The Fermi Thread - Part 3

Quote:

---

Originally Posted by GoldenTiger

It's 40% faster on average at 2560x1600 with some AA (2x-4x) in general still for a stock 5970 compared to a 5870, from what I've seen with newer drivers (I haven't seen a good article on 10.3a's yet). Regardless, 40% on a slower 5970's GPU's (core clocks/etc.) isn't 40% over a 5870, anyway: the 5870 has higher stock speeds and can clock higher when oc'ing as well. Crossfire tends to spike your max FPS making an awesome-looking average, while not really helping the minimums a whole lot for the most part. I don't need 20min, 75avg. 140 max, I need 50min, 80avg, 110 max! :)

---

When crossfire has a good profile for a particular game the scaling is usualy more then 40%. Minimum framerates go up as well, usually as much or more then the average framerate does. I wouldn't use dual GPU cards if they didn't improve the average AND the minimum.

Since you guys trust SKYMTL I used his 5970 review for the basis of the numbers: http://www.hardwarecanucks.com/forum...review-10.html
I included the % faster 5970 is versus 5870 in average and minimum framerates:

Code:

---

COD:WAW
1920x1200 4xAA/16xAF: +54% AVG, +41% MIN
2560x1600 4xAA/16xAF: +45% AVG, +38% MIN

COJ
No crossfire profile

Crysis: Warhead (DX10)
1920x1200 2xAA: +57% AVG, +62% MIN
2560x1600 2xAA: +48% AVG, +0% MIN

Dawn of War 2
No profile

Fallout 3
CPU Bound

Far Cry 2 (DX10)
1920x1200 4xAA: +45% AVG, +60% MIN
2560x1600 4xAA: +46% AVG, +55% MIN
2560x1600 8xAA: +44% AVG, +108% MIN

L4D
1920x1200 4xAA/16xAF: CPU bound
2560x1600 4xAA/16xAF: +49% AVG, +54% MIN
2560x1600 8xAA/16xAF: +50% AVG, +51% MIN

Hawx
1920x1200 4xAA: +64% AVG, +79% MIN
2560x1600 4xAA: +62% AVG, +0% MIN
2560x1600 8xAA: +43% AVG, +29% MIN

---

It was just a quick review of the numbers so I may have messed a few up. And that review was with old drivers so some games have been fixed or improved. The point is that you will get different results out of CF/SLI depending on the game and resolution you use. But on average if the game isn't CPU bound or the drivers lacking a profile the scaling of a 5970 is usually significantly better then 40%.

Also we should take into consideration that ATI isn't just going to sit around waiting. They are encouraging board makers to make higher clocked versions and there is the looming prospect of a revision. Even if a 5970 was only 40% faster then 5870 and if a Fermi was enough to match it, would it be enough to match an extra 15% or more in the newer cards? Ultimately we won't know until some reviews, which aren't that far off anyway.

But this is quite off topic. We aren't here to discuss 5970 scaling. We are here to discuss Fermi and only ATI products in so far as they relate to Fermi. If you want to talk multi-GPU scaling further I'm sure we could start a thread about it.

Quote:

---

Excellent? :rofl: :rofl:

---

I wouldn't call ATI's drivers excellent. But I haven't been having any worse trouble then I have with Nvidia drivers lately. After swapping my GTX285 and a friend's 5970 a number of times I can say I am not really impressed with either company's drivers and am quite disappointed in the downturn in quality on the NV side (I used to hold them in high regard).

Quote:

---

Originally Posted by GoldenTiger

Just because I post that I had bad experiences with ATI products doesn't make me a fanboy of nvidia's stuff :rolleyes:. I tried the 10.2 drivers prior to selling my 5870, as well, same issues mostly, as I said in my posts if you had actually read them instead of just going "OMG must be an nV fanboy!!!!" at first glance ;). Just because someone doesn't like one product doesn't make them a zealot of the competitor automatically, unless you are a fanboy of the one they don't like you wouldn't construe comments like I make as that.

