AT fermi review is up!

[2March]

its not a bug its a feature...



actually it seems to have been due to the architecture being borked

Most likely both. The drivers can't get the card into 2D power saving mode and there is something wrong with the chip itself aswell.

Sounds like a rush job allright.

 

[Janooo]

Hell... it gets to be 30C here in Canada in a few months... Who knows how that will mess with things..

Fermi will be responsible for the global warming.

 

[Xarick]

I am actually very surprised to see card makers actually putting out reference design cards. I would expect better cooling solutions right off the bat.

 

[Qbah]

People who are asking about the 2 screen 90C thing... did any of you bother to visit the link this information was referring to? There's like a blue-colored paragraph with nVidia's response... Seriously, at least bother to read the source...

"We are currently keeping memory clock high to avoid some screen flicker when changing power states, so for now we are running higher idle power in dual-screen setups. Not sure when/if this will be changed. Also note we're trading off temps for acoustic quality at idle. We could ratchet down the temp, but need to turn up the fan to do so. Our fan control is set to not start increasing fan until we're up near the 80's, so the higher temp is actually by design to keep the acoustics lower." - NVIDIA PR

 

[Skurge]

I am actually very surprised to see card makers actually putting out reference design cards. I would expect better cooling solutions right off the bat.

That reference cooling is pretty beefy in its own right, like the 5970. Except without the noise.

 

[Skurge]

People who are asking about the 2 screen 90C thing... did any of you bother to visit the link this information was referring to? There's like a blue-colored paragraph with nVidia's response... Seriously, at least bother to read the source...

Yeah, but idle power is still quite alot.
I'd have it that way though. higher temps at idle for less noise.

 

[sciwizam]

Has anyone considered the ambient temps during reviews are in the lows 20s. temps here in Africa usually hover around 35-40c so It could become a serious problem.

Hell... it gets to be 30C here in Canada in a few months... Who knows how that will mess with things..

Already posted in another thread, though probably using Furmark:

That's very nice I'm sure, but 100% irrelevant to people who play video games, particularly as I understand they've majorly gimped all that for the 'gaming' cards

As it is, it then appears to be deadweight, and damn power-hungry deadweight at that...

EDIT: Perhaps more importantly, do you think I could run a GTX 470 on a reasonably decent 500W PSU? Q6600@3.4, 4GB DDR2, 1 HDD, 1 optical drive is the rest of the system, in an Antec Sonata III with the stock 500W PSU (Earthwatts 500W with 2x 17A 12v rails).

You don't have to understand French, but take a look at these temp. readings with that same case. They have an ambient temp. of 26C. I'm pretty sure it gets hotter than that in Australia.

 

[RussianSensation]

A 4850 sells today on Newegg for $89.99 with free shipping.

That's the best value in the history of video cards.

Don't forget dual 5770s for < $260. That's as fast a single 5870!
There's also a lonely BFG 8800GT for $45 on Newegg (right now out of stock though). 2 of those budget cards for $90 should be decent too

 

[lopri]

For 8x MSAA, NVIDIA just made some ROP changes. They have better compression and the additional ROPs help offset the penalty for AA'ing smaller primitives (which don't compress well).

As for why we're seeing the performance gap close between the cards at higher resolutions, at this point I don't have a good answer for that.

Thank you for the explanation. I thought it's kind of counter-intuitive since those two tend to go hand in hand. (high res & AA performance)

 

[lopri]

I don't mean to beat up the latest and greatest from NVIDIA, but this chip does have some problems, it seems. Even without comparing head to head against Cypress, there are glaring issues for a microprocessor from what I've read so far.

1) Idle power consumption and temperatures @50 MHz. Realistically, 50 MHz is, if at all, likely the lowest frequency a modern ASIC can maintain, and I doubt it can go any lower while functioning. And at the very minimal (possibly borderline) frequency, the chip is already burning.

2) If you watch the noise/power recording of the 480 @H, you can see the power consumption continuously rising (up to extra 50W) under same load as temperature rises. It's a known characteristics of 'leaky' chips.

3) In the same vain, the SLI power consumption according to AT review shows something is very off. The difference between single 480 and dual 480s is 372W, which is way over the official TDP. I'm guessing both cards under SLI are running outside the spec, and the magnitude is alarming.

