Printable View
Well they took these benches...Quote:
Originally Posted by Sam_oslo
http://images.trustedreviews.com/ima...12254-far4.gif
http://images.trustedreviews.com/ima...2254-cod44.gif
Then made their own graph with the data, and added a column for gtx480
To get the 480 scores, they took the gtx295 data and multiplied it times 1.5.:down:
So it's a real graph of some fake scores. :ROTF:
Then everything's a failure... AMD's magny cours... intel's gulftown... amd 5990 :p: windows 8... xbox 370 ... ps4 :p:Quote:
Originally Posted by h0bbes
Yeah, but nVidia provided them with that x1.5 info, right?Quote:
Originally Posted by jaredpace
In case, it is a really interesting and valuable graph in this stage. Because even nVidia should have been optimistic with 1.5 x GTX 295-performance, these numbers are showing a really superior GTX 480. Or what?
Errr, I think most would agree that Fermi was later than what nvidia was hoping for.Quote:
Originally Posted by BenchZowner
Its a fake graph, why are you acting like it matters?Quote:
Originally Posted by Sam_oslo
I don't know if they are real, but the guy is calming that nVidia has provided the info, and in case they matter a lot . Who told you they are fake? source please.Quote:
Originally Posted by MrMojoZ
I think you kinda misunderstood how rumors work.Quote:
Originally Posted by Sam_oslo
If you spread a rumor, the burden is on you to prove that it's right. It's not for others to "disprove" a baseless claim.
Speak for yourself.Quote:
Originally Posted by Tenknics
Quote:
Can't everyone just agree Fermi is late and is for a reason ? Its a failure, no matter how you try to justify it.
THEY AREN'T REAL RESULTS FROM ACTUAL TESTS. Hence fake. You can prove it yourself with a little math. Wait for some real info. I'm sure Fermi will be very fast, it is the price/power/heat/stock levels issues that determine if it is a good product or not.Quote:
Originally Posted by Sam_oslo
Did I say this a proven fact?Quote:
Originally Posted by Teemax
I'm trying to find out where these numbers comes from, and what they mean. I'm holding an open mind with all possibilities on the table.
But maybe you should say this to those who are trying to say it is proven fake?
@samoslo: I've see everything from 15% slower than a 295 to 50% faster than a 295. But I will just go by the GTX470 3dmark vantage score of P17,000. And another good judgement of performance, IMO, is the GF100 farcry2 and darkvoid benchmarks. Makes it ~65% faster than the 285, and up to 100% faster if using high physx lots of compute + rendering simultaneously. Sounds like it's going to be similar to GTX260 vs. HD4870, where the GTX470 and HD5870 will compete and trade blows. And the GTX480 will be above all in a league of its own, ~25% faster, fastest single gpu, and between HD5870 + HD5970.
If you were to use an application like (extreme compute, really high Phys-x games like batman, cryostasis, darkvoid, etc, raytracing, the rocketsled demo, folding, crunching, or unigine heaven w/tessellation, or tessellation demos), then the GTX480 would demonstrate the 100% lead over the GTX285 because of its raw compute power. With traditional games like modern warfare, farcry, etc, it should be like the FC2 demo, with a ~65% lead.
If it was really %50 faster than a GTX295, then it would blow away a 5970. If that remark is true, then they are probably quoting the 295 at high resolutions with filtering where the 295 starts to slow down a bit. This is also where the gf100 would pull ahead because of it's quad triangle setup. If you were to run 30" and pile on the 8xAA or 32xAA, the 295 would choke, and the GF100 could very well be twice as fast. This wouldn't say the same for gameplay at the more popular resolutions.
You see in this stage we don't have real benchmarks, and this guy is claiming that the info is provided by nVidia based on some calculations/predictions that 480 GTX will perform 1.5 times better than GTX 295. In case this is the most interesting data in this stage.Quote:
Originally Posted by MrMojoZ
This may be fake too, I don't know, but I don't don't get why so many are afraid of discussing the possibilities and meaning av this. is it that good?.