---

I agree, personal attacks are unwarranted. We should all remember that not everyone has the same experiences, needs, and desires. My anecdotal evidence doesn't necessarily apply to you any more then yours applies to me.

Quote:

---

Originally Posted by annihilat0r

No, the high end is just that, the HIGH and the END. $700 is high end. $500 isn't.

---

Doesnt high-end represent the highest gpu in the range? Isnt that then the GTX480? Or are you saying that nvidia isnt even making a high-end gpu this time around? Are 470 and 480 both mid-range? But there is no range above 480 .....

:up:

Quote:

---

Originally Posted by Eastcoasthandle

The new profile update released on the 19th apparently addresses that. I wonder who will use it as part of their showdown reviews?

---

Unfortunately it's one or the other at this point. 10.3a makes AvP all but unplayable while many people (myself included) have reported major issues with the Profile + BF BC2 on Crossfire / Dual GPU cards. I am going to try and reinstall the whole shebang later today so I am crossing my fingers.

Quote:

---

Originally Posted by Solus Corvus

When crossfire has a good profile for a particular game the scaling is usualy more then 40%. Minimum framerates go up as well, usually as much or more then the average framerate does. I wouldn't use dual GPU cards if they didn't improve the average AND the minimum.

Since you guys trust SKYMTL I used his 5970 review for the basis of the numbers: http://www.hardwarecanucks.com/forum...review-10.html
I included the % faster 5970 is versus 5870 in average and minimum framerates:



I agree, personal attacks are unwarranted. We should all remember that not everyone has the same experiences, needs, and desires. My anecdotal evidence doesn't necessarily apply to you any more then yours applies to me.

---

Good post. I must have misremembered the reviews I read initially at release... so it is definitely better than 40%, looks more like 65-70% on the 5870 crossfire scaling assuming a profile is present. I remembered having read the minimums as not improving that much offhand, my bad :(.

On your last paragraph, thank you for being one of the couple to have respect on forums lately... I've been careful to (most of the time) restrict comments to include a phrase like "for me" or "in my system" so as not to blanket-statement things. I'm sure it works 100% awesomely for many people, it just doesn't for me. If people could keep themselves in check ego-wise like you are and actually respond with facts/analysis instead of personal attacks, forums would be a lot more useful and better in general (some have gone VERY much downhill such as hardforum lately, I appreciate that the people here even when being hostile are not going to those levels).

Quote:

---

Originally Posted by Solus Corvus

I hope you guys are kidding. Nvidia has been anything but quiet about fermi with one sneak peek, intentional leak, demonstration, and architecture review after another. They were so eager to start showing off fermi that they couldn't wait for a real card for their CEO to hold up in front of everyone.

What ATI did with 4000, 5000, and (so far) 6000 series was keeping a tight lid on things. What Nvidia is doing with fermi is the diametrical opposite of that.

---

No sorry I wasn't kidding at all, I was asking if it would be a fair comment to say, I guess you don't feel it is.
A small flash back

Quote:

---

Originally Posted by kgtiger

Would it be a fair comment to say.
The reason behind nvidia having such a strong NDA and being so tight lipped with there information on the 470s/480s, has alot to do with the fact that ati where able to keep the development
of eyefinity a secret from nvidia right up to the card being released?

It just seems to me like nvidia are making the ponit that they can keep things on the down low as well if they choose to.

---

I can see with hindsight I should of said "Final specifications" and not just information.

As for your statements.
"Nvidia has been anything but quiet about fermi with one sneak peek," Just as well then they had the good sence to keep it down to just one sneak peek then, hay.
" intentional leak," Well as there is no info or link about this I am not sure but I guess they could have.
"demonstration," Humm I do love a good demo, especially if it's about an up and comming new card.
"and architecture review after another." As to have lots of reviews on the net about nvidias new architecture,
I think that has a lot to do with the fact theres alots of review sites on the net that are interested and want to make money.
"They were so eager to start showing off fermi that they couldn't wait for a real card for their CEO to hold up in front of everyone." Well I guess they didn't have a real one to show everyone, in say that,
it's one thing to have a mock-up card and to let people know it's just a mock card.
If in fact the CEO was trying to give the impression that it was in fact a real working card, then thats just plain Dumb, I have no idea about this, only read here say.

