NEW! ATI FirePro™ V8800 Professional Graphics
Antonio Huerta
Now you can play that elusive Crysis 2 at full settings !
Jim
features enabled+different drivers. The memory is 50MHZ slower and the 2GB
5870 just launched so this isn't an improvement. Crysis is a The Way its
Meant To Be Played game so the GTX480 is likely going to be the faster
anyway.
John Lewis
Hope they beefed up the Voltage Regulator modules over those on the
5870. Also hope that they improved the cooling over that of the 5870.
The Catalyst drivers from 9.xx onwards are kludged to throttle back
the clock rate on the 5870 by about 25% if the GP/GPU Stress program
Furmark is detected to avoid smoking the board. A great piece of
conservative design... NOT!! Furmark is a recursive algorithm, a very
common occurrence in GPGPU applications, which are also presumably
targeted by the V8800, otherwise why charge this ridiculous price??
The nVidia drivers have no such throttling and Fermi (GTX480/470) is
designed to run continuously at 90 degrees core-temperature for
maxed-out GPGPU applications. I doubt if many V8800 will be sold.....
John Lewis
Benjamin Gawert
> Hope they beefed up the Voltage Regulator modules over those on the
> 5870. Also hope that they improved the cooling over that of the 5870.
What for? How does the cooling and the Voltage Regulator (there are no
VRMs on gfx cards) limit the performance of any 3D application or game?
> The Catalyst drivers from 9.xx onwards are kludged to throttle back
> the clock rate on the 5870 by about 25% if the GP/GPU Stress program
"GPU stress programm" only. There is no "GPGPU" in FurMark.
> Furmark is detected to avoid smoking the board. A great piece of
> conservative design... NOT!!
It is in fact not a bad thing to prevent the card from damaging itself,
especially when looking at what damage FurMark did to some Nvidia GTX cards.
> Furmark is a recursive algorithm, a very
> common occurrence in GPGPU applications,
Nope. Recursion is not a common occurence in GPGPU applications, and
FurMark is a crap test for GPGPU performance as it is nothing more than
a synthetic OpenGL graphics benchmark which uses extreme parallelism to
create a load that is higher than any real-world application will ever
reach. Something which you could have found out easily had you bothered
to look at its website:
<http://www.ozone3d.net/benchmarks/fur/#techdata>
"What is FurMark?
FurMark is a very intensive OpenGL benchmark that uses fur rendering
algorithms to measure the performance of the graphics card. Fur
rendering is especially adapted to overheat the GPU and that's why
FurMark is also a perfect stability and stress test tool (also called
GPU burner) for the graphics card."
Using FurMark to judge the GPGPU performance of a gfx card is not just
naive, it is plain stupid. It is a program made to deliberately overload
the GPU, nothing more.
> which are also presumably
> targeted by the V8800, otherwise why charge this ridiculous price??
This might not be understandable for someone who barely knows anything
about consumer gfx market (and obviously nothing about the professional
gfx market) but there is a reason why the price of FirePro (and Nvidia
Quadro) cards are more expensive than your standard Radeon 5870 (which
is what the V8800 basically is), and no, it is not GPGPU. The simple
reason is that these cards are certified for many professional
applications, and often enough a requirement to even have the software
manufacturer's support talk to you.
> The nVidia drivers have no such throttling and Fermi (GTX480/470) is
> designed to run continuously at 90 degrees core-temperature for
> maxed-out GPGPU applications.
In your theory maybe. All modern GPUs use throtteling, and so does the
Fermi, a huge monster chip that converts almost 300W in heat. I would
assume that Fermi indeed can run GPGPU applications (if not this would
just anotther Nvidia embarassment), and so can the Radeon 5870 because
unlike what you think FurMark is *not* a GPGPU application and as
representative for one as is a snail for a sports car.
> I doubt if many V8800 will be sold.....
Yes, but then, you obviously don't have a clue about the market this
card is made for.
Benjamin
Rin Stowleigh
wrote:
Heh! I was going to ask "is that true?" about Catalyst drivers
governing the clock on Futuremark, but I will take your word for it.
What's funny is that recursion, by itself, has nothing to do with how
hard a processor works when comparing recursive vs non recursive
algorithms.. Recursive scenarios could potentially keep a processor
running at 100% for exponentially longer periods of time based on some
other condition incrementing only slightly (thus producing a condition
that might not have been expected, like the operation taking longer
and the processor getting hotter as a result), but even still if you
have some code that's going to execute a few hundred thousand
instructions, and a recursive routing that would (based on some input)
produce the same number of instructions, at the end of the day the
processor has worked just as hard.
