BRIGHT SIDE OF NEWS About | Advertise | Contact BSN USER Login
| Register 		SUBSCRIBE Newsletter | RSS Feeds
Saturday, May 18, 2013
Email this to a friend.
Your friend's e-mail:
Your Name:
Your e-mail:
Message subject:
Comments
Phermi by: Anonymous on 11/25/2009
"To say "wait until 2010 for our OpenCL" is irresponsible behavior when it comes to multimillion dollar investments. Business doesn't wait."

I couldn't agree more with you. But then again you jump over previous conclusions mentioned in above text (i hope you didn't just cp them from somewhere)
_"... and like NV30 paved way for NV40/45/47/RSX [essentially, GeForce 5800, 6800, 7800, 7900 and RSX are the same]. Do note that NV35 [FX5900] was "an anomaly", with added fixed-function shader hardware and wider memory controller to make GeForce 5800 architecture usable."_

The whole Brook C business was way longer present in nVidia science oriented community even when you're all laughed on there NO2 upgraded NV25 architecture with all that fabulous things that undermined their performance in 3D applications they had with their previous GF4Ti (NV25) generation. Rarely anyone of you so called IT journalists covered anything of the nV idea that was going on back then trying to brought us GPGPU.
So to conclude nV has their trials since 2002 in GPGPU business while ATi for the same reason, that they were late to market with similar solution up to 3yrs, nobody took seriously with their GPGPU trials and in fact they have to adopt things that might or might not be good for their GPGPU development. So don't jump to some short sight conclusions about ATi business in OpenCL. Yep i hate that they still don't deliver and cope withe seasonal changing strategies in their GPGPU business. But it's hard knock life for them too when there's not enough dev flock in their herd, and ever changing strategies certainly doesn't help them in that matter.

And NV30 was just an upgrade onto an old NV10-NV25 architectural path, an original GeForce hybrid if you want. While NV40 was *the real thing* which architectural developments are still there in even _freshly designed_ G300


@Kyocera

"However, I must ask myself, can Nvidia be trusted? Are the GeForce GT300 chips first choice ones? Or just leftovers; inferior and semi-defective, deemed good enough for PC users ?
While Nvidia makes a nice pile of first choice chips for their Professional line due few months later?"

That's the one of the problem i stuttering about when AMD comes into my mind. They are doing such policy since first K8 see daylights in server racks and all companies try to follow that kind of lead. AMD never even try to develop or tweak some part of their K8/K8L to meet desktop uses demand when it comes to performance they just said we develop server things and the junk leftovers will be served to desktop market. So not that we don't have somewhat better single socket performers but we also stuck with power hungry damaged chips that couldn't compete themselves into server racks. And it's funny thing to see how some Magny-Cours chip could easily reach 3.2GHz on 1.05V while for some desktop future hexa Phenom II (codenamed Thuban) users we'll probably be happy to see how it reaches same only on 1.2V or above. And to add to all that sometimes even single socket desktop offerings are priced higher 30% or more than some top Opteron HE offerings.
by: Kyocera on 11/20/2009
Great synopsis Editor in Chief!!

(you should put it in a separate article, since it will resurface many times in upcoming month).
RE: by: Theo Valich on 11/20/2009
We will be taking a detailed look in NV100 architecture, and also look into nVidia as a company. NV100 is an interesting generation of products, especially given the look into ECC penalty across the board. But everone should be aware of the fact that this is the first generation product this complex and it works. Yes, there are issues, but it works. Intel took more than three billion to make Larrabee work and is still much farther than NV100 is now, nVidia did the same in around 400-500 million budget.

As far as comments and constructive criticism goes, yes, we do try our best to get to the bottom of the story. We're not perfect, but we're trying our best to give you the look at what's really going on. We are going to be right in a lot of things and we are going to be wrong from time to time. What we can do is continue to push with the content and have "yield" as close to 99%. But neither nVidia, AMD, Intel or just about anyone else is able to be 100% correct. The list of errata in Intel and AMD CPUs just confirms that.