Anyhow please correct me if I am wrong, but none of the above statements you have made came under the nvidias NDA.

The whole point of this was from what I know (and thats not much) in the past by now a lot more information about the final specification and benchmarks would have been known by now.

As ati was able to keep eyefinity a secret right up to it's release date this whole lockdown of information about the final specification and benchmarks just feels like a knee-jerk reaction by nvidia.

Quote:

---

Originally Posted by insurgent

$500USD isn't high end now? Is that mid-range? Wow.

---

nvidia still need to get a 495 out on the market wich would be the ultra high end LOL


............... imagine the price on that thing

Quote:

---

Originally Posted by kgtiger

No sorry I wasn't kidding at all, I was asking if it would be a fair comment to say, I guess you don't feel it is.
A small flash back

---

There was no question mark so I took that as a rhetorical statement rather then a question. Sorry if that's not the case.

Quote:

---

I can see with hindsight I should of said "Final specifications" and not just information.

---

Ok, I agree that we have very poor data on the final specifications and game benchmarks. There are too many rumors but nothing solid. But we do have lots of other information about fermi - much more then we did about the 4000 and 5000 series. We even already have significant information about the architecture.

Quote:

---

As for your statements.
"Nvidia has been anything but quiet about fermi with one sneak peek," Just as well then they had the good sence to keep it down to just one sneak peek then, hay.
" intentional leak," Well as there is no info or link about this I am not sure but I guess they could have.

---

Well maybe I didn't use the right choice of words. But I'm referring to the pics Nvidia put up on their facebook page. There is surely, IMO, also more information being leaked and/or intentionally distorted by both ATI and Nvidia.

Quote:

---

"demonstration," Humm I do love a good demo, especially if it's about an up and comming new card.

---

They have demoed fermi cards several times in the past monts. SC09, PDXLAN15, CES2010.

Quote:

---

"and architecture review after another." As to have lots of reviews on the net about nvidias new architecture,
I think that has a lot to do with the fact theres alots of review sites on the net that are interested and want to make money.

---

Lots of review sites would love to start talking about HD6000 series too but ATI hasn't given us much info yet, that's my point. Nvidia had two (IIRC) arch previews with one as far back as september 09. The review sites published the information because they want visitors to their site, but the information was provided by nvidia - it wasn't a leak.

Quote:

---

"They were so eager to start showing off fermi that they couldn't wait for a real card for their CEO to hold up in front of everyone." Well I guess they didn't have a real one to show everyone, in say that,
it's one thing to have a mock-up card and to let people know it's just a mock card.
If in fact the CEO was trying to give the impression that it was in fact a real working card, then thats just plain Dumb, I have no idea about this, only read here say.

---

I don't know what they were thinking only what they did. In either scenario you present they still wanted to show the card off rather then keep it under wraps.

Quote:

---

Anyhow please correct me if I am wrong, but none of the above statements you have made came under the nvidias NDA.

---

I'm not sure what you mean. Do you mean that architecture information, board layout and cooling, and box design isn't under NDA but performance is? Well then they aren't keeping a very tight lid on that information either. We have some numbers for Far Cry 2 and Unigine from Nvidia themselves and leaked numbers for Dirt 2.

Quote:

---

The whole point of this was from what I know (and thats not much) in the past by now a lot more information about the final specification and benchmarks would have been known by now.

---

I'm willing to bet that someone already knows and has stated or hinted at those specifics. But we won't know until after release who was right and who was merely spreading disinformation.

Being the card has two DVI and one mini-HDMI outputs, will they be able to support two monitors (via the DVI's) and a TV (via the HDMI) at the same time?

Quote:

---

Originally Posted by ShadowFox19

Being the card has two DVI and one mini-HDMI outputs, will they be able to support two monitors (via the DVI's) and a TV (via the HDMI) at the same time?