Aren't you getting it? That chart is calculated by multiplying the GTX 295 scores with an amount Nvidia told some time. They are not actual benchmarks. They don't tell anything more than that Nvidia once upon a time allegedly told someone something about GTX 480 performance.Quote:
Originally Posted by Sam_oslo
ok some serious info 470 is faster like 25 percent from gtx 285 and gtx 480 is close to 5970 in both performance and power usage
OK.Quote:
Originally Posted by Sam_oslo
The guy heard a "claim" from NVIDIA without any logical explanation whatsoever for the miraculous performance, and he went ahead and fabricated charts to ... back up his claims.
This is screaming fake for me.
Yet, you want to "discuss" it as facts. Interesting.
GTX 480 will be 40-50% faster than 5870. And would you know vhy...
Because Nvidia drivers for DX11 would have more performance in DX11&DX10 games...
You all know that DX11 for 5770(an example) lowered the performance..
I mean 5770 shoud have about same performance( in DX 10 games) with 4890(same specifications) , but it don't have it.
5830>=4890.
So GTX 480 will beat 5870 with the "power of drivers to"..:)
I am thinking this thing is going to be one leaky boat of hot transistors. There are obviously problems as it would not have taken this long to put the product out. I am awaiting patiently to see this thing finally come out of the proverbial closet to face the challenges of what customers will put it through. I am also anxious to see what this can do when watercooled as compared to air cooled because I think there is going to be a thermal limit on things here for sure. It's a big chip and it is going to be hot. There are no two ways around that unless you lower clocks significantly. Hence why I am waiting patiently as I want to see a F/C waterblock on this and see what happens when the thermal barrier is knocked down and a couple voltage tweaks added. As much as I do not like Nvidia's business practice of BS I have still sat by and waited and did not pick up the ATI counter part. If Fermi fails then it fails and I will pick myself up a nice Cypress package.
Looking forward to seeing that as well. Won't be me, but I'm sure someone on this forum will do it. :) I bet it will be a screaming beast at that point.Quote:
see what happens when the thermal barrier is knocked down and a couple voltage tweaks added.
Holy cow, have you guys seen this? http://www.youtube.com/watch?v=VpZXhR1ibj8
It is the Hitler video and his reaction is to the bad news about Fermi. HAHA, man it is classic.
Hahaha I thought that video was used once before on another product...Man that kills me...
:rofl: :ROTF: :clap:Quote:
Originally Posted by jaredpace
Thanks jaredpace :D
Indeed thanks very much Jaredpace, I just laughed my butt off (and given how dissapointing things have been regarding this new Nvidia card I needed a good laugh)!
That was damn funny as hell! :rofl:Quote:
Originally Posted by jaredpace
I swear that video with different captions never gets old :)Quote:
Originally Posted by Sadasius
Agreed! :D I especially liked the part about JHH and wooden screws! :clap: :rofl:Quote:
Originally Posted by 96redformula7
cuda will solve everything lol damn so sad how nvidia became such a joke material after that mighty g80 arch
What was the "wheelchair wheel" reference?Quote:
Originally Posted by jaredpace
These videos really never get old, although I do have to turn the sound off of course.
I dunno, I didn't get that one either. And in the thread i saw this link in at hardforum, they were all asking, "wheelchair wheels?" :shrug: :ROTF:Quote:
Originally Posted by thatdude90210
GTX 470 20-25% faster than GTX 285, GTX 480 equal to HD 5970?
http://www.donanimhaber.com/GeForce_...ilir-18363.htm
Translated
close not equal
I dont think so. There are 99 reasons why not. And only reason why yes.Quote:
Originally Posted by annihilat0r
yup. The parallels are eerie. Except, back in R600 days, AMD was going monolithic (1 huge chip), for both CPU and graphics.Quote:
Originally Posted by saaya
Now, AMD graphics is always small chip strategy. And even CPU like Athlon X4 are aiming to be smaller.