 

[Silverforce11]

I don't mean to beat up the latest and greatest from NVIDIA, but this chip does have some problems, it seems. Even without comparing head to head against Cypress, there are glaring issues for a microprocessor from what I've read so far.

1) Idle power consumption and temperatures @50 MHz. Realistically, 50 MHz is, if at all, likely the lowest frequency a modern ASIC can maintain, and I doubt it can go any lower while functioning. And at the very minimal (possibly borderline) frequency, the chip is already burning.

2) If you watch the noise/power recording of the 480 @H, you can see the power consumption continuously rising (up to extra 50W) under same load as temperature rises. It's a known characteristics of 'leaky' chips.

3) In the same vain, the SLI power consumption according to AT review shows something is very off. The difference between single 480 and dual 480s is 372W, which is way over the official TDP. I'm guessing both cards under SLI are running outside the spec, and the magnitude is alarming.

Those are good points. Charlie claimed they have a poor architecture on 40nm that is very leaky, especially at high temperatures. That was months ago. And its definitely true because the 480 has been benched to pull over 300W in 3dmark and near 300W in games. Its well over the official TDP. The 372W (AT) was taken from the wall, so an efficiency of 87%, its ~320W which is what it peaks at in furmark.

Also, the leaky nature means that even at idle speed the voltage/amps still remained high else it would cause instability.

Seriously though, if the card ran cooler and used around 250W peak, it would be a fantastic GPU. A 10-15% lead over a 5870 for a slightly more power usage and cost is fine. But more power hungry and hotter than a 5970 with that kind of performance, its bad. That is ultimately the final analysis.

 

[SolMiester]

Performance linear scaling never happens unless if there's an extreme bottleneck situation. Those numbers seems fishy to me, because the HD 5870 had a much higher overclock and didn't yield the same scaling results. So far, the PCGH is the worst review of Fermi I've seen.

This reads to me...... because ATi doesnt scale linear, NV cant, so PCGH is the worst review...??

 

[dguy6789]

This reads to me...... because ATi doesnt scale linear, NV cant, so PCGH is the worst review...??

Do you seriously think Nvidia transcends basic laws of physics?

 

[SolMiester]

Do you seriously think Nvidia transcends basic laws of physics?

Hey, I'm not saying anything, but commenting on how evolucion8's quote came across to me...however PCGH said 18%oc = 18% performance increase....

 

[cbn]

2) If you watch the noise/power recording of the 480 @H, you can see the power consumption continuously rising (up to extra 50W) under same load as temperature rises. It's a known characteristics of 'leaky' chips.

This is why I am interested in seeing the watercooling results for 480 GTX.

Seriously, we might see some real gains under water. This architecture might have a lot more left in it.

 

[toyota]

Do you seriously think Nvidia transcends basic laws of physics?

considering it wasnt just the core/shaders that was overclocked it makes sense to me.

 

[Nemesis 1]

Agreed. Its sad that the heat issue happened and what ever. If this thing would have come out with 512. It would kick ass. Prettty much. But thats not what happened.

But it does create any interesting whats next for AMD/ATI . WE know or I think AMD/ATI knows not to bring northernlights out on a new node. I don't believe that,so leaves the question Whats NL going to appear as 40nm (Middle performance) or highend.

. But it better be good because when NV gets Fermi to the 28 node. I can see at least a 75&#37; gain in what we have seen so far . But Fermi I is same as PH I its a pass until II. So whats ATI refresh going to be . ATI/AMD not making alot of noise I smell a clubing coming for Femi I . But AMD better duck cause its clear hear Fermi II is going to stomp some ass. Plain as day to see. This thing has legs.

 

[Nemesis 1]

Just to add If northern lights is basicly ready and its good AMD could score a knock out. What if AMD has the middle performance NL ready to go and its faster than Fermi I

. All I can see is After 6 months Fermi should be a sitting duck for AMD . AMD had to been expecting more from Fermi. If AMD could get middle p[erformance NL out in short order and beat the 480 it could be a hugh win for ATI/AMD. It should happen this way to. Unless AMD fell asleep at the wheel.

