![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Fermi Working Samples for CES?
- Thread starter FrgMstr
- Start date
realistically, the 5870 runs on par with the 295. I know people argue about 1 gpu versus 2 on a card but it is one card vs one card. (if i scan reviews you see the 295 is actually a little faster). So as a single card solution NV still has the top spot. Is it dx11 compatible? Nah for the games out now and the next few months that does not really matter. I mean lets be honest all NV has to do is drop prices on their 2xx line (260 and above) to battle the shortage of 58XX cards. I would sli two 285's if they were 200 a piece 
...
Not that I am a NV or ATI fan boy, I have both..just from a marketing perspectinve NV has the product line out now to compete with the new ATI cards out now..all they have to do is shift focus from performace to price (a 5870 for 400 or a 295 for 300?)..
What NV is going to have to do is when the 300 line does come out - smoke the competition and do it at a decent price point. Say the gt380 is 100% faster than a 5870,...if they are selling it for 600 bucks I don't think a lot of people will care.
...Not that I am a NV or ATI fan boy, I have both..just from a marketing perspectinve NV has the product line out now to compete with the new ATI cards out now..all they have to do is shift focus from performace to price (a 5870 for 400 or a 295 for 300?)..
What NV is going to have to do is when the 300 line does come out - smoke the competition and do it at a decent price point. Say the gt380 is 100% faster than a 5870,...if they are selling it for 600 bucks I don't think a lot of people will care.
They wouldn't but it won't be 100% faster at gaming.
yeah also gtx 295 is multi gpu setup with much more power draw sli problems etc and how much you think nvidia can afford reducing prices ? (not that ati can't answer to it with its 4000 line)realistically, the 5870 runs on par with the 295. I know people argue about 1 gpu versus 2 on a card but it is one card vs one card. (if i scan reviews you see the 295 is actually a little faster). So as a single card solution NV still has the top spot. Is it dx11 compatible? Nah for the games out now and the next few months that does not really matter. I mean lets be honest all NV has to do is drop prices on their 2xx line (260 and above) to battle the shortage of 58XX cards. I would sli two 285's if they were 200 a piece
Not that I am a NV or ATI fan boy, I have both..just from a marketing perspectinve NV has the product line out now to compete with the new ATI cards out now..all they have to do is shift focus from performace to price (a 5870 for 400 or a 295 for 300?)..
What NV is going to have to do is when the 300 line does come out - smoke the competition and do it at a decent price point. Say the gt380 is 100% faster than a 5870,...if they are selling it for 600 bucks I don't think a lot of people will care.
realistically, the 5870 runs on par with the 295. I know people argue about 1 gpu versus 2 on a card but it is one card vs one card. (if i scan reviews you see the 295 is actually a little faster). So as a single card solution NV still has the top spot. Is it dx11 compatible? Nah for the games out now and the next few months that does not really matter. I mean lets be honest all NV has to do is drop prices on their 2xx line (260 and above) to battle the shortage of 58XX cards. I would sli two 285's if they were 200 a piece
Not that I am a NV or ATI fan boy, I have both..just from a marketing perspectinve NV has the product line out now to compete with the new ATI cards out now..all they have to do is shift focus from performace to price (a 5870 for 400 or a 295 for 300?)..
What NV is going to have to do is when the 300 line does come out - smoke the competition and do it at a decent price point. Say the gt380 is 100% faster than a 5870,...if they are selling it for 600 bucks I don't think a lot of people will care.
at $100 a die for 295 its already 200+ just for the die cost.. then add everything else on top of it.. no way they can sell a 295 @ $300
*snorts*
Okay, that was funny
It's a real shame that nVidia is taking so long this time around. Fermi looks like the long-awaited shift towards a more generic GPU/vector processor design Intel has been hyping for years with its Larrabee. I have no doubt that Fermi as cGPU heralds the future of GPUs, it's just annoying that the future has been postponed for a few more months
Yeah...and how is that potential market working out for Nvidia so far?