Simply, AMD changed. nVidia still hasn't.
That is because BOTH nVidia and AMD "fix bugs" in games just before new product launch to "slow down" their old products, to make new products look better. You can even use it to make rebranded products appear faster :up:Quote:
Originally Posted by Dainas
SECRET OF GTX480/FERMI PERFORMANCE.
Look at these graphcs comparing 4890 vs 5870. Similar clocks, but 2x more shader power:
http://www.xbitlabs.com/articles/vid...hd5870_16.html
But performance typically only 40-50% higher.
And 5970 has about 4x shader power of 4890:
http://www.xbitlabs.com/articles/vid...0_8.html#sect1
But once again that translates into only 70-90% higher fps.
Fermi has at best 512/240 = 2.1x more shader power.. probably a bit lower if less SP or lower clocks. Although they will definetly be faster than GTX285, and certainly close to 5870, I think comparisons to 5970 are wishful thinking (unless secretly those 512SP are really 1024SP -_-)
Anybody hear of Amdahl;s Law? Diminishing returns?
Some games, especially RTS, are so CPU bound they show very little gain. Some like L4D2 are not very demanding, and doesn't matter if you're getting 150fps or 200fps.
The only games where 5870/5970 really make a difference are those where such a card makes the game playable: Crysis/Warhead, Stalker: COP, HAWX, Battleforge.
So like I said so many times before, its "Crysis" syndrome. 90% of games that everyone can play on any mainstream card, and a few not so great games with "special" effects feature that makes them so slow, even ultra-high-end struggle. Bottom line, unless Fermi has some top secret features, its just a (late) replacement for 5870... ie doesnt change anything.
Quote:
Originally Posted by xdan
You've sure posted a lot for someone who was done commenting, lolQuote:
Originally Posted by xdan
Invalid comparison. The reason ATI is getting diminishing returns from doubling the SPs is because their 5-way SPs are optimized for independent instructions, which can't be continuously generated for an unlimited number of SPs. RV870 can only execute 320 dependent instructions, which is where the biggest chunk of performance on the SPs comes from because performance increases in a more linear manner with an increasing ability to execute dependent instructions.Quote:
Originally Posted by ***Deimos***
Each of GF100s SPs can execute a dependent instruction, for a total of 512.
Even if it's not it won't be the end of the world. A 480 shader / 60 tmu part should still hold its own well enough. It all comes down to clocks.Quote:
Originally Posted by annihilat0r
Haha, that argument is such a joke. You take this crap way too seriously. Fermi is late, I don't care what you call it, a uARCH, a product whatever. The uARCH is what the product is based off of right? We were told it'd be out in 09, its not. That's my point. You are way too literal and it seems the fanboy in you will pick apart anything anyone says just to find something to argue with. I guess r600 wasn't late either..It wasn't a product, just a uARCH. And Tukwilla? Nah, not 3 years late. It was never a product, just a uARCH! 'Quote:
Originally Posted by Chumbucket843
I would sure like to make new uARCHs for a living. That way I can just keep working on it indefinitely and not have worry about it ever being late. uARCHs are immune from time, they can't be late or early! They aren't products after all~ :rofl:
Whether info is official or not, something is obviously wrong. Deny it all you want, have fun in whatever world you're living in. I live in reality.
Point to me where I am blatently insulting other people, putting them down and trying to start stuff? ??!?!Quote:
Originally Posted by ElSel10
why do you have such an aggressive response to my argument? quit being so emotional. i'm not picking at your words, you just post ridiculous things.Quote:
Originally Posted by Tenknics
exactly. its not a hard concept to grasp. the delays were process related too. they put too much on their plate but it payed off in the end. tukwila is another story altogether.Quote:
I guess r600 wasn't late either..It wasn't a product, just a uARCH. And Tukwilla? Nah, not 3 years late. It was never a product, just a uARCH! '
you are very funny but i dont know how you could have reached this conclusion based on my statement. fermi taped out 6 months ago so the architecture is finished. they had 3 metal spins which were for debugging.Quote:
I would sure like to make new uARCHs for a living. That way I can just keep working on it indefinitely and not have worry about it ever being late. uARCHs are immune from time, they can't be late or early! They aren't products after all~ :rofl:
that something would be yields. i thought that was well known by now.Quote:
Whether info is official or not, something is obviously wrong. Deny it all you want, have fun in whatever world you're living in. I live in reality.
look up.^^;)Quote:
Point to me where I am blatently insulting other people, putting them down and trying to start stuff? ??!?!
