall from NVidia and ATI was "graphics drivers are
way too hard, open-source amateurs couldn't do it anyway." That was
disproven with Intel drivers and some of the other IGP drivers. Then
it was "our driver code contains other company's IP which we're
contractually obligated not to release." Then AMD bought ATI and
suddenly that problem went away because they started releasing
documentation instead of their (presumably crappy) code.

Does NVidia have a specific argument against releasing code or
documentation now? If they were leaps and bounds ahead of ATI, I
could see them not wanting to reveal some of the technical features of
their designs... but as far as I can tell there are few if any novel
features to strongly differentiate their products.

Dan

Dan Lenski writes:
Nvidia, on the other hand, is still cramming its fingers in its ears and
chanting LALALA as loudly as it can... Â*(there are certainly Nvidia
engineers who support FOSS, but management is still strictly in the
stone ages) Â*Not Very Promising.

What is their rationale for *not* supporting FLOSS at this point??
The first one I recall from NVidia and ATI was "graphics drivers are
way too hard, open-source amateurs couldn't do it anyway." That was
disproven with Intel drivers and some of the other IGP drivers. Then
it was "our driver code contains other company's IP which we're
contractually obligated not to release." Then AMD bought ATI and
suddenly that problem went away because they started releasing
documentation instead of their (presumably crappy) code.

I don't keep close track of what Nvidia's claims on the matter, but I
imagine the actual reasons are a mixture of cluelessness and arrogance
in top management (engineers and researchers tend to be a bit better, at
least on average), and the various practical legal issues like the "IP
argument" you mention above. I suppose the most important thing is
probably a cultural suspicion of _any_ "non business" relationships
(an attitude I encounter regularly at my work, especially among older
higher-level managers).

I think the IP argument is overblown, but not entirely nonsense -- one
of the limiting factors on the AMD/ATI release of public documentation
is apparently the speed with which they can document things from scratch
(using a relatively small team) and have the lawyers go over it with a
fine toothed comb. I don't know exactly what the potential problems
they're trying to avoid are, but I suppose their non-public
documentation is a mismash from various sources and contains info about
proprietary libraries or whatever...

Anyway, it seems fairly clear that although AMD is committed to the
process, they're still scared (perhaps unnecessarily) of potential
problems, and are moving very carefully as a result.

There are occasional posts by AMD people on places like phoronix.com,
which give some info about this kinda thing...

-Miles

--
Abstainer, n. A weak person who yields to the temptation of denying himself a
pleasure. A total abstainer is one who abstains from everything but
abstention, and especially from inactivity in the affairs of others.

On Mon, 31 Aug 2009 05:55:33 +0900, Miles Bader wrote:

I don't keep close track of what Nvidia's claims on the matter, but I
imagine the actual reasons are a mixture of cluelessness and arrogance
in top management (engineers and researchers tend to be a bit better, at
least on average), and the various practical legal issues like the "IP
argument" you mention above. I suppose the most important thing is
probably a cultural suspicion of _any_ "non business" relationships (an
attitude I encounter regularly at my work, especially among older
higher-level managers).

So... the usual. It just seems *so shortsighted* in this day and age.
It's not like they'd be heading into uncharted waters. Tons of hardware
companies have been cooperating with open-source projects for years and
years, with almost no bad results that I can think of. Heck, even ATI
and Intel are opening up their drivers, right in the same market niche.

I think the IP argument is overblown, but not entirely nonsense -- one
of the limiting factors on the AMD/ATI release of public documentation
is apparently the speed with which they can document things from scratch
(using a relatively small team) and have the lawyers go over it with a
fine toothed comb. I don't know exactly what the potential problems
they're trying to avoid are, but I suppose their non-public
documentation is a mismash from various sources and contains info about
proprietary libraries or whatever...

Interesting. But in the process of this clean-room-ish documentation
process, they'll actually get *better* documentation out of it. So it's
a win-win. ATI will have better, legally unencumbered documentation for
their products, and so will the FLOSS community.

Anyway, it seems fairly clear that although AMD is committed to the
process, they're still scared (perhaps unnecessarily) of potential
problems, and are moving very carefully as a result.

There are occasional posts by AMD people on places like phoronix.com,
which give some info about this kinda thing...

Yeah, Phoronix is a good source for this stuff. As long as they're still
moving forward, even if cautiously, I'm quite pleased. As far as I can
tell, they've already released enough docs to do 2D and video support
for /all/ their current and past hardware. And 3D is underway.

Dan

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi,

As one possible explanation...

Dan Lenski pravi:
What is their rationale for *not* supporting FLOSS at this point??

We have been noticing a situation on the market for some time, where
certain manufacturers and/or software makers tend to "stick together"
and help each-other lock out their respective competition.