They were examples - and it's not like they would be the first company to sell a product at or below cost (hi SONY!! Hiya Microsoft!). I'm not knocking ati the 58XX cards are damn impressive - but honestly are they lightyears beyond the 2XX or even the 48XX cards?
Yeah, see the post above yours. I brain farted on that one.erm... Juniper is 5750 and 5770... Whatever happened to our NVDA stock argument? and Q1 vs. Q2 fermi release... Vengeance you're normally sharper than this...
Die shrunk and clocked higher, but pretty much.
2x 285s would be faster and at 200$ each wouldn't be a bad deal.realistically, the 5870 runs on par with the 295. I know people argue about 1 gpu versus 2 on a card but it is one card vs one card. (if i scan reviews you see the 295 is actually a little faster). So as a single card solution NV still has the top spot. Is it dx11 compatible? Nah for the games out now and the next few months that does not really matter. I mean lets be honest all NV has to do is drop prices on their 2xx line (260 and above) to battle the shortage of 58XX cards. I would sli two 285's if they were 200 a piece
The reason they aren't dropping their prices is because of zero supply on the 5870 front.
$300 would be a loss leader for sure. I'd go snag one for that price. If they gave them away for free, they would take all focus off ATI's new chips too.
You're hyperbowling but yeah. If Nvidia comes out with Femri, clocks it at 650Mhz, and sells it at 400$ they will have a shot depending on what the 5890 refresh cards look like. That'll put it on paper at 20% faster than a 5870 (2.4X a 280). However, thats high for Nvidia clocks, and it assumes a CUDA core is the same power as a GT200 core. I don't think it is a bad assumption to think a GT200 core is at least equal to a CUDA core clock for clock on paper. I suspect there will be a LOT in the way of driver optimization to be done. I don't think there is nearly as much room for the 5870 to mature because of how similar the architecture is to the 4870 in terms of preformance gains post launch.
It is MUCH more power. So much so that the power draw is a real case. I've never been a fan of people talking about 20 watts difference in power draw. This however is an order of magnitude different.
The die cost is probably closer to 50-60$. However I agree they aren't going to sell it at 300$.
Nothing potential market. Professional GPUs like the Quadro have been selling well for years, now Tesla cards are selling like hotcakes in as far that's possible in a high profit margin professional market. Especially hospitals, clinics, research facilities, oil companies and HPC people are the biggest GPGPU market for nVidia at this point.
crustybooger
Limp Gawd
- Joined
- Apr 3, 2004
- Messages
- 436
so I take it I'll be able to buy one in june 2010?!
jediknight73
Bad Trader
- Joined
- Mar 8, 2006
- Messages
- 388
why should price's drop with no competition??
Yeah, Kyle and Charlie are kind of renowned for their knowledge of the inner workings of nvidia/aibs.
Well, at least pounding the pavement to try to figure out what the hell is going on.
While we didn't know jack about G80 until a couple weeks before launch, we knew WHEN it was launching.
Finally, the common consensus is 20% so, for which I agree. Someone in the know at B3D was saying the word was 70% of 5870 crossfire iirc (Trini may recall better than I), which would be ~25%.
I don't know what speed you expect 48ROP/128TMU/1024 Flops/cycle to be running at, but barring some Dawn dust, it's going to be a slower core speed than 5870, as it's a supposedly a ~60% larger die, so it starts to average out.
My personal theory put it at 700/1750 and 640/1600 for the full and 448sp (360) parts with a 2.5/1 ratio which keeps everything at the same pixel/tex/flop ratio as GT285, with the same performance differential between GTX280 and GTX260 for the two respectfully. At those speeds we're talking 70% and 35% faster theoretical than GTX285, not counting the scheduler enhancements and all that jazz in the new architecture that could improve efficiency. Using simple math:
GTX285 was ~10% faster than 4890
GTX380 may be ~70% faster theoretical than GTX285.
5870 is ~60% faster than 4890.
10+70-60= 20% faster. Efficiency enhancements to SFU and such could be anything but ~5% sounds reasonable...So there, there's a realistic theory to back Charlie/B3D guy up.