If that's true, will we see a 512 SP high clocked GTX 480 vs 384 SP decent clocked/416 SP low clocked GTX 470 ? The performance differential is too wide. :shrug:Quote:
Originally Posted by onethreehill
I hope this is true and it is what I have been suspecting all along.Quote:
Originally Posted by Zed_X
that would make the 470 faster than a 5850 and about as fast if not faster than a 5870, right?Quote:
Originally Posted by eric66
yeah, i liked the part when hitler asks everybody except for the top guys to leave the room, and they dubbed it : everybody who owns or has owned an ati card, leave the room... and almost everybody leaves :lol:Quote:
Originally Posted by 96redformula7
i wonder if youll still laugh when we hear about the final price and when you try to actually buy one though heheh :DQuote:
Originally Posted by Zed_X
about fermi being late... i dont think that necessarily means fail...
how long did G80 last? it could have launched 6 months later and still would have been a kick4ss gpu and dominated the market...
no news about GF104? :(
I hope you aren't basing that off PR slides! :ROTF:Quote:
Originally Posted by Zed_X
he's been right before...Quote:
Originally Posted by zerazax
I am 99% sure GTX480 only has 15 TPC.
He has also been wrong...Quote:
Originally Posted by 570091D
well, it doesnt really matter what nvidia calls them, but there probably will be 16TPC 512core cards...
im pretty sure they will come in very limited numbers though...
probably less than 280s when nvidia launched gt200, and back then the whole launch allocation for one month for taiwan was only 3 dozen cards...
sure, taiwan isnt a big country, but they were sold out within a day or two :D
His post history doesn't indicate that. At least here at XS. What has he been wrong about so far? In any case there should be leaks galore within a week or two.Quote:
Originally Posted by LordEC911
What about B1 revision pals ? Is it the retail version ?
And GF104 has been taped out or no ?
And thanks OTH for the donanimhaber.com's link, hope that's true :D
edit: I have some very reliable source, saying that GTX 480 will be 480 sp's and above 700MHz in clockspeed
GTX 470 will have 1280MB memory, and GTX 480 1536MB memory.
This due the memory buses(like G80 series) 320bytes(640-1280) and 384bytes(768-1536)...:)
Comments about launch...Even so i don't "speak" very much like others..Quote:
You've sure posted a lot for someone who was done commenting, lol
I think it's better if you don't post at all. :DQuote:
Originally Posted by annihilat0r
--->
Quote:
Originally Posted by annihilat0r
This looks really good, and several sources are indicating the same in past days. Lets hope it holds true, and the possibilities is raising ;)Quote:
Originally Posted by onethreehill
I can almost hear all those non-scene denials that would rather to deny this by some wild insider-info claims, again. :p:
It's not his fault. He just has multiple 100% accurate sources that are saying completely different things. It seems to be affecting a lot of people these days ;)Quote:
Originally Posted by Tim
Quote:
Originally Posted by Sam_oslo
If NV can make GTX480 equal to HD5970,why they make GTX470 slower than 5870?
Last time GTX280 is 15%-20% faster than GTX260.
:DQuote:
Originally Posted by Tim
What can I do, Nvidia is so indecisive. Looks like the final decision of cutting the card to 480SP was made within the last couple of days.
Quote:
Originally Posted by mindfury
I think it is fare to say 480 will be about ~20% better than 470. Then if 480 GTX matches/beats HD5970, we can expect 470 to beat the 5870 by ~20%.
