FERMI Benchmarked?

[bladerash]

But bladerash there is a clue in the name of that website... SEMI-accurate lets hope they hit 600Mhz and wanted to go the whole at least 700Mhz hog with A3!

Well.. if the slides are true, it will instantanous turn HD5970 from fastest to most suckiest GPU due to power consumpion at ~300 watt, while the GTX 380 only eat 225 watt with way better performance.

[Hornet331]

meh about this, but if performance is somewhat near that i have to sell my HD5870 and get myself a 380gtX

[skugpezz]

Nice try nvidia lol

[JohnZS]

Well.. if the slides are true, it will instantanous turn HD5970 from fastest to most suckiest GPU due to power consumpion at ~300 watt, while the GTX 380 only eat 225 watt with way better performance.

Exactly, which is why I hope the slides are true... however we all know that people can use photoshop and make things up

[skugpezz]

if the 5950 get out before fermi I am buying the 5950. Nice attemp st sabotaging the competition nvidia lol.

[Utnorris]

The reality is that Fermi is coming out 6 months or more after the HD5800 series, so it has to have better performance or otherwise it will be a BIG OLE FAIL. As far as the numbers go, I can see these being plausible, especially on games that favor Nvidia, which is probably why these were used. Notice that there isn't any DX11 comparison in there and that would have been just as easy. As I stated already, Fermi should beat the HD5800 series, but they first have to get them on the shelves and they better be ready for a price war. I guarantee that as soon as Fermi is released the HD5800 series will drop in price to $199 (5850), $299 (5870) and $399 (5970). They raised the prices up and have been careful with the supply to keep inline with keeping demand high, this will allow them to make those price cuts when the time comes.

[prava]

Nah, totally nonsense.

Perf/W from GT300 would have to be utterly amazing compared to Evergreen which will never happen (we never had this difference before) as they rely on TSMC.
We can expect 30% gain from GTX295, which could get nearly to 5970 with an "Ultra" model. Obviously those massive gains from Evergreen are impossible, as we never had such difference between same gen. cards.

[ajaidev]

The tomshardware slides are totally fake, fermi will most likely come in 786MB and 1.5GB of memory... AFAIK fermi has 6 memory controllers so 256*6=1.5gb

[Solus Corvus]

Nice if true. But I want to see benchmarks from an objective 3rd party. If these are marketing slides from nvidia or just the creation of a fanboy, they certainly don't come from an objective source.

[Hell Hound]

They need something to counter Eyefinity because we already have plenty of power in 295/5970.When crysis 2,lost planet 2 and games like Battlefield 3 come The 6 series will be among us.With better Eyefinity support.I've also read that Simbin is working on a new title,unless Nvidia is giving away Triple Head 2 Go with their cards,I don't see them having the upperhand.

[Vit^pr0n]

These are old and fake. Saw these on OCuk like 2-3 days ago. Nothing new.

The results for the 5800's are also questionable in this fake benchmark. A 5870 gets over 100 fps at those settings in RE5 for example.

Thread should be closed imo. Until we get benchmarks from actual sources like Nvidia themselves, then this is all just nothing but hype.

[grimREEFER]

faster than 5970= more expensive than 5970

they could get away with selling this thing at $650, maybe even more if ati doesnt cut 5970 prices

[Solus Corvus]

Does Nvidia usually publish their slides in jpeg format or PNG?

[annihilat0r]

absolutely fake. Fermi is a single card and what'll it TDP be? 200W? You can't beat a 300W card with a 200W one at the same node. It just doesn't make sense.

[Solus Corvus]

absolutely fake. Fermi is a single card and what'll it TDP be? 200W? You can't beat a 300W card with a 200W one at the same node. It just doesn't make sense.

You could if the 200W one was a more power/performance efficient architecture. But for reasons I have stated in the past, I don't think Fermi is going to be more power efficient then Cypress.

[annihilat0r]

Obviously you could, but it would have to be INCREDIBLY efficient in comparison to ATI's current architecture. And from what we hear about optimization towards GPGPU, I don't see that happening.

[dengyong]

Thread should be closed imo. Until we get benchmarks from actual sources like Nvidia themselves, then this is all just nothing but hype.