The companies involved have to stand by their words, it is as simple as that. nVidia can say all they want, but the fact is that the company is not honest nor open, and they want to go ahead and compete against Intel who gives as open roadmap as possible. We know Gulftown performance and clocks since May 2009, their estimated TDP was correct etc.

When it comes to AMD and their GPGPU strategy, I don't think anyone can say they're executing as they should - all the talk of OpenCL is nice, but AMD - show me the hardware and software.

To say "wait until 2010 for our OpenCL" is irresponsible behavior when it comes to multimillion dollar investments. Business doesn't wait. Hence the reason why the world most powerful computer in 2011 will probably use AMD CPUs, but an array of tens of thousands of NV100-based GPUs.

Lastly, I don't see that we are taking the route set forth by alleged competition. BSN* has a clear vision of covering the technology and tech impact and we're not interested in games, which all companies working with us are feeling.

What I will do is admit that it is taking much longer to bring all the features online, as this is my first company where I am CEO and Ed-in-Chief. But steps are taken to increase the quality of the site, and I hope that the team as we have it now will continue to expand with such unique individuals.

My biggest challenge is to ensure the steady flow of content to the site, as we set forth to be a news generator, rather than being an aggregator. We all hope that you enjoy the content here and will continue to push in the future.

Ed.
by: Kyocera on 11/20/2009
In order to save energy?? down-clock?

Buy yourself a solar-photovoltaic panel.
:)) :))

and eat less meat; the animals produce more greenhouse gases then all cars on earth together.

http://animals.howstuffworks.com/mammals/methane-cow.htm

And you may also contribute; a plug??
Fermis Specs by: Anonymous on 11/20/2009
To my knowledge the Fermi for Supercomputers consumes 190 Watt tipically and more than 200 Watt during peak usage. This is already more than the 188 Watt of Cypress and if we take into account that the Fermi card is already a downclocked version the Fermi Card for gamers must draw even more, perhaps 250 Watt which I consider way too much for a single core card. In order to save energy nvidia will be forced to downclock its gamer cards too with significant impact on performance. I do not believe that Fermi will perform any miracle. More likely they will be hardly better than a 285.
Tflops... by: Anonymous on 11/19/2009
din anyone knows what's AMD GPU's efficiency when it comes to tflops?

it's close to 50% :)

take FAH for example :)
by: Kyocera on 11/18/2009
Comments on update 1 and 2.
As a client I expect fair tretmant from a company. No matter if the product sells for 10 dollars or the company presumes, I'll be only an one-time client.

Your company, or better said your colleagues in the GeForce department, Mr.Humber and Mr. Keane, are a flowery example of turpitude.
It is meaningless to go into why Nvidia's repute took a plunge among common users, neither to go into your marketing practices (including the price roller coaster with the 8000 series). Apparently they and their decried practices enjoy support in the management.
It is too much to aks from Nvidia to give a general idea what to expect? Nobody needs secrets; it is enough to give a general idea what to expect if things go as planned and the steps (silicon respins) to guarantee a really valid product.
Instead you decided to enjoy in a game; a diseased game of subterfuges, wandering, deception, double-dealing,... One would think, that your GeForce department employes suspense writers for life-drama or horror films. The only thing missing is the sound scenery.
All this was not enough; you cared to go into an injuring explanation how small customers account for 75% of volume but negligible profit; just in the middle of the heated anticipation talk from home users-thank you; at least we know now, that we are not worth much to you.
The only thing that you forgot in your greed and sublimity is, that You need volume; no foundry will buy machinery and employ engineers to fabricate that few supercomputers and professional workstations; no matter how good you plan to peel them (update 1 from BSN).

A great deal of customers did not care (any more) to get their heads filled by Nvidia, they just turned away and purchased ATI. It's good that BSN decided to look at the product and not at the company, since they are keeping it above water from early this year, while your own colleagues try to prove that stones float on water.

Without TSMC problems, you would probably finish as a fish on dry ground (the "famous" Charlie even wrote a few days ago, that the TSMC problems have a very advantegeous timing; but this are his Nvidia complots).

When some of the PC users applied just a drop of you medicine on yourself, with the GFlops discussion, you reacted quite quickly. Or were you only woried that your "drama" game took a wrong turn on you? (end of manipulation?) The Facebook picture?