But donanimhaber said GTX470 only 20-25% faster than GTX285.:shrug:Quote:
Originally Posted by Sam_oslo
Above 700 is a lot higher than what most rumours have been hinting at. I really hope there is a 512 part tho.Quote:
Originally Posted by annihilat0r
Nothing is for sure yet, but several other sources are indicating that 480 GTX will be at least as good as HD5970 (if not not better).Quote:
Originally Posted by mindfury
Fixed ;)Quote:
Originally Posted by Sam_oslo
These conflicting sources is all we have for now.
We must try to deal with it until we get some more reliable sources and benchmarks. But until then, I personally would rather investigate them with an open mind.
That's all we have, and we should deal with it in respectful manner. I would also suggest that we all show respect for others meaning and idea without fixing their Quote.
As far as I could gather from what some people (who gave reliable info in the past, not sure of the present :D) have been saying, the decision was 512SP with 650Mhz clockspeed but then it was ditched just within the last couple of days in favor of 480SP, 700+ clockspeed and higher availability of the cards.Quote:
Originally Posted by takamishanoku
Obviously every such info could be wrong but I don't think they could be far off.
In terms of computing power, 512 x 650 is as much as 480 x 700 but we'll see..
Thanks man :) The "open mind" part had me gigling ;)Quote:
Originally Posted by Sam_oslo
An "open mind" doesn't jump on anything that makes Fermi look good and screaming "LOOK AT THIS!", even though the underlying numbers or arguments are completely false.
May I suggest that you use a better choice of words next time? :)
You may not suggest what i write. I chose my own words based on my ideas.Quote:
Originally Posted by Teemax
The whole thread is full av denials about a good Fermi. Too much Fermi-bashing and denial of anything that can cast a better light on Fermi. I'm trying to bring balance to this non-sense. Not because I'm sure Fermi is good, but because I'm sure many of these Fermi-bashing is as unsure, too.
As I have repeated several times before, "nothing is for sure yet" and those of us interested in high-end GPU and Fermi should keep an open mind, both for Fermi and for others ideas and meanings.
oh my god are you serious :ROTF: so according to these 100 percent accurate sources nvidia changes specs every day :rofl:Quote:
Originally Posted by annihilat0r
So let's see, nV figures out 512 SP is no go so they say fsckit lets cut it down to 480 something and raise clock from 600 to 750! They all start cheering, hugging and taping themselves on the back 'How the fsck didn't we thought of this sooner!!! It's even faster now!
Heck why not cut it down to 300 SP and raise clock to 2Ghz!!!! That would WTFBBQPWN!!!!
Get real guys.
what a mess!!Quote:
Originally Posted by annihilat0r
They can do one of the two things:
a) turn off cores via BIOS (leaving possibility to unlock them)
b) fuse them off in testing facility!
If the option is b) that would mean that chips are still in factory, not in assembly line, and far, far away from AIB partners and journalists (not to mention customers).
Also if NV is chasing quantities, they would fuse of cores, so that they don't drain power (BIOS disablement can't cut of power to the disabled units). And that also means that they're upping voltage to make as much as possible chips come alive with as much as possible working cores. Upping the frequency would not be in line with the goal of making as much as possible usable chips, and NV DOESN'T have time for precise binning!
So, point is:
a) decision of the core numbers were put in place a month or two months ago, not yesterday
b) decision was made yesterday, and that pushed launch from planned CeBIT time-frame to PAX, and availability in April - May
Anyhow this is sign of total lack of any planned execution...
Of course it's full of Fermi bashing. NVIDIA is having real problem delivering the cards, and most of the things we hear are bad news, except for a few random (and baseless) "claims" of a miraculous Fermi's performance, which you ask people to accept as gospels "with an open mind"...Quote:
Originally Posted by Sam_oslo
If there are real, substantial good news, bring them out for us to rejoice! Do NOT ask people to accept obvious fakes "with an open mind" just like that. It doesn't work, and it's pathetic.