Yes, that's just an idea of what it COULD be, based on what we know and think we know, and could very well be very wrong. Point is though, don't expect miracles that are unrealistic. Look at all the facts before you call him nuts: The specs (white paper), look at the amount of transistors and therefore size comparatively speaking to former nVIDIA gens and chips on the process, the speeds capable per TDP on the given process, and there you go. Charlie doesn't just 'guess'....well, not ALL the time.
I feel he usually has a grip on the sitch one way or another through sources or looking at the facts himself. Doesn't mean I don't think he's sensationalist in his writing, but you gotta give credit where it's due when he has a source or breaks/publicizes a story, IE the AIBs to not expect Fermi until April/May.
Last edited:
Read the financials. The professional Quadro business at nVidia is 0.2% of its revenue. The margins are nice, but the volume is tiny
Unless that market segment starts growing explosively starting Q1 of 2010, and keeps exploding for several quarters in a row, it will be irrelevant to Fermi's success
Maybe a Fermi successor in 2012-2015 starts to bring significant revenue. I guess that's what's keeping nVidia investing in those extra transistors
Since graphic card generations get obsolete every 18-24 months, nVidia would never recover the cost of designing Fermi, if it had to come from revenue from the professional segment during 2010-2011. The professional segment starting size is too small, and the life of graphic generations too short
Good try to distract investors on presentation day, but every professional investor could see through it next day when they were back at their offices and run the revenue spreadsheet projections
Oscar Meyer
Limp Gawd
- Joined
- Oct 9, 2009
- Messages
- 326
Actually, it's not cherry picking bullshit. Graphics make what % of AMD's sales? and Graphics makes what % of NVDA's sales? AMD investors have always been sweating cause AMD as a company sucks...bleeds money everywhere. However, IF Cyress failed, AMD probably would not. If Fermi failed, NVDA could easily kick the bucket and be on sale for a cheap price. You're statement is like comparing Intel with ATI. It's apples and oranges. If anything fruitful were coming in the coming months for both AMD or NVDA, they're stock prices would reflect it. Obviously there isnt, despite the success of the 5800's.
Last edited:
/sigh. Stop internet road raging for a min.
I was not talking that the 5870 was 20 watts more than a 295. I was refering to the past, where people talked about power draw between cards like a 4870 vs a GTX 275 which from your graph is 31 watts apart. I was talking about the difference between a pair in SLI like 2x 275s and a 5870 which is arguably close on preformance. Yet there is a 243 watt difference in power. I.e. an order of magnitude off the 20 watts for similar preformance I had stated earlier in the post.
TheMadHatterXxX
2[H]4U
- Joined
- Sep 7, 2004
- Messages
- 3,022
It's gonna be what-atleast a 3-5 month gap before NVidia puts something new out since the 58x0 architecture was released? forget by January, think of it this way theres gonna be a crapload of people who will take take their holiday money and buy a 58x0 card and the majority of those people probably will not make another vid card purchase for another 6 months after...Nvidia definitely loses this round.
I don't mind waiting till March for Nvidia's new cards, if that's the case. While nice, the 5870 isn't a big enough boost for me - however for 4850 and 8800GT owners I see it as a very enticing upgrade. I would consider getting the 5870x2 (or 5900 series, whatever the final name ends up being) but it looks like it won't fit in my case.
Flexion
[H]ard|Gawd
- Joined
- Jul 20, 2004
- Messages
- 1,607
That is if they can find 58xx cards before the holidays... Sigh... Want one more...
That is if they can find 58xx cards before the holidays... Sigh... Want one more...
Something a lot of people seem to be missing. Buying a 58xx right now is like playing whack a mole. And judging by what TSMC is saying, it's not going to be any easier to get a 5870 any time soon.
Well I have no idea what fermi is going to do.
oh and this math was odd you can't just add them together since one is a superset of the other. it is 70% of x+10%
GTX285 was ~10% faster than 4890
GTX380 may be ~70% faster theoretical than GTX285.
5870 is ~60% faster than 4890.
I used 2 as the base because 1 does screwy things to math.