---

I don't think so, if it could then there wouldn't be a need for two cards to run triple monitor, you'd just use one monitor with the HDMI.

Quote:

---

Originally Posted by aussie-revhead

Doesnt high-end represent the highest gpu in the range? Isnt that then the GTX480? Or are you saying that nvidia isnt even making a high-end gpu this time around? Are 470 and 480 both mid-range? But there is no range above 480 .....

:up:

---

High-end means the most expensive and best product, which is 5970. A GTX 480 is an enthusiast product but it's not high end.

I believe some people will say "BUT GTX480 ARE NVIDIAS BEST GPU SO IT HAVE TO BEING HIGHEND LOL???". Lol indeed, Fermi hasn't been released yet so is GTX 295, a product on par with a $400 ATI card, a high end card because it's the best Nvidia has to offer right now?

Quote:

---

Originally Posted by annihilat0r

High-end means the most expensive and best product, which is 5970. A GTX 480 is an enthusiast product but it's not high end.

I believe some people will say "BUT GTX480 ARE NVIDIAS BEST GPU SO IT HAVE TO BEING HIGHEND LOL???". Lol indeed, Fermi hasn't been released yet so is GTX 295, a product on par with a $400 ATI card, a high end card because it's the best Nvidia has to offer right now?

---

High-end means high-end, not TOP-end/bleeding edge. GTX480 is a high-end product. So is a GTX 280, Radeon 5870, and GTX 295, frankly.

Quote:

---

Originally Posted by ShadowFox19

Being the card has two DVI and one mini-HDMI outputs, will they be able to support two monitors (via the DVI's) and a TV (via the HDMI) at the same time?

---

I don't see why not? It should work like this

Quote:

---

Originally Posted by MrDiaz

I don't see why not? It should work like this

---

It won't work.

Quote:

---

Originally Posted by Chumbucket843

how do you know how 400 series costs much more? you should do some research first. even if the die costs 2x more the other costs will lower difference in BOM.

---

Common sense maybe?

Larger die = more expensive
Larger memory interface = higher complexity PCB = more expensive
Higher TDP = higher complexity PCB, better/more power components, larger heatsink = more expensive
More memory = more expensive

I said it before... GF100 will cost Nvidia more than ~2x to make than Cypress and that is being conservative.

Quote:

---

Originally Posted by GoldenTiger

Good post. I must have misremembered the reviews I read initially at release... so it is definitely better than 40%, looks more like 65-70% on the 5870 crossfire scaling assuming a profile is present. I remembered having read the minimums as not improving that much offhand, my bad :(.

On your last paragraph, thank you for being one of the couple to have respect on forums lately... I've been careful to (most of the time) restrict comments to include a phrase like "for me" or "in my system" so as not to blanket-statement things. I'm sure it works 100% awesomely for many people, it just doesn't for me. If people could keep themselves in check ego-wise like you are and actually respond with facts/analysis instead of personal attacks, forums would be a lot more useful and better in general (some have gone VERY much downhill such as hardforum lately, I appreciate that the people here even when being hostile are not going to those levels).

---

Speaking bluntly, if you keep speaking of your own experiences as purely anecdotal, then why do you debate about people on driver issues? It's like you're trying to assert what you've experienced is general and are trying to influence other people by that 'fact.'

Quote:

---

Originally Posted by MrDiaz

I don't see why not? It should work like this

---

Fermi probably has only 2 display pipelines.

Quote:

---

Originally Posted by LordEC911

Common sense maybe?

Larger die = more expensive
Larger memory interface = higher complexity PCB = more expensive
Higher TDP = higher complexity PCB, better/more power components, larger heatsink = more expensive
More memory = more expensive

I said it before... GF100 will cost Nvidia more than ~2x to make than Cypress and that is being conservative.

---

Only Problem is nvidia does not produce graphics boards. :)
Even GPUs are not produced by nvidia. nvidia is fabless.
Let all this be the problem of the board makers.