A typical bond you would see was Intel-Microsoft (per CPU server license
anyone?). Since AMD took over ATI, nVidia has been leaning towards Intel
(their website contains this "interactive system builder" that only
provides matches for Intel CPUs). It is not too far fetched to believe
they would try Microsoft too, by only providing good drivers for Windows
(there is a Windows 7 ad on the nVidia front page).

LP,
Jure
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFKnaSjB6mNZXe93qgRApoZAJ9lksRKVJ0O+98tGFW/5RJELkwFmgCgxjrv
HXb5jhnn3bRwX5ecCmKPt3I=
=I/BJ
-----END PGP SIGNATURE-----

On Wed, 02 Sep 2009 00:48:03 +0200, Jure Sah wrote:

Hi,

As one possible explanation...

Dan Lenski pravi:
What is their rationale for *not* supporting FLOSS at this point??

We have been noticing a situation on the market for some time, where
certain manufacturers and/or software makers tend to "stick together"
and help each-other lock out their respective competition.

A typical bond you would see was Intel-Microsoft (per CPU server license
anyone?). Since AMD took over ATI, nVidia has been leaning towards Intel
(their website contains this "interactive system builder" that only
provides matches for Intel CPUs). It is not too far fetched to believe
they would try Microsoft too, by only providing good drivers for Windows
(there is a Windows 7 ad on the nVidia front page).

Interesting. So, you're saying that nVidia feels it has to cozy up to
Intel as a counterweight to AMD-ATI? That could explain nVidia caving
and allowing SLI for some Intel chipsets.

With nVidia's chipset business tanking, I think they must be hoping to
keep their revenue up by licensing SLI. My understanding is that SLI
should work on any system with dual PCIe x16 slots and appropriate BIOS
support (like ATI CrossFire does), and that software driver lockout is
the only thing that restricts it. This would explain why they don't want
to openly document their 3D hardware, as it would quickly lead to SLI
working on chipsets whose makers haven't coughed up for the SLI license.

I can also see the idea of nVidia cozying up to Microsoft.

Weird... I had been thinking in terms of nVidia as a market leader
guarding its advantage, but these strategies instead suggest a company
clinging to its proprietary advantage a dwindling market. Hmmm. Too
pessimistic?

Dan

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Daniel Lenski pravi:
Interesting. So, you're saying that nVidia feels it has to cozy up to
Intel as a counterweight to AMD-ATI? That could explain nVidia caving
and allowing SLI for some Intel chipsets.

I would call that a definite possibility, considering the point that AMD
and ATI are now indeed very supportive of eachother.

It's a great shame architecturally since nVidia technology kinda fits on
AMD CPUs. For the obvious, AMD + nVidia chipsets were natively
compatible since they both used HyperTransport very well... but there
was also the point that nVidia technology depends on high I/O
performance of the CPU, especially now that we have PCI-Ex, and Intel
never put much emphasis on this. Intel would blatantly sacrifice I/O
performance in favor of faster in-CPU processing (better "CPU --
cache", poorer "memory -- CPU -- hardware", with AMD the reverse is
true), which works much better than the alternative in cases of poor
programming.

With nVidia's chipset business tanking, I think they must be hoping to
keep their revenue up by licensing SLI. My understanding is that SLI
should work on any system with dual PCIe x16 slots and appropriate BIOS
support (like ATI CrossFire does), and that software driver lockout is
the only thing that restricts it. This would explain why they don't want
to openly document their 3D hardware, as it would quickly lead to SLI
working on chipsets whose makers haven't coughed up for the SLI license.

As far as I could find out, SLI and CrossFire are actually quite a bit
different from eachother. It goes a little like this...

PCI-Ex is basically a switched network (like a LAN), you have a switch
in the chipset and multiple connections going from it to the individual
slots (16x slots got 16 such connections, 1x got 1, etc). This design
obviously allows any device connected to communicate with any other via
the switch, though they usually all communicate with the CPU in a PC.

In SLI and CrossFire, the two graphics cards use this connection to
communicate to eachother, so they can use eachother's memory to render
3D (if you think this is a poor design, think twice about using SLI or
CrossFire, because you're right).

The thing with the little bridge is though that in CrossFire, the link
is actually the DVI data that would otherwise go to the screen, the
first graphic card in the sequence thus simply renders the missing parts
black, and the second one modulates this signal to include these parts
(or, AFAIK, they can also each render one frame and thus have to render
at a slower FPS, all depends on your settings). This is how they split
work. This is also a lot like what multi-GPU setups used to be in the
Voodoo 2 days (though then the connection was analog).

In SLI though, the bridge is a much simpler binary connection that the
cards use to help themselves communicate and it actually not always
necessary. Instead the cards communicate most of their data including
images using the PCI-Ex bus, the 'drawback' being that they cannot do
this alone but need a special chip present in the chipset (which also
connects to the PCI-Ex network) that helps coordinate them.

The conclusion being that while it may all just be in the settings with
CrossFire, SLI actually needs an extra piece of hardware to work.