Contrary to some believes, small (too)informed customers know the difference between a rack system (cluster load/durability/failure %) demands and the completely different PC usage pattern.

Don't understand me wrong; your product is amazing in every respect. ATI is just serving a warmed-up meal, spiced with a shovel of very spicey Mexican spices, to cover the main dish underneath.

However, I must ask myself, can Nvidia be trusted? Are the GeForce GT300 chips first choice ones? Or just leftovers; inferior and semi-defective, deemed good enough for PC users ?
While Nvidia makes a nice pile of first choice chips for their Professional line due few months later?
FUD is FUD by: Anonymous on 11/18/2009
History Revolvo Ipsum - History repeats itself To us, NV100 i.e. GT300/Fermi architecture is looking more like NV30 every day.

Pure FUD statement: and a Direct quote from your article.

The higher clocked Quadro's and the still faster clocked GeForce's of the Fermi have not yet been released.

You have also received information from nVidia that the Tesla had ECC enabled that had significant performance penalties.

Will you be retracting your above conclusion?

I expect Charlie to distort and spread the FUD why is BSN following his style?

And why did you not contact nVidia's General Manager of Tesla Business before drawing your conclusion about the whole GeForce, Quadro, Tesla Fermi architecture?
by: Anonymous on 11/17/2009
if everybody say that 5870's 2.7 TFLOPS is slower than 295's 1.8 TFLOPS.

then how to explain 295's 1,8 TLOPS against Fermi's 1.26 TFLOPS. its number is slower than 295's....
by: Kyocera on 11/17/2009
I think that the following video demonstrates quite nicely the Fermi consumer approach and marketing till today.

http://www.youtube.com/watch?v=UEHVbD-cttc&feature=related
by: Kyocera on 11/17/2009
First reviews from ATI 5970 are coming online on far east; I hope you'll follow soon.

By the way; Intel's new G2 Postville SSDs are on market only on paper; the whole of Europe was not able to order supplies for more than a month; the majority of German shops have listed mid-December as the first incoming shipment....
Nobody is reporting or commenting the issue.
The wholesale is speaking about heavy production difficulties....

Why are there no reports and coverage???
@FUD by: Theo Valich on 11/17/2009
We have cited both C2050 and C2070, as the presentation states that both cards are 520-630 GFLOPS, with the difference in memory size and price [$2499 for C2050 and $3999 for C2070].

The numbers are a let down form the previous hardware generations:
NEW: GeForce 7800GTX - 160 GFLOPS
Refresh: GeForce 7900GTX - 184 GFLOPS
NEW: GeForce 8800GTX - 330 GFLOPS
Refresh: GeForce 9800GTX+ - 476 GFLOPS
NEW: GeForce GTX 280 - 933 GFLOPS
Refresh: GeForce GTX 285 - 1063 GFLOPS
NEW: GeForce GTX 380 - 1260 GFLOPS?

If you want to compare Tesla generation, gladly:
Tesla C870 - 518 GFLOPS / No DP
Tesla C1060 - 933 GFLOPS / 78 GFLOPS DP
Tesla C2050 - 1040 GFLOPS / 520 GFLOPS DP
Tesla C2070 - 1260 GFLOPS / 630 GFLOPS DP

If we don't look at DP numbers, it is obvious that NV100 GPU comes with a significant price/performance penalty.
C870 - 800 USD
C1060 - 1499 USD
C2050 - 2499 USD
C2070 - 3999 USD

The math is fairly easy to calculate.

Ed.
all this is specualtion by: Anonymous on 11/17/2009
There is NO PRODUCT to benchmark.

All this jaw flapping, is just that. All talk and no substantial information to base any of the comments upon.

Time to put the nVidia attack dogs back in their kennels.
I call FUD by: Anonymous on 11/17/2009
Why did BSN purposely choose the lower clocked C2050 and not the higher clocked C2070 to form their opinion. Only one reason: FUD.

The Tegra C2070's DP 630GF equates to SP 1260GF or 35% faster SP than the C1060.
by: Kyocera on 11/17/2009
""It'll be interesting to see if they're going to even market this card in the consumer space.

