I'd dislike it very much if they really abandoned the highest end GPUs for now. To bring the problems of AFR further down the product line is just bad. I hope they make the largest chip that is economically feasible.
Where do you get 2x? GTX580 does 1.581TF, well over 2TF is less than 2x. 2TF would just be 20%, so basically nothing. If that is indicative of gaming performance, that is.
Or they are only talking about DP performance in this piece, then it would make sense. On the other hand, the talk about a multi-GPU chip - and that is only for gaming aka SP performance (for the simplicity of argument).
Or they are only talking about DP performance in this piece, then it would make sense. On the other hand, the talk about a multi-GPU chip - and that is only for gaming aka SP performance (for the simplicity of argument).
Its DP.
Anyway, seems they are something is not listed, maybe there is a trick up nvidias sleeve. Where is the GK100? i dont see them scrapping that name.
uhmmmmm????
Anyway, seems they are something is not listed, maybe there is a trick up nvidias sleeve. Where is the GK100? i dont see them scrapping that name.
uhmmmmm????
http://www.fudzilla.com/graphics/item/25015-nvidia-kepler-roadmap-shows-up
Looks like Fudzilla is now reporting this as well (Noting the same source). I don't know if it makes it any more accurate, but it's still the latest gossip. If true, we're going to see expensive 7000 series cards. Even if they are released a reasonable prices, they'll end up going up after launch like the 5000 series did, I'm afraid.
Looks like Fudzilla is now reporting this as well (Noting the same source). I don't know if it makes it any more accurate, but it's still the latest gossip. If true, we're going to see expensive 7000 series cards. Even if they are released a reasonable prices, they'll end up going up after launch like the 5000 series did, I'm afraid.
If, as so many people like to say on here, that 90% of Nviida's profits in the video card segment come from selling to the professional market (i.e. their Quadro and Tesla cards), then doing a bottom-to-top release does not make as much sense for them. I don't believe this roadmap is true.
If Nvidia was bent on releasing smaller 28nm chips before the Kepler flagship comes out, it will be with die shrunk Fermi's for the laptop space, to fill the gap between now and Ivy Bridge.
If Nvidia was bent on releasing smaller 28nm chips before the Kepler flagship comes out, it will be with die shrunk Fermi's for the laptop space, to fill the gap between now and Ivy Bridge.
If they end up late it suggests desktop graphics isn't their imediate priority - perhaps it's taken a back seat to tegra or denver? perhaps they are designing one of the next gen console gpu's?
I suspect that the 512b bus has more to do with the needs of NVs professional products (especially for compute purposes).
blech, these rumors are driving me nuts
I agree, with the slow down in the pace of computing advancement, the amount of rumors has increased and become even less reliable, IMO.
Unless your 5870 dies in the meantime. lol A 470 died on me after 1 1/2 years. They don't make em like they used to. Lifetime warranties are going to be worth more with life cycles increasing. I would expect a small premium increase for video cards that have them.
Denver isn't silicon yet - it's not even going to be on 28nm. Tegra development does not affect Geforce development, or vice versa. They have different engineers working on both projects.
JTsyo
Lifer
- Nov 18, 2007
- 12,032
- 1,132
- 126
hmm I guess the 560 did come out at $200. I might still wait 2 or 3 months to shake out which the one to go with. I'm partial to egva.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
Without going into specific financial information, generally speaking, the discrete graphics division is also very important. Certainly, NV does not make only 10% of its profits from discrete graphics vs. professional graphics cards. While the profit margins are incredible in the professional space, the volume is much lower.
The delay most likely has to do with manufacturing/power issues, etc. I don't think NV would purposely delay Kepler knowing HD7000 series is just around the corner. NV's goal is probably to launch as soon as it's financially feasible (i.e., once yields on their chips are good enough).
Again, just like last time when Fermi was 6 months late, it barely affected NV, despite the amount of false information being spread on our forums (i.e., NV is finished, etc.). It's public information that NV revised GPU generations to 24 months from 15-18 months (as was stated by them several times). Also, I don't see what the big deal is for one firm launching faster than the other, even if it's by 3-4 months. It has happened in the past. If Kepler is faster or offers better price performance, then there are still at least 18 months of sales before next big generation. The doom and gloom of launching by a quarter later than HD7000 series is blown way out of proportion - Fermi has shown that NV was easily able to make a competitive product lineup and maintain a 58-59% discrete GPU market share even after being 6 months late....
Not to mention, the market for $450+ GPUs is pretty small. NV can cut prices of GTX560/570/580 cards to still make them competitive with HD7850/70 series. Also, by the looks of it, the lower end HD7000 series won't be anything special either. In the short-term, NV only stands to lose the performance crown at the $300+ level. Even then, HD7000 series is launching in the historically slow for hardware sales Q1. It's not the same as when HD5800 series launches in the Fall of 2009, way before holiday season!!
