The GT300/Fermi Thread - Part 2!

[Lightman]

@annihilat0r
new cat.10.3beta give 1000+ point in P preset of Vantage on HD5870, so you're right.
Besides we don't know if these scores are PhysiX on or off ...

[annihilat0r]

If the 3D mark vantage scores are correct this is bad. I mean in extrema GTX 470 actually loses out to 5850 Toxic "765 clock instead of 725"

But i also hope its a driver thing, because if its not a mild boosted 5870 would be able to compete against a GTX 470

I would prefer 19xxx and 9xxx from GTX 4xx !!

I'm not so sure about the those Tweaktown scores. I mean, other scores around the Internet give HD5870 around 16500P and 8000X. But maybe that article was written with better drivers or a better configuration.

[annihilat0r]

@annihilat0r
new cat.10.3beta give 1000+ point in P preset of Vantage on HD5870, so you're right.

Wow, that's amazing. And I still haven't installed those drivers! Such powerful a card is HD5870, I didn't even need to overclock it whatsoever or need better performance at all. No PC exclusive games = No need for better performance.

[trinibwoy]

And what is a reliable source ? B3D forum or what ?

I don't know how an entire forum can be considered a source but I'm not sure what you're asking. You do know what reliable means right? Trustworthy? Dependable?

The only SKU announced so far has 448 shaders. All the talk about 512 shader parts being available in the first batch of Geforce cards has been from forum posts and rumour mongering sites - i.e. unreliable sources.

[ajaidev]

"GeforceGTX470 in somehow in the 3Dmark Vantage is divided into sub-X7400 around, GPU score of 7100 points, while the same platform Overall score for 6400 run RadeonHD5850 around, GPU score of 6200 or so. 同时据称GeforceGTX470的核心/显存运行频率为810/3200MHz，采用320Bit GDDR5显存。 At the same time allegedly GeforceGTX470 core / memory operating frequency of 810/3200MHz, using 320Bit GDDR5 memory."

http://translate.google.co.in/transl...en-US:official


ahhahhaa 810/3200MHz wazza ??

[mindfury]

810MHz is impossible,even GTX480 core clock will not pass 750Mhz....


PS:
Core clock in GF100 is GPC clock（half of hot clock），not L2/ROP clock...

[annihilat0r]

Lol what, GTX 470 is 810 MHz? No way!

And if those scores are really from a 810MHz 470, then all the worse!

[zalbard]

If the 3D mark vantage scores are correct this is bad. I mean in extrema GTX 470 actually loses out to 5850 Toxic "765 clock instead of 725"

But i also hope its a driver thing, because if its not a mild boosted 5870 would be able to compete against a GTX 470

LOL, talk about getting your expectations high.

[ajaidev]

Since both scores are around 7k lets give some slack to nvidia and say its x7400 overall. Now my friends 4870 x2 gives about x7350 and my GTX 275 OCed gives about x5500, 5850's with better drivers such as 10.3 beta may give x7200 and 5850 overclocked may be able to give x7300.


So overall overclock 5850 would give about the same score as the proposed GTX 470 score? This is very troubling indeed also i am about sure that the 810 Mhz is wrong.

[dartaz]

From when 3Dmark vantage scores were important ?

I want to see a real gaming benchmark

[Olivon]

I don't know how an entire forum can be considered a source but I'm not sure what you're asking. You do know what reliable means right? Trustworthy? Dependable?

The only SKU announced so far has 448 shaders. All the talk about 512 shader parts being available in the first batch of Geforce cards has been from forum posts and rumour mongering sites - i.e. unreliable sources.

Xbit Labs is reliable or not ?

The flagship Fermi graphics processor will feature 512 stream processing engines (which are organized as 16 streaming multi-processors with 32 cores in each) that support a type of multi-threading technology to maximize utilization of cores. Each stream processor has a fully pipelined integer arithmetic logic unit (ALU) and floating point unit (FPU). The top-of-the-range chip contains 3 billion of transistors, features 384-bit memory GDDR5 memory controller with ECC and features rather unprecedented 768KB unified level-two cache as well as rather complex cache hierarchy in general

[FischOderAal]

Flagship doesn't necessarily mean GTX 480. For what we know the yields of the next revision might be good enough to harvest quite a few fully functional chips with 512 CUDA-cores. These chips then might be used for a Ultra 490.

Pure speculation on my part.

Additionally, the part of the report is just a list of what is known from the whitepaper... Xbit Labs doesn't claim GTX 480 will have 512 CUDA-cores either.

[trinibwoy]

Xbit Labs is reliable or not ?

They are just regurgitating high level specs from the whitepaper. I'm talking about the specific configuration of the GTX 480 SKU.

[Zed_X]

GTX 480 has 512SP, i thought it is a taken as a facts, where are you was last few weeks/months triniboy?

[Sushi Warrior]

Xbit Labs is reliable or not ?

I'd say not as their story says it's 384-bit, it is almost certainly 448.