So if what you're saying is true, it means that playing games too long
on these ATI cards will smoke them according to ATI's own tests.
Also, why would they throttle clock rate based on the application
using it instead of some other condition like temperature?
Shawk
Don't mind JL. He only wakes up to rant about ATI and Steam. He's
often wrong about Steam too.
Benjamin Gawert
> Don't mind JL. He only wakes up to rant about ATI and Steam. He's often
> wrong about Steam too.
So he's basically a bastard (a hybrid) of "Steamkiller" aka "PC GAMER"
and 1000s-of-Dells-administrator/Nvidia-chief-developer "Johns"?
Benjamin
Juarez
> It is in fact not a bad thing to prevent the card from damaging itself,
> especially when looking at what damage FurMark did to some Nvidia GTX
> cards.
All it did was show that many of the 7900GTs had ram that the
manufacturer factory OC settings couldn't do. I had to send 2 of those
suckers back for RMA due to under spec ram.
John Lewis
wrote:
>Am 08.04.2010 19:35, * John Lewis:
>
>> Hope they beefed up the Voltage Regulator modules over those on the
>> 5870. Also hope that they improved the cooling over that of the 5870.
>
>What for? How does the cooling and the Voltage Regulator (there are no
>VRMs on gfx cards) limit the performance of any 3D application or game?
Ah, yes our self-proclaimed PC-technology EXPERT replies.
Ummm, if there are no VRMs on graphics cards, then how does the +12V
input(s) get converted to the ~1-2volts that the GPU silicon requires?
VRM means Voltage Regulator Module, of which there are normally more
than one to share the low-voltage load current, considering that the
peak low-voltage GPU current can potentially exceed 150 amps
on a top-end graphics card.
I personally own a factory-overclocked eVGA GTX260 Core 216, clocking
at 675MHz instead of the default 576MHz. No problem with Furmark
stability test, no throttling whatsover. GPU Core temperature levels
out at 80 degrees C, with fan speed around 70%. Runs happily for the
duration of the default Furmark Stability test time ~ 15-20 minutes.
Great room-warmer.
Oh, by the way, Furmark is also a great check on the +12V
load-capability of the PC's main power-supply. Especially when
simultaneously combined with a CPU core- and main-memory-saturating
custom-profile of simultaneous instances of Prime95.
>Nope. Recursion is not a common occurence in GPGPU applications, and
>FurMark is a crap test for GPGPU performance as it is nothing more than
>a synthetic OpenGL graphics benchmark which uses extreme parallelism to
>create a load that is higher than any real-world application will ever
>each.
That's odd. A personal friend of mine is running massiively-parallel
real-time-video processing algorithms in CUDA on a GTX260. His unit
does get just a little toasty. I wonder why? Not quite as toasty as
with Furmark, but at least he is reassured that there is no hidden
driver throttling with massively-parallel computations. He will be
upgrading to Fermi shortly.
>Something which you could have found out easily had you bothered
>to look at its website:
I am very well-acquaintanced with the Furmark website. Thanks for the
arrogance. Hopefully you are not French.
John Lewis
Shawk
Except of course that they are way more entertaining...
Benjamin Gawert
>> It is in fact not a bad thing to prevent the card from damaging itself,
>> especially when looking at what damage FurMark did to some Nvidia GTX
>> cards.
>
> All it did was show that many of the 7900GTs had ram that the
> manufacturer factory OC settings couldn't do. I had to send 2 of those
> suckers back for RMA due to under spec ram.
Yeah, right.....
<http://www.geeks3d.com/20091209/geforce-gtx-275-vrm-damaged-by-furmark/>
Benjamin
Benjamin Gawert
> Ummm, if there are no VRMs on graphics cards, then how does the +12V
> input(s) get converted to the ~1-2volts that the GPU silicon requires?
It comes from a Voltage Regulator you moron. Gfx cards have Voltage
Regulators, but since they are soldered on the PCB they are just that,
Voltage Regulators. VRMs as the name already suggests are Voltage
Regulators on a *Module*.
>> Nope. Recursion is not a common occurence in GPGPU applications, and
>> FurMark is a crap test for GPGPU performance as it is nothing more than
>> a synthetic OpenGL graphics benchmark which uses extreme parallelism to
>> create a load that is higher than any real-world application will ever
>> each.
>
> That's odd. A personal friend of mine
Oh yes, just another hearsay story.
> is running massiively-parallel
> real-time-video processing algorithms in CUDA on a GTX260. His unit
> does get just a little toasty. I wonder why?
Sure you wonder, because you don't have a clue. Otherwise it would came
to your mind that even code that is completely different than FurMark
could also lead to high GPU load. But I guess in your mind just because
"it gets hot, too" means to you "it's the same". Sorry, it isn't.