1- Who said it was 100 percent accurate. You're making fun of words you yourself are putting in my mouth.Quote:
Originally Posted by eric66
2- Who said Nvidia changed specs every day, I told of only 1 (reads "one") change of spec. Don't forget that the specs from long ago (750mhz and such) were only targets which were missed, and clock speed testing of the last revision (A3) was completed only recently. So anything you have heard before were only rumors / lies / targets.
As far as I know, Nvidia had completed testing the latest revision only a little time ago and decided upon 512SP / 650MHz. Before that there weren't any specs only targets and rumors. They have changed the spec ONCE and that is to 480SP / 700+MHz.
Surely those could be all wrong. But one change of spec isn't really an impossible thing to happen. Especially if you have missed your targets horribly and are trying to figure out what to do best with the parts you ended up with; and with the ultra high TDP and thermal death problems they have now.
http://www.digitimes.com/news/a20100225PD202.htmlQuote:
Second-tier makers unlikely to start mass shipping Nvidia Fermi cards until April
Second-tier graphics card makers are unlikely to start mass shipments of their Nvidia GeForce GTX 480/470-based (Fermi) graphics cards until April despite that Nvidia plans to announce the GPU on March 26 since most of the them have not yet received complete reference board designs, according to sources from graphics card makers.
The sources expect Nvidia will give supply priority to first-tier makers or makers that only produce Nvidia cards. Makers including XFX and PNY are already offering pre-orders for these upcoming cards with GTX 480 priced at around US$679.99 and GTX 470 at about US$499.99.
Some market watchers originally expected AMD to drop its graphics card price to counter the launch of Nvidia's new cards, but since the prices of Nvidia cards are higher than AMD's existing cards, watchers now believe any price war between the two chip makers is unlikely to start until after May.......
So the SabrePC "prices" indeed fooled everybody? Or were there other places which sold XFX and PNY 480's?
Cos you know, it looks kinda interesting to me that XFX and PNY are offering those cards only on a site that is ranked 6 million in Alexa.
I got from that..that the 5970 was going to be the faster card. Meaning the GTX480 could not beat the 5970. And yeah, the 470 would beat the 285 by about 20%~25%. That's really about 5870 levels. Depends on which game you're looking at.Quote:
Originally Posted by mindfury
you could be forgiven for thinking ati had been tipped off, that fermi was close to the 5970's performance.(unlikely at this point) :rolleyes:
so ati have released an enhanced spec for its partners to produce a card 20% faster with 4gb gddr5... :up:
or they might just be playing it safe and have decided they are going to keep the performance crown no matter what.:clap:
cant wait for the real bench's because we have so much misdirection...:(
what we know for sure is nvidia is really sucking ati's balls really hard at this point...:bananal::cord:
Quote:
Originally Posted by annihilat0r
Yeah, they were going to with a 650MHz, 512SP ..then at the last minute they decided to change it to a 950MHz, 1024SP. :ROTF:
The GTX 4xx has allready started effecting 5870's price nice i love it...
HIS 5870 FOR $380.....
From computerbase latest test:Quote:
Originally Posted by Your_Boss
http://img.photobucket.com/albums/v4...r/a9707a7c.jpg
http://www.computerbase.de/artikel/h...formancerating
24-hour special , nothing to do with GTX 4xxQuote:
Originally Posted by ajaidev
You sound so sure of nVIDIA Fermi's crappiness.Quote:
Originally Posted by purecain
Do you have any private info that we don't ?
I hope Fermi fails because it's going to be very hard for you to eat your words :D
So now "the pore yelds, broken, unfixable" can do 750MHZ core? :D
exactly!Quote:
Originally Posted by xdan
it would be achievement if 480SP enabled core can achieve 600MHz for descent supply...
:p:Quote:
Originally Posted by annihilat0r
Oh man, that would suck so hard, seriously, if that's true, I'm thinking about just forgetting about getting Fermi after all. Maybe I will pick up a cheap 5870, or even cheaper 285 to tide me over. Well....I don't need them actually, maybe Crysis 2. Gives me some time to see how this will pan out.
yeah but the initial target was 750mhz with 512sp.Quote:
Originally Posted by xdan
they ended up with 480sp and even that can't do 750mhz.