4890 A2 = 2
GTX285 N2 = 2.2
GTX380 N3 = 2.2 + 1.54 = 3.74
5870 A3 = 2 + 1.2 = 3.2
so numbers are close just not sure if they have any meaning.
But one of the market nvida wants to exploit is the the companies using hundreds of chips would would love to cut that number in half or even quarters which from the white papers looks possible. They could be blowing smoke up our butts but hey no ones knows. That market lives on high margin gains. Tesla are making money and fermi should be better at the same game.
Next day traders make life interesting... and I don't trade stock and even I understand market caps, splitting stock and such. Nvidia has more non-preferred stock out there than AMD, they are also in the black as opposed to AMD which is in major debit. I understand from an accounts point of view it is better to have people owe you money than have cash on hand (still don't think this makes sense), but owing people money you don't have is probably not a good thing. AMD's graphics divsion is making most of AMD's money right now. So they need hits. Nvidia has enough money they could afford to lose money for two quarters in a row before the OEMs start hurting nivida's bottom line.
Most people at hardocp want new hardware now, but I'd rather prefer that nvidia wait until they have an 8800 instead of a 5800. What is interesting is nvidia prior to the 8800 was in real bad shape they thought to have no chance and the few specs that got leaked everyone laughed at saying there was no way that even a quarter of that would be a good card. The specs were right we got an insane card that ATi had to spend several generation and several engineers from the AMDs processor section to beat.
Now the specs we have are:
The 512 CUDA cores = 16 SMs (x 32 cores)
six 64-bit memory partitions (384-bit memory interface)
up to 6 GB of GDDR5 DRAM
g200 vs fermi
30 FMA ops / clock vs 256 FMA ops /clock
240 MAD ops / clock vs 512 FMA ops /clock
What these numbers mean when translated to game performance is totally up in the air. I'm guessing that the first card is going to be held back by memory speed. Is nvidia going to try and save power and use less sockets on the gamers cards and bandwidth and more on the fermi cards where power is not an issue? samsungs new gddr5 is 1.3v which is cheaper (power and heat cost) than the existing 1.5v and much cheaper than the 1.8v gddr3. Which mean that they can run it faster. The new stuff is supposed to achieve 7GHz which results in 296GB/speak bandwidth compared to the 256 bit and 512bit quoted in the article. Which is on a 40nm process. Does samsung have their own foundry or are they also in the TMSC that is getting bogged down with their 40nm process?
oh and this math was odd you can't just add them together since one is a superset of the other. it is 70% of x+10%
GTX285 was ~10% faster than 4890
GTX380 may be ~70% faster theoretical than GTX285.
5870 is ~60% faster than 4890.
I used 2 as the base because 1 does screwy things to math.
4890 A2 = 2
GTX285 N2 = 2.2
GTX380 N3 = 2.2 + 1.54 = 3.74
5870 A3 = 2 + 1.2 = 3.2
so numbers are close just not sure if they have any meaning.
But one of the market nvida wants to exploit is the the companies using hundreds of chips would would love to cut that number in half or even quarters which from the white papers looks possible. They could be blowing smoke up our butts but hey no ones knows. That market lives on high margin gains. Tesla are making money and fermi should be better at the same game.
Next day traders make life interesting... and I don't trade stock and even I understand market caps, splitting stock and such. Nvidia has more non-preferred stock out there than AMD, they are also in the black as opposed to AMD which is in major debit. I understand from an accounts point of view it is better to have people owe you money than have cash on hand (still don't think this makes sense), but owing people money you don't have is probably not a good thing. AMD's graphics divsion is making most of AMD's money right now. So they need hits. Nvidia has enough money they could afford to lose money for two quarters in a row before the OEMs start hurting nivida's bottom line.
Most people at hardocp want new hardware now, but I'd rather prefer that nvidia wait until they have an 8800 instead of a 5800. What is interesting is nvidia prior to the 8800 was in real bad shape they thought to have no chance and the few specs that got leaked everyone laughed at saying there was no way that even a quarter of that would be a good card. The specs were right we got an insane card that ATi had to spend several generation and several engineers from the AMDs processor section to beat.