If NV's Kepler generation is worse than HD7000, then NV has a problem. But ever since 8800GTX, NV has had the faster lineup every generation, forcing AMD to compete on price/performance (outside of dual GPU cards). Considering Fermi is already a scalar architecture, and its Tessellation performance is miles ahead of HD6900 series, there is a lot more risk with GCN rather than with Kepler. Fermi architecture has excellent scaling with higher clock speeds. NV can increase Fermi's specs without any architectural improvements, and it will already be a beastly GPU. But of course, NV is going to improve the architecture even further.
Last edited:
exar333
Diamond Member
- Feb 7, 2004
- 8,518
- 8
- 91
Generally production costs get lower the longer something is built as well, so it's easier to lower the prices. I agree though, this is a competitive price-cut, and makes sense. If the 580 is reduces to ~$400 permanently, that tells us where AMD 7000 performance may lie. And if the 7970 doesn't outperform it, then Nvidia has the older hardware ('spec' shoppers like the newest) but a better price plus better availability early on.
Availability is important because if you only have 5 in-stock products competing against 30 on a big-site like Newegg, guess where the sales will be?
It's all about their respective approaches to the HPC market.
Intel - CPU focused
AMD - CPU + GPU focused
nVidia - GPU focused
Fermi's dual use gpu/hpc design was a response to the announcement of AMD's fusion strategy followed by Intel's similar strategy, which would inevitably lock nVidia out of the low end and mainstream x86 market. NVida's answer was to focus on the growing HPC market.
AMD having a CPU + GPU HPC strategy can afford a considerably more GPU focused design for their new GPU chips meaning nVidia has the same fundamental disadvantages againt AMD in the next gen GPU boards as it did in the last. NVidia chips will be more complex and have a larger die area than AMD chips for comparable pure grahics performance and, as with Fermi, be releasing on a complex new node with teething pains.
Hence the AMD advantage in time to market for their next gen GPUs. Charlie D's article AMD had a 6+ mo. tapeout advantage on nVidia apppears to be on the mark and while TSMC is having problems ramping up their yields on a new process, AMD's current stepping on their HIGH end chip appears to be sufficient for mass production with nVidia lagging sustantially in achieving their final stepping for their LOW end chip. AMD starting with it's high end chip and nVidia with it's low end chip making a pretty convincing argument AMD is FAR ahead of nVidia with the next gen GPU timing.
A six months lag for nVidia on it's 69xx class chips is probably a best case scenario.
Intel - CPU focused
AMD - CPU + GPU focused
nVidia - GPU focused
Fermi's dual use gpu/hpc design was a response to the announcement of AMD's fusion strategy followed by Intel's similar strategy, which would inevitably lock nVidia out of the low end and mainstream x86 market. NVida's answer was to focus on the growing HPC market.
AMD having a CPU + GPU HPC strategy can afford a considerably more GPU focused design for their new GPU chips meaning nVidia has the same fundamental disadvantages againt AMD in the next gen GPU boards as it did in the last. NVidia chips will be more complex and have a larger die area than AMD chips for comparable pure grahics performance and, as with Fermi, be releasing on a complex new node with teething pains.
Hence the AMD advantage in time to market for their next gen GPUs. Charlie D's article AMD had a 6+ mo. tapeout advantage on nVidia apppears to be on the mark and while TSMC is having problems ramping up their yields on a new process, AMD's current stepping on their HIGH end chip appears to be sufficient for mass production with nVidia lagging sustantially in achieving their final stepping for their LOW end chip. AMD starting with it's high end chip and nVidia with it's low end chip making a pretty convincing argument AMD is FAR ahead of nVidia with the next gen GPU timing.
A six months lag for nVidia on it's 69xx class chips is probably a best case scenario.
JTsyo
Lifer
- Nov 18, 2007
- 12,032
- 1,132
- 126
Unconfirmed timeline
Looks like we still have a while to wait, especially those looking for the high end.
Looks like we still have a while to wait, especially those looking for the high end.
That's the chart that stirred things up and has been showing up at more and more HW sites. We'll see.
just saying
-what if the chart is wrong [names -descriptions]
-shouldn't the dual card be at the top to replace the 590 dual card and fitting into the timing of dual cards after the high end singles.
-the high end 512 bit single card replacing the 580 single-
-if so the gtx580 replacement could be 3-4 Q ?
-GK110 could be the single
-GK112 could be the dual in the same time frame of the chart.
-was going to mention that when I first looked at it , but did not think it would be around for this long.
-what if the chart is wrong [names -descriptions]
-shouldn't the dual card be at the top to replace the 590 dual card and fitting into the timing of dual cards after the high end singles.
-the high end 512 bit single card replacing the 580 single-
-if so the gtx580 replacement could be 3-4 Q ?
-GK110 could be the single
-GK112 could be the dual in the same time frame of the chart.
-was going to mention that when I first looked at it , but did not think it would be around for this long.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 396
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
Facebook
Twitter