I'd like ALL this info to be wrong and Fermi show up with 750mhz 512sp... with 200w TDP and for only $500 but then I'd like people to elect me the President as well.
But I do think GTX 480 won't be more than $500, because its performance level won't justify anything more.
GF100 will NOT be featured on Quadro's: http://www.xbitlabs.com/news/video/d...ar_Nvidia.html
They should release some Ultra-GTX480
750 mhz - 512 sps
It would sell, and they could charge a :banana::banana::banana::banana:load for it
And they would basically match the 5970
I think this would be Quadro ones :yepp: Desktop parts would be cutted, slower or both as there is way more defective cores (but working with some cutting-down).Quote:
Originally Posted by orangekiwii
Not surprising at all. Besides, it is possible that Nvidia wants to 'please' the most loyal partners giving them a headstart.Quote:
Originally Posted by onethreehill
Awesome! Fermi better be cheap as well! :eek:Quote:
Originally Posted by ajaidev
It's a pity 5970's prices aren't really dropping yet. :(
Are 5970's even available for sale anywhere? It needs to be in stock before we get price drops.
I remember the 8800 Ultras @ ~$900. Lets not hope for that with the 480. I'm holding out for $300 5870 or $500 5970 or $400 480, whichever comes first.
$680 for the GTX480? Who would pay for that if it's in 5970 territory? That's just silly.
Quote:
Originally Posted by annihilat0r
higher base clock is better for TMU/ROP no?Code:750 700 650 600
512 3.84 3.58 3.33 3.07
480 3.60 3.36 3.12 2.88
448 3.36 3.14 2.91 2.69
but since shader power is proportional, every 50Mhz is tradeoff like a 32SP block.
I'm surprised by recent "news" too. I also would consider a 480SP part in >10k to be a huge success, even at 600Mhz.Quote:
Originally Posted by Nedjo
Remember the "amazing" 8800GT launch on mature 65nm process. It was 112 out of 128SP, 600 out of 738Mhz (GTX+), and exact same architecture as year old GTX.
Back before 5970 I was adamant it would require special low voltage binning, low clocks, or lower SP count, or all above.
I just find it very very difficult to believe Fermi's 3B transistors can be SO MUCH more efficient that you'd get very high SP AND 750Mhz. Like having cake and eating it too. Too good to be true. Yield (# defective chips) issues may improve over time, but physics doesnt.
480SP x 600Mhz is more realistic because:
- less power/heat - not as much trouble getting PCIE and power supply certified
- 480SP drastically improves availability. Extra cluster makes marginal performance difference.
- room for future "ULTRA" version. Look at GF6/7/8/200.. they always make room for ULTRA.
- TSMC transistor variance issues. If you're going for a hike up a mountain with olympic athletes, all within sec of each other, you can safetly go at high pace. Throw in some stragglers and you're forced to lower pace.
- teething in new architecture. Face it, 8800GTX didn't break any clock records. It was ONLY 575Mhz. Why expect a miracle out of the 2nd biggest chip every made in history of mankind? Getting a fully working chip is astonishing enough engineering feat like Burj Dubai
- 480 is "only" 2x SP on GTX285. And "only" 40-50% more bandwidth. Thats actually HUGE. Fermi is so hyped beyond belief, as if CPU and other system bottlenecks don't apply to it.
Quote:
Originally Posted by Sam_oslo
Quote:
Originally Posted by Sam_oslo
Quote:
Originally Posted by Sam_oslo
Quote:
Originally Posted by Sam_oslo
:rofl::ROTF::clap:Quote:
Originally Posted by Sam_oslo
back on topic, guys, do you think nvidia just sent out cards to its partners without telling them how many sps they have and whether they are 470s or 480s? they could also have sent them at different clocks, to then check their feedback and see what partners think about what configuration and what clocks and prices etc... kinda like testing their partners feedback what prices and speeds would be acceptable.
im not sure if gpuz can read fermis configuration, even reading clockspeeds is tricky on some new cards before they are officially released... so this might explain the confusion...