> I am very well-acquaintanced with the Furmark website.
Maybe, but obviously not with the content, otherwise you wouldn't have
made yourself look like an idiot because of your silly claims. Next time
better do some basic research *before* spreading any nonsense.
> Thanks for the arrogance.
Thanks for the show. You have been quite entertaining ;-)
>Hopefully you are not French.
No. I mean, really, no.
Benjamin
John Lewis
wrote:
>Am 09.04.2010 19:10, * John Lewis:
>
>> Ummm, if there are no VRMs on graphics cards, then how does the +12V
>> input(s) get converted to the ~1-2volts that the GPU silicon requires?
>
>It comes from a Voltage Regulator you moron. Gfx cards have Voltage
>Regulators, but since they are soldered on the PCB they are just that,
>Voltage Regulators. VRMs as the name already suggests are Voltage
>Regulators on a *Module*.
The term VRM is currently used not only for modules containing
discrete elements on a mini-ecb or hybrid substrate, but also for
integrated silicon that not only contain the power-regulating
transistors/mosfets but also the controlling elements on the same
silicon substrate.
>
>>> Nope. Recursion is not a common occurence in GPGPU applications, and
>>> FurMark is a crap test for GPGPU performance as it is nothing more than
>>> a synthetic OpenGL graphics benchmark which uses extreme parallelism to
>>> create a load that is higher than any real-world application will ever
>>> each.
>>
>> That's odd. A personal friend of mine
>
>Oh yes, just another hearsay story.
>
Ah, yes.. you would make a great politician but a lousy lawyer.
>> is running massiively-parallel
>> real-time-video processing algorithms in CUDA on a GTX260. His unit
>> does get just a little toasty. I wonder why?
>
>Sure you wonder, because you don't have a clue.
the "I wonder why?" was rhetorical. Your prejudices cripple your
comprehension/deductive skills.
>Otherwise it would came
>to your mind that even code that is completely different than FurMark
>could also lead to high GPU load. But I guess in your mind just because
>"it gets hot, too" means to you "it's the same". Sorry, it isn't.
>
Furmark's parallelism makes it the ideal candidate for corner-testing
a GPU intended for GPGPU computations - just as Prime 95 if correctly
configured is a superb corner-test for multicore CPUs. Maybe you would
like to suggest a more searching/representative GPGPU test that will
simultaneously reasonably exercise all the stream-processors?
>> I am very well-acquaintanced with the Furmark website.
>
>Maybe, but obviously not with the content, otherwise you wouldn't have
>made yourself look like an idiot because of your silly claims. Next time
>better do some basic research *before* spreading any nonsense.
>
.... in your oipinion, of course.
In the near future, I expect to be running a Furmark profile on a
GTX470 specifically intended for GPGPU applications. Will keep you
posted with regard to throttling or any other unusual behavior..
>> Thanks for the arrogance.
>
>Thanks for the show. You have been quite entertaining ;-)
>
>>Hopefully you are not French.
>
>No. I mean, really, no.
>
Oh, good.
>Benjamin
>
>
John Lewis
John Doe
Benjamin Gawert <bgawert gmx.de> wrote:
> Path: news.astraweb.com!border5.newsrouter.astraweb.com!news.glorb.com!news2.glorb.com!fu-berlin.de!uni-berlin.de!individual.net!not-for-mail
> From: Benjamin Gawert <bgawert gmx.de>
> Newsgroups: comp.sys.ibm.pc.games.action
> Subject: Re: NEW! ATI FireProT V8800 Professional Graphics
> Date: Thu, 08 Apr 2010 22:26:15 +0100
> Lines: 71
> Message-ID: <826vvjFo2rU1 mid.individual.net>
> References: <dc313744-5328-48b5-98e5-756e5de99231 u22g2000yqf.googlegroups.com> <qP2dnRDiMY3iISDWnZ2dnUVZ_hednZ2d giganews.com> <4bbe1e40.6804140 news.giganews.com>
> Mime-Version: 1.0
> Content-Type: text/plain; charset=ISO-8859-15; format=flowed
> Content-Transfer-Encoding: 7bit
> X-Trace: individual.net xtkK/6v3SkyI/IZC2hsG3g+5Tt254LITX75pihO9DxTdStYK4=
> Cancel-Lock: sha1:ClRSCyPs5kWsjF0UF0FpmNAN72s=
> User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; de; rv:1.8.1.12) Gecko/20080213 Thunderbird/2.0.0.12 Mnenhy/0.7.5.0
> In-Reply-To: <4bbe1e40.6804140 news.giganews.com>