Exactly my thoughts.

They might put it on the market for a very competitive price against ATI without providing more then very very limited quantities.
In this manner they can sell each hard awaited silicon from TMSC as a 2.500 dollar Tesla unit. Considering that the market for Tesla is going be huge in the first 10 months from introduction,....""

I hope Nvidia does not read BSD; otherwise they just got an idea (additional money will certainly buy a Mac for the dog of the CEO).
by: Anonymous on 11/17/2009
"For all intensive purposes Nvidia has been cutting back rendering for performance under the user's nose for years now, people are just too jaded to care."

*facepalm* it's "for all intents and purposes" not "for all intensive purposes"...making yourself look like an idiot...
by: Kyocera on 11/17/2009
Requiem.
(for consumer GPU; sacrificed on the altar of high money flow supercomputing).
by: Frantz on 11/17/2009
@anonymous below,

Whatever the ratio difference between the 2 arch, the resulting values for DP and SP are:

ATI Radeon 5870: 544 GigaFlops in DP
2.7 TeraFlops in SP
NVIDIA Fermi: 630 GigaFlops in DP (best case)
1.5 TeraFlops in SP
by: Anonymous on 11/17/2009
I love how everyone's comments are reviewing and critiquing the cards and new architecture before they're even out or before any official benchmarks have been released.
by: Anonymous on 11/17/2009
In ATi's dreams...
ATi's cards have about 1:5 ratio between SP and DP FLOPS. A 1.2 TFLOPS SP Radeon gets about 240 GFLOPS DP.
nVidia has a 1:2 ratio in this new architecture.
Failmi by: Anonymous on 11/17/2009
So can I buy one today?

No.

I *might* be able to buy one in FIVE MONTHS time.

Never mind that the theoretical single precision is lower than the actual achievable SP performance in ATI's currently shipping product line, and the double precision isn't too far off.
by: Kyocera on 11/17/2009
Gaming performance? Advancement?
Nothing till 28nm shrink?
by: Anonymous on 11/17/2009
There are plenty of logical reasons.
Point is that FLOPS is not an accurate assessment of real-world performance. It's a purely theoretical figure, even less meaningful for real-world performance than synthetic tests.
We've known this in the CPU-world for decades... In the early days CPU performance was expressed in terms of MIPS (Millions of Instructions Per Second), the integer-counterpart of FLOPS. Even though standardized synthetic benchmarks were developed for MIPS/FLOPS, such as whetstone/dhrystone, in the end it was abandoned because as CPUs become more advanced, these tests lost their meaning. MIPS was closely related to the MHz myth... and with GPUs and FLOPS we're going through that all over again.

And the last thing we need is tin-foil-hat characters such as you, telling us that developers must be cheating if the card with the highest FLOPS doesn't win every benchmark. Get a clue.
by: Anonymous on 11/17/2009
@post below me:

The GT200 series won benchmarks by cheating and cutting corners in rendering.

If a game is coded properly it will utilize the FLO output on a consistent scale.

There is no logical reason for a card with a lower FLO output to perform better than one with higher output unless the operation required some proprietary standard or API like CUDA.

For all intensive purposes Nvidia has been cutting back rendering for performance under the user's nose for years now, people are just too jaded to care.

so if a game performs better on weaker hardware, it isn't the fault of the hardware.

It's the fault of the devs for making a poor engine.
by: Anonymous on 11/17/2009
The raw speed in TFLOPS might not have improved much, but that doesn't mean the GPU isn't faster than a GTX285 anyway.
As we know from 4870/4890 vs GTX280/GTX285, the card with the highest TFLOPS rating isn't always the fastest.

If nVidia has improved efficiency of their architecture, it could indeed be faster than a GTX285.
I'm quite sure that efficiency has improved lots in GPGPU tasks at least, because with the full C++ support in Cuda you can write much more efficient code than before.
by: Anonymous on 11/17/2009
welp

http://www.youtube.com/watch?v=gyZDZCGQJf8
© 2009 - 2011 Bright Side Of News*, All rights reserved.
© 2009 - 2013 Bright Side Of News*, All rights reserved.