Seems to me Fermi is going to fail, if I was nvidia I would dump fermi (like intel did with Larrabe) and go for the next-gen gpu, at this point fermi will born dead, since ATi is cooking their next-gen too and if Fermi performance is on par with ATi current offerings is going to be even worse for the green guys...I don't know what is passing through nvidia mind, they had problems with GT200 at start because the chip was gigantic, they had problems jumping to 55nm too and now the do the same mistake. AFAIK you learn from the mistakes you did in the past to no do it again, but seems nvidia doesn't...
:rofl::ROTF::clap:Quote:
Originally Posted by saaya
back on topic
Target is 750mhz with 512sp.
this isn't Photochoped, it's actual banner from partners website!
this prophet just don't give up :D
http://img690.imageshack.us/img690/1...uterlaptop.jpg
Hi my friend, good to see you again. One of the great inventions of this forum is the ignore list. Just type in his name, like the rest of us, and it's like he's not here at all. Just like some kind of magic... :-)Quote:
Originally Posted by saaya
Thank you for addressing and illuminating the various issues raised and for not loosing your patience with us.
Rob
i think thats too pessimistic man... :DQuote:
Originally Posted by Blacky
think about it, even a 600mhz 480sp part will be a double gtx280... its not going to be slow... it will be less than nvidia hoped for and less than some expect, but it will be able to compete with 5870 and 5970... its not like nvidia will lose the performance battle... if anything it will be a draw with both ati and nvidia winning some here and some there...
cancelling fermi would not make sense... you wouldnt be able to release a new gpu faster by cancelling fermi, you would just lose all the money you invested in it and let ati reign free... im sure they are focussing a lot on their next gen project right now, but fermi wont be cancelled...
performance isnt the problem to begin with, its cost...
and nvidia has two good chances to improve on that, they can improve 40nm yields with a new stepping which they are already working on, and there is 28nm...
those are both risky though... their main problem is cost, and the only safe way to improve on that is cutting fermi down... i really hope they will get a cut down fermi out asap... its good for nvidia, and its good for us... there wont be notable price impacts from a 470 and 480 in low quantities... its even possible that prices dont change at all after fermi launches... : /
no price drops = hw industry slows down even more, consoles eat more market share = worse gaming overall... cause lets face it, consoles CANT replace pcs when it comes to gaming...
hey dude :DQuote:
Originally Posted by Rob Halford
yeah i know, but even geforceti4200 and some other guys who were much worse than him didnt make me use that function :D
plus with all the strange comments he posts he gets quoted frequently so ill end up seeing his posts anyhow... :D
thats what fermi causes jensen for sure, cold sweats and fast heartbeats... and i bet it causes him quite some sleepless nights too :DQuote:
Originally Posted by Nedjo
but luckily he knows he doesnt make mistakes and he just has to insist on his ideas for long enough to be successful :D
You are right, low quantities won't matter. Because no one will be buying it
i dont think so, people WILL buy it, demand will outstrip supply big time, which will drive prices up and give people a headache... ultimately ati wont have to lower prices cause it wont be easy to find fermi cards and they will be expensive...Quote:
Originally Posted by damha
thats what i think will happen for the first month, maybe even until june...
The price of Fermi would depend on the completion, not the speculation on umber of cards. and if the ATi-competition comes with 86-series later this year, or with a 28nm next year, it won't help.
If ATi has an existing or new/refresh GPU to match the Fermi-performance at launch, then will see good prices. But a superior Fermi with no competition won't come cheap.
i dont care what ya'all are saying, my nvidia will come through for me, they allways have and allways will, i for one have £500 UK sitting here just waiting from the nod from Jen-Hsun, i expect an email from him any day, my old mucker