Quote:

---

Originally Posted by mapel110

Only Problem is nvidia does not produce graphics boards. :)
Even GPUs are not produced by nvidia. nvidia is fabless.
Let all this be the problem of the board makers.

---

Their partners must love them... they'd probably rather cut prices on the Radeon 5ks and sell more and still earn money, since those boards are so easy to produce.

Quote:

---

Originally Posted by mapel110

Only Problem is nvidia does not produce graphics boards. :)
Even GPUs are not produced by nvidia. nvidia is fabless.
Let all this be the problem of the board makers.

---

BOM is set by Nvidia...
Nvidia has contracts with TSMC, they pay TSMC money.
Nvidia sets the reference design for the cards, they purchase the components from other manufacturers and sell them in kits to AIBs.
All AIBs have to worry about is the final assembly, packaging, warranty/service and distribution of the cards.

Quote:

---

Originally Posted by LordEC911

I said it before... GF100 will cost Nvidia more than ~2x to make than Cypress and that is being conservative.

---

They are definitely more expensive, but don't forget, that they are salvage parts, so the yields are better. More memory, but slower afaik, that can also make a difference. And lastly, they don't have the same contract as ATI with TSMC. They are producing more chips, so they can have much better deal.

Quote:

---

Originally Posted by LordEC911

Common sense maybe?

Larger die = more expensive
Larger memory interface = higher complexity PCB = more expensive
Higher TDP = higher complexity PCB, better/more power components, larger heatsink = more expensive
More memory = more expensive

I said it before... GF100 will cost Nvidia more than ~2x to make than Cypress and that is being conservative.

---

Die in between 280 & 285 dies = similar cost to last gen assuming die size alone is the sole factor
PCB less complex than their previous 285 with smaller memory bus
Lower binned ddr5 offsets costs of larger bus

I say we can all armchair build a full product and assume manufacturing costs its not so hard....

Quote:

---

Originally Posted by Vardant

They are definitely more expensive, but don't forget, that they are salvage parts, so the yields are better. More memory, but slower afaik, that can also make a difference. And lastly, they don't have the same contract as ATI with TSMC. They are producing more chips, so they can have much better deal.

---

Depends on the salvage part...
Correct with the TSMC contract, we don't know anything about either contracts, though I suspect AMD/ATi has been getting a decent discount as well over the last few years.
I also highly doubt Nvidia is seeing 60% yields right now...

Quote:

---

Originally Posted by highoctane

Die in between 280 & 285 dies = similar cost to last gen assuming die size alone is the sole factor
PCB less complex than their previous 285 with smaller memory bus
Lower binned ddr5 offsets costs of larger bus

I say we can all armchair build a full product and assume manufacturing costs its not so hard....

---

40nm wafers are more expensive than 65/55nm wafers...
Lower binned GDDR5? It is clocked slower but I don't think it is a lower bin.

so many semiconductor pro's here .....

Quote:

---

Originally Posted by cegras

Speaking bluntly, if you keep speaking of your own experiences as purely anecdotal, then why do you debate about people on driver issues? It's like you're trying to assert what you've experienced is general and are trying to influence other people by that 'fact.'

---

Actually, other way around: they keep saying "Well it must be user error ATI IS INFALLIBLE YOU NUTCASE" so I am simply defending myself.

lol agreed, you wonder if the peeps in this forum where in charge we would have had any problems at all with the yield on Fermi silicon :rofl:

Quote:

---

Originally Posted by LordEC911

40nm wafers are more expensive than 65/55nm wafers...

---

this comment is purist logic in its finest form, and truly smacks of GPU genius....im only joking Lord EC :up:

anyways, a blast from the past, remember this pic....seems like years ago now lol,

http://www.hardwaresphere.com/wp-con...engine-gpu.jpg

heres the link to the so called 'digital citizens' that where supposedly rendered from scratch, but where in fact actors in still shots animated to look like live motion, on Fermi of Course,

http://www.onesize.nl/projects/playgrounds-titles-2009