That's no reason to close the thread.

There's nothing else to discuss on the matter right now, so why not this.

We should be able to discuss whatever here without being childish.

[Smartidiot89]

absolutely fake. Fermi is a single card and what'll it TDP be? 200W? You can't beat a 300W card with a 200W one at the same node. It just doesn't make sense.

The manufacturing process doesn't matter as much as the architecture But yes it does sound very unreasonable and I'll call fake until we hear something from Nvidia.

[RPGWiZaRD]

This reminds me of the hype just before G80 and then it actually turned to be true.

1.8GB and 2GB sounds like a lot of memory for GPUs today but NVIDIA usually pushes the capacity to new heights for a new gen, 768MB for a 8800GTX was an insane amount when it was released as well.

However one thing I can't simply believe in is the power efficiency, either the TDP is very inaccurately given as to what realworld figures would show or these numbers are just made up. As for performance, it's definitely looks a bit too positive but not totally impossible but due to the respins/delays, I'm really doubtful. If performance figures would be accurate nvidia could charge like $699 and $599 respectively unless ATI would adjust prices, that alone says a lot...

[ajaidev]

This reminds me of the hype just before G80 and then it actually turned to be true.

1.8GB and 2GB sounds like a lot of memory for GPUs today but NVIDIA usually pushes the capacity to new heights for a new gen, 768MB for a 8800GTX was an insane amount when it was released as well.

However one thing I can't simply believe in is the power efficiency, either the TDP is very inaccurately given as to what realworld figures would show or these numbers are just made up. As for performance, it's definitely looks a bit too positive but not totally impossible but due to the respins/delays, I'm really doubtful. If performance figures would be accurate nvidia could charge like $699 and $599 respectively unless ATI would adjust prices, that alone says a lot...

But how AFAIK fermi has 6 memory controllers so that means 786MB and 1.5GB of memory...

[Teemax]

For me, it's pretty much 100% fake for a simple reason: NVIDIA PR dude said "Dual Fermi coming soon".

If GTX 380 is in a position to compete against HD 5970, NVIDIA wouldn't have to rush a dual Fermi out so soon.

[highoctane]

The performance would be insane if true but that would also mean fermi would probably have to be clocking like a beast or have specs beyond what we currently assume which seems doubtful to me.

I would expect it to be faster than the 5870 but it seems to be reaching for it to surpass the 5970 in situations where crossfire implementation is working.

[annihilat0r]

For me, it's pretty much 100% fake for a simple reason: NVIDIA PR dude said "Dual Fermi coming soon".

If GTX 380 is in a position to compete against HD 5970, NVIDIA wouldn't have to rush a dual Fermi out so soon.

This.

Plus, what's going to be Fermi's TDP? Any info on that?

[Vit^pr0n]

That's no reason to close the thread.

There's nothing else to discuss on the matter right now, so why not this.

We should be able to discuss whatever here without being childish.

Well this isn't exactly news. News have sources, and this source is just a 'forum member' that posted slides out of thin air. If sources could be pretty much anyone now, this forum would be filled with tons of BS news.

[tajoh111]

I think its very plausible, but the source is dubious. So I agree with anyone professional and take this with a huge grain of salt.

Honestly everyone expected and believed the 5870 was going to have perfect crossfire scaling of 4890s i.e exactly twice the performance of the 4890 series, I don't think a single person on this board doubted that. I though it was going to have perfect CF scaling and when the card came out and it didn't, I was let down and didn't want to purchase a 58xx anymore. When the 5970 came out, I was even more let down because from AMD slides, I was expecting significantly better performance. But the increase in price and decrease in performance, really made me not want to buy the card anymore.

When benchmarks essentially claim the same thing, but from NV, people are claiming it is too fast, it impossible, even when fermi specs, indicate, it has more than twice the shaders, those shaders are 16% faster and on top of this, its a new architecture.

So why do people believe it when it comes to AMD perfect CF scaling is possible, but from NV, it can't be, its impossible, when specs and a change in architecture indicate the NV scenario is more likely to be true?

http://www.xtremesystems.org/forums/...d.php?t=233621

Heres the original thread.