Now the specs we have are:
The 512 CUDA cores = 16 SMs (x 32 cores)
six 64-bit memory partitions (384-bit memory interface)
up to 6 GB of GDDR5 DRAM
g200 vs fermi
30 FMA ops / clock vs 256 FMA ops /clock
240 MAD ops / clock vs 512 FMA ops /clock
What these numbers mean when translated to game performance is totally up in the air. I'm guessing that the first card is going to be held back by memory speed. Is nvidia going to try and save power and use less sockets on the gamers cards and bandwidth and more on the fermi cards where power is not an issue? samsungs new gddr5 is 1.3v which is cheaper (power and heat cost) than the existing 1.5v and much cheaper than the 1.8v gddr3. Which mean that they can run it faster. The new stuff is supposed to achieve 7GHz which results in 296GB/speak bandwidth compared to the 256 bit and 512bit quoted in the article. Which is on a 40nm process. Does samsung have their own foundry or are they also in the TMSC that is getting bogged down with their 40nm process?
veedubfreak
2[H]4U
- Joined
- Dec 10, 2007
- Messages
- 2,070
It does everything? But for the first 2 years costs 600 bucks?
jeremyshaw
[H]F Junkie
- Joined
- Aug 26, 2009
- Messages
- 12,511
and how will this be any easier with nVidia, with their larger die sizes?
Gorankar
[H]F Junkie
- Joined
- Jul 19, 2000
- Messages
- 11,107
Not only that, but right now, AMD cards are, for the most part, priced appropriately against their competition. The 5870 is a $100 less, give or take, and a bit slower than the 295 in most games. The 5850 competes well against the 285 for quite a bit less money. The only pricing glitch AMD seems to have right now is the 5770, which is a bit too close in price to the 260, and performs on par to a bit slower. If anything they have room to raise prices on the 5850 and 5870.
Last edited:
What i think is:
Nobody thought about how starving the (gaming-relevant graphics) market was.Nvidias parts were old, i mean, rebatching and shrinking them, didn't actually make them much faster.And AMD were just about catching up with them.
On the other hand, we got more taxing Games,than there were 18 months ago.(or lets say graphically interesting)
As Soon as somebody pushes a card out, that really shows better performance,than previous generations, everybody wants one of these.What a gamer wants, is playing games at full eyecandy, highest possible resolution,and as fluid as water.
That is exactly what Ati concentrated on. (and delivered)
Fermi is something planned for "computing graphics and physics".Nice for Industry and science, and i expect the full-on monster(the 380) to be way more expensive, than my complete high end computer. No need to go that far for a gamer, even with a 30-inch monitor, or 3 of them, for that matter.lol
But if you are a scientist, you might get the same computational power than with a server cluster, for half its price, might be worth it,indeed.
For it to work, it ll need its own toolbox and that is something, Nvidia is really good at.Their success is not to a small part due to their drivers/tools.
Question is: Is the Market asking for it? and is it big enough to feed Nvidia enough?
Nobody thought about how starving the (gaming-relevant graphics) market was.Nvidias parts were old, i mean, rebatching and shrinking them, didn't actually make them much faster.And AMD were just about catching up with them.
On the other hand, we got more taxing Games,than there were 18 months ago.(or lets say graphically interesting)
As Soon as somebody pushes a card out, that really shows better performance,than previous generations, everybody wants one of these.What a gamer wants, is playing games at full eyecandy, highest possible resolution,and as fluid as water.
That is exactly what Ati concentrated on. (and delivered)
Fermi is something planned for "computing graphics and physics".Nice for Industry and science, and i expect the full-on monster(the 380) to be way more expensive, than my complete high end computer. No need to go that far for a gamer, even with a 30-inch monitor, or 3 of them, for that matter.lol
But if you are a scientist, you might get the same computational power than with a server cluster, for half its price, might be worth it,indeed.
For it to work, it ll need its own toolbox and that is something, Nvidia is really good at.Their success is not to a small part due to their drivers/tools.
Question is: Is the Market asking for it? and is it big enough to feed Nvidia enough?