I could be wrong though. For obvious reasons I can't confirm this data
with anything. It's not like one can't reverse engineer a BIOS though,
so if we want to be sure, we might as well just check.


I can also see the idea of nVidia cozying up to Microsoft.

Weird... I had been thinking in terms of nVidia as a market leader
guarding its advantage, but these strategies instead suggest a company
clinging to its proprietary advantage a dwindling market. Hmmm. Too
pessimistic?

nVidia is good, but that was never really enough to stay on the market
(Cyrix anyone?).

As the Intel example teaches us, it can be more important what your
consumers think is better than what actually is better. The motherboard
makers play a part in this too... they try to pick the 'best' hardware
to put in what they make and then sell only that. The resellers play a
similar role (Centrino laptops anyone?).

There is a lot of dirty tricks in use and the market leaders always were
best at using them.

LP,
Jure
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFKn+m3B6mNZXe93qgRAqDZAJ0VnJ9PDzYhZM7XI8OHeS CfxajTfACfQban
ySXZlEfCIG7o1Xr2SNPDOIk=
=V8Ah
-----END PGP SIGNATURE-----

On Thu, 03 Sep 2009 18:07:19 +0200, Jure Sah wrote:

Daniel Lenski pravi:
Interesting. So, you're saying that nVidia feels it has to cozy up to
Intel as a counterweight to AMD-ATI? That could explain nVidia caving
and allowing SLI for some Intel chipsets.

I would call that a definite possibility, considering the point that AMD
and ATI are now indeed very supportive of eachother.

Yeah, on the other hand I don't see how ATI+AMD really *prevents* nVidia
from making chipsets and graphics for AMD processors. Mostly, it just
seems to have raised the bar for the competition... the available
chipsets for AMD processors have improved a lot since the merger.

It's a great shame architecturally since nVidia technology kinda fits on
AMD CPUs. For the obvious, AMD + nVidia chipsets were natively
compatible since they both used HyperTransport very well... but there
was also the point that nVidia technology depends on high I/O
performance of the CPU, especially now that we have PCI-Ex, and Intel
never put much emphasis on this. Intel would blatantly sacrifice I/O
performance in favor of faster in-CPU processing (better "CPU --
cache", poorer "memory -- CPU -- hardware", with AMD the reverse is
true), which works much better than the alternative in cases of poor
programming.

Right. My first AMD64 box (http://www1.shopping.com/xPF-Acer-ASE360-MT-
A3500-1GB-200GB-DVDRW-NVIDIA-XP-MCE but with 1gb RAM) had the nForce4
chipset, and it was really impressive. Not expensive, but great I/O
performance and tons of features. GeForce 6100 IGP worked well too. If
nVidia made something equally good and cheap for Socket AM2+, I would've
probably bought that.

So you're saying Intel cranks up the CPU performance, benefiting poorly-
written CPU-bound code, while AMD cranks up the I/O performance?

As far as I could find out, SLI and CrossFire are actually quite a bit
different from eachother. It goes a little like this...

PCI-Ex is basically a switched network (like a LAN), you have a switch
in the chipset and multiple connections going from it to the individual
slots (16x slots got 16 such connections, 1x got 1, etc). This design
obviously allows any device connected to communicate with any other via
the switch, though they usually all communicate with the CPU in a PC.

In SLI and CrossFire, the two graphics cards use this connection to
communicate to eachother, so they can use eachother's memory to render
3D (if you think this is a poor design, think twice about using SLI or
CrossFire, because you're right).

The thing with the little bridge is though that in CrossFire, the link
is actually the DVI data that would otherwise go to the screen, the
first graphic card in the sequence thus simply renders the missing parts
black, and the second one modulates this signal to include these parts
(or, AFAIK, they can also each render one frame and thus have to render
at a slower FPS, all depends on your settings). This is how they split
work. This is also a lot like what multi-GPU setups used to be in the
Voodoo 2 days (though then the connection was analog).

In SLI though, the bridge is a much simpler binary connection that the
cards use to help themselves communicate and it actually not always
necessary. Instead the cards communicate most of their data including
images using the PCI-Ex bus, the 'drawback' being that they cannot do
this alone but need a special chip present in the chipset (which also
connects to the PCI-Ex network) that helps coordinate them.

The conclusion being that while it may all just be in the settings with
CrossFire, SLI actually needs an extra piece of hardware to work.

So if I understand you correctly... CrossFire transfers most of the inter-
card data over the bridge which is totally separate from the PCIe bus.
While SLI transfers the inter-card data over the PCIe bus itself, thus
requiring some chipset support?

I could be wrong though. For obvious reasons I can't confirm this data
with anything. It's not like one can't reverse engineer a BIOS though,
so if we want to be sure, we might as well just check.

Interesting! Did you do some hacking yourself, or read about these
CrossFire/SLI distinctions somewhere else?

Dan