 

[blanketyblank]

Just to add If northern lights is basicly ready and its good AMD could score a knock out. What if AMD has the middle performance NL ready to go and its faster than Fermi I

. All I can see is After 6 months Fermi should be a sitting duck for AMD . AMD had to been expecting more from Fermi. If AMD could get middle p[erformance NL out in short order and beat the 480 it could be a hugh win for ATI/AMD. It should happen this way to. Unless AMD fell asleep at the wheel.

If AMD did have something good they'd be stupid to release it now when their 58xx cards are still competitive with Fermi and they basically own the midrange. No one would want thei old 58xx cards if they knew a better chip was coming out so they'd be their own worst enemy. At best I see them working with the old chips and providing new PCBs and more memory to compete with Fermi.

I only see them unveiling 6xxx card after NV has released some midrange cards and probably a dual GPU solution. That would allow them to maximize their investment in the 5xxx cards and then crush NV's new line.

 

[Cookie Monster]

The first video card implementation of Fermi isnt what everyone was expecting. Architecturally its got alot of potential, but the physical implementation is a mess. But once they solve the power issue (which will naturally lead to noise/heat issues to be solved) and GDDR5 memory controller I think then we will see the true potential.

I also think there could be more TMUs in GF100 but are disabled at this point in time. Well thats the current rumour spinning in B3D when the rumoured GF104 specs surfaced.

This however means more waiting.. but atleast they got the damn thing outta the door. Kind of reminds me of some engineering projects ive done back at college. Start of the project = set high bars, ludicrous ideas etc end of project = get the thing to work asap to meet basic requirements, forget all other fancy features.

 

[Nemesis 1]

If AMD did have something good they'd be stupid to release it now when their 58xx cards are still competitive with Fermi and they basically own the midrange. No one would want thei old 58xx cards if they knew a better chip was coming out so they'd be their own worst enemy. At best I see them working with the old chips and providing new PCBs and more memory to compete with Fermi.

I only see them unveiling 6xxx card after NV has released some midrange cards and probably a dual GPU solution. That would allow them to maximize their investment in the 5xxx cards and then crush NV's new line.

I don't know their really quite but there is this, Its not much but AMD has had Time on their side this round. If ya hit a guy and he gets up knock him down right away .

http://news.softpedia.com/news/AMD-Working-on-New-GPU-Architecture-132855.shtml

 

[Cookie Monster]

I wonder if its a new architecture. Its been roughly 4 years now since R600 so its a possibility. Or it could be another incremental step which seems to work out rather well for AMD.

 

[Attic]

Seriously though, if the card ran cooler and used around 250W peak, it would be a fantastic GPU. A 10-15% lead over a 5870 for a slightly more power usage and cost is fine. But more power hungry and hotter than a 5970 with that kind of performance, its bad. That is ultimately the final analysis.

This paragraph sums up everything very nicely.

 

[Hacp]

Agreed. Its sad that the heat issue happened and what ever. If this thing would have come out with 512. It would kick ass. Prettty much. But thats not what happened.

But it does create any interesting whats next for AMD/ATI . WE know or I think AMD/ATI knows not to bring northernlights out on a new node. I don't believe that,so leaves the question Whats NL going to appear as 40nm (Middle performance) or highend.

. But it better be good because when NV gets Fermi to the 28 node. I can see at least a 75% gain in what we have seen so far . But Fermi I is same as PH I its a pass until II. So whats ATI refresh going to be . ATI/AMD not making alot of noise I smell a clubing coming for Femi I . But AMD better duck cause its clear hear Fermi II is going to stomp some ass. Plain as day to see. This thing has legs.

Why shouldn't ATI stick with its current strategy of high yield, cheap-midrange parts with the high end being covered by crossfire? Its a good business strategy.

 

[Compddd]

Also, I don't think anyone else mentioned this: 90C idle with dual displays.

I was planning on getting a GTX 480 until I read this. That is UTTERLY UNACCEPTABLE. I am forced to run 1 main display and 1 virtual display because my sound card is hooked up to my receiver via HDMI. This is a total deal killer for me on the GTX 480.

It sucks but Nvidia failed