That must have been one serious boo-boo over at TSMC.
5870 listed to be in stock on 11/30/09
http://www.newegg.com/Product/Product.aspx?Item=N82E16814161301
If ATI can't even get a working design out there's no chance Nvidia is going to have anything in retail for a while.
5870 listed to be in stock on 11/30/09
http://www.newegg.com/Product/Product.aspx?Item=N82E16814161301
If ATI can't even get a working design out there's no chance Nvidia is going to have anything in retail for a while.
crustybooger
Limp Gawd
- Joined
- Apr 3, 2004
- Messages
- 436
yup, looks like it's gonna be a long wait. :|
honolululu
n00b
- Joined
- Jul 19, 2007
- Messages
- 44
So, as I see it, this TSMC situation is very good for Nvidia. If they can get enough product for testing, and by the time Nvidia's engineer's work out the hardware, TSMC will be ready to pump em out. Along with AMD's refresh, setting up a nice price war and we win!
ApaThy_NeXT2
[H]ard|Gawd
- Joined
- Sep 7, 2002
- Messages
- 1,225
Can just imagine Jen-hsun in a ninja suit sabotaging TSMC machinery.
WBurchnall
2[H]4U
- Joined
- Oct 10, 2009
- Messages
- 2,622
Yes, but please less spandex.
That quote you posted from Ask Nvidia A Question thread is quite interesting. If nothing else, nVidia as a company has a great marketing team and honestly their answers made it sound like nVidia was still relevant? competitive? and planning something really 'special' with Fermi. Especially when they roughly discussed that Fermi has a ...method to calculate tessolation that they'd be 'Announcing soon!'. I guess maybe they will try to use some of those Cuda Cores to do tessolation? Who knows.
Personally, I think it was really great marketing spin and great marketing is as valuable as fertilizer...to someone who isn't a farmer...or a gardener...or well...you get the point. I think someone needs to ask in that thread, 'Why have we not seen a single benchmark of Fermi? Does Fermi play games?'
That quote you posted from Ask Nvidia A Question thread is quite interesting. If nothing else, nVidia as a company has a great marketing team and honestly their answers made it sound like nVidia was still relevant? competitive? and planning something really 'special' with Fermi. Especially when they roughly discussed that Fermi has a ...method to calculate tessolation that they'd be 'Announcing soon!'. I guess maybe they will try to use some of those Cuda Cores to do tessolation? Who knows.
Personally, I think it was really great marketing spin and great marketing is as valuable as fertilizer...to someone who isn't a farmer...or a gardener...or well...you get the point. I think someone needs to ask in that thread, 'Why have we not seen a single benchmark of Fermi? Does Fermi play games?'
ChrisRay and Sean asked nVidia to help on the "software only Tessellator rumor".
Got Nvidia to weaken a bit on their resolve about Fermi graphics. With our latest questions we pushed to have the Fermi Tessellator rumor put to silence for good.
Yes Fermi has a dedicated hardware for for tessellation. So glad I can stop biting my tongue about this one
jeremyshaw
[H]F Junkie
- Joined
- Aug 26, 2009
- Messages
- 12,511
Better then the 'hardware assisted' vertex shaders for the mx440 (go, for me ) that were finally dropped in the post 90.xx drivers. (iirc, too long ago).
) that were finally dropped in the post 90.xx drivers. (iirc, too long ago).The margins are nice indeed. Forget revenue, look at % profit per unit sold and look at the further potential in the rapidly growing GPGPU market. Most of the profit at this point for nVidia and ATi is in mainstream and budget GPUs whereas high-end units are loss-leaders. The GPGPU market could potentially reverse or at least more equalize this situation. Future super computers could be constructed out of thousands of cGPUs like Fermi instead of racks of quad-core computers. This has been done once already and so far it's a huge success. Making GPUs more general-purpose only makes this a more attractive option in the future.
I'm just saying
technogiant
Limp Gawd
- Joined
- Aug 17, 2009
- Messages
- 181
All this talk about the Fermi delay......If you can't actually buy a 5870....whats the difference?