[zalbard]

Since both scores are around 7k lets give some slack to nvidia and say its x7400 overall. Now my friends 4870 x2 gives about x7350 and my GTX 275 OCed gives about x5500, 5850's with better drivers such as 10.3 beta may give x7200 and 5850 overclocked may be able to give x7300.

So overclocking from 725MHz core to 1GHz is only 100pts? I think not.

[trinibwoy]

GTX 480 has 512SP, i thought it is a taken as a facts, where are you was last few weeks/months triniboy?

According to who? We have a million rumours about everything including the color of Jensen's underwear but hardly any real info on launch configurations, performance or pricing. We all know the chip has 512 SPs but that doesn't guarantee they will launch a card with all SM's enabled on March 26th.

[ElSel10]

Perhaps you need to read my question again. I didn't ask about the GF100 architecture. I asked about a specific product based on it.

Are you joking? First of all, the codename GF100 explicitly refers to the GeForce line of cards. And the Deep Dive was dedicated to the top GeForce part as well. And in the whitepaper, it says it refers to the top GF100 part.

What are you suggesting, that nvidia lied in the whitepaper and at Deep Dive?

We all know the chip has 512 SPs but that doesn't guarantee they will launch a card with all SM's enabled on March 26th.

Now you're just contradicting yourself and from the looks of it, making an attempt to troll. I thought the whole point you were asking was because we didn't know the chip has 512 SPs? And going from Nvidia's last generation, the GTX*80 was the top end single GPU part. There is no reason for that to be different this time around and we know the 480 and 470 are launching at the same time.

And when even Charlie has said it will have 512 SPs that's when you know it's time to stop arguing otherwise.

[mapel110]

The 480 will have 512 SPs. There is no need for a discussion about this fact.

[trinibwoy]

Are you joking? First of all, the codename GF100 explicitly refers to the GeForce line of cards. And the Deep Dive was dedicated to the top GeForce part as well. And in the whitepaper, it says it refers to the top GF100 part.

The education system is really failing humanity. GF100 is a chip. GTX 480 is a part based on that chip that may or may not have all SM's enabled. You know how the top announced Tesla SKU only has 14 of the SM's enabled though it's based on the same chip? Why is such a simple thing so difficult for you to understand?

@mapel, it's only a fact when Nvidia announces the specs. So far we only have random people on forums asserting "facts".

[mapel110]

@mapel, it's only a fact when Nvidia announces the specs. So far we only have random people on forums asserting "facts".

http://www.nvidia.com/object/gf100.html
Look in both PDFs. 512 Cuda Cores everywhere.

I am following the GPU-Buiness since 10 years. There was never the case, that the high end part was a cut down version.
And so it will be in March. Even if nvidia got only 10.000 512SPs at start, it would be dumb to not release this part.

[RaZz!]

http://www.nvidia.com/object/gf100.html
Look in both PDFs. 512 Cuda Cores everywhere.

I am following the GPU-Buiness since 10 years. There was never the case, that the high end part was a cut down version.
And so it will be in March. Even if nvidia got only 10.000 512SPs at start, it would be dumb to not release this part.

this time we have tesla cards as well as geforce cards. when nvidia talks about gf100 we don't know to which series of cards they refer to. there were rumors that the tesla parts would be higher binned cores than their geforce counterparts.

the only thing we can do is just wait for the final official specs to be announced.

[kadozer]

Are you joking? First of all, the codename GF100 explicitly refers to the GeForce line of cards. And the Deep Dive was dedicated to the top GeForce part as well. And in the whitepaper, it says it refers to the top GF100 part.

What are you suggesting, that nvidia lied in the whitepaper and at Deep Dive?



Now you're just contradicting yourself and from the looks of it, making an attempt to troll. I thought the whole point you were asking was because we didn't know the chip has 512 SPs? And going from Nvidia's last generation, the GTX*80 was the top end single GPU part. There is no reason for that to be different this time around and we know the 480 and 470 are launching at the same time.

And when even Charlie has said it will have 512 SPs that's when you know it's time to stop arguing otherwise.

Wow just wow. Stop taking Nvidia PR as gospel, they're almost as bad as Charlie.
Triniboy said it best.

[BenchZowner]

You don't know who's correct

You guys express in every possible way how much you hate nVIDIA and their tricky marketing plot lately, but you still post and fight over who's right or wrong in a meaningless posting war...
Since you're bored and tired of nVIDIA and Fermi, why can't you just stop posting and wait for the launch day ?

[mapel110]

this time we have tesla cards as well as geforce cards. when nvidia talks about gf100 we don't know to which series of cards they refer to. there were rumors that the tesla parts would be higher binned cores than their geforce counterparts.

the only thing we can do is just wait for the final official specs to be announced.

GF100 Graphics Architecture White Paper <-- do you expect to read something about Tesla is such a PDF?

And Tesla will be slower versions. Because these boards have more vram and need to be able to run 24/7 in computing enviroments.