Yes, also like the original Timna design.Quote:
Originally Posted by xlink
The next step we are taking now after the IMC is PCIe and GPU ondie.
Printable View
Yes, also like the original Timna design.Quote:
Originally Posted by xlink
The next step we are taking now after the IMC is PCIe and GPU ondie.
Where are you getting this info? Links/info please.Quote:
Originally Posted by Shintai
nVidia must slash prices but part of me feels that the performance of the GTX 280/260 is going to improve immensely over the next few months as nVidia have a track record for raining on ATi's parade. Already nVidia have pulled out some driver magic for G80/G92 cards in their latest OFFICIAL 175.19 WHQL drivers. I gained 400 3dmark Vantage Points and 700 3dmark06 points. Gaming FPS was up as well.
I would not be surprised if just before the R700 launch or just after the HD4870 launch nVidia release a driver for their GT200 series which increases performance across the board as well (and perhaps adds official physx support).
John
.........As if ATI can't unleash "magical" drivers........:stick:
Of course they can, and so can NVIDIA... the difference is who has the most to "unleash"... which party and architecture has the most "untapped performance" yet to offer.Quote:
Originally Posted by Kingcarcas
Of course they do. Nehalem is basically Shanghai with less cache. Although Nehalem does have 3 memory channels and hyper threading. If AMD can get a dual socket system running for cheap ... it should be a superior solution.Quote:
Intel is so far ahead of AMD in CPU land that I wonder if AMD has a chance of ever being competitive again.
You forgot the 20-50% faster IPC aswell, and higher clock, better interconnect :ROTF:Quote:
Originally Posted by 0g1
ummm sorry i dont think fusion is just IGP on a CPU...Quote:
Originally Posted by Shintai
intel boasts abt not requiring GPUs
nvidia boasts abt not needing CPUs
the only sane person in this whole argument seems to be AMD.. who say both are necessary.. all theyre doing is implementing both gpu and cpu on the same level. its is well known that cpu and gpu are co-existent and cannot be run without the other.. my point is im sure they pulled out all those techies to put into fusion research for a reason and not to only have another implementation of "IGP on Processor"... of course this is speculation but i dont think you can just call it another IGP on CPU....
anyone know any technicals on this... is there any way amd can match performance of discrete with fusion ?? possibly used the integrated memory controller ? wat ? any thoughts ?
Quote:
Originally Posted by eleeter
<cough> Pot, meet kettle.Quote:
Originally Posted by eleeter
Not even close to that I bet.Quote:
Originally Posted by loutsos
How would you pack a CPU and a discrete performing GPU in 1 package? Specially for mobile and value. Fusion is like value Nehalem, IGP on a CPU package/die. Also an IGP already use the IMC...Quote:
Originally Posted by saveus222
:DQuote:
Originally Posted by gojirasan
.
ZZZ stop chanting that "System-on-a-chip FTW" bs. Its not a feasable idea for anything but the most specialized usage. Huge hunks of silicon arent a bit part of the future, it's starting to become obvious. Pushing down to the theoretical tiniest production scale sure as hell wont help with that.Quote:
Originally Posted by Shintai
Discrete GPUs do have a future (likely to perform much more functions than now tho). Unless you mean about the year 2100, which sounds to be a decent point in time for those quantum computers to become mainstream :p:
Given the sheer quantity of AMD fanboism that has overtaken xtremesystems I feel it is necessary for me to state emphatically that I am NOT an AMD fanboy. However, the more I think about things the more it seems that AMD may have gotten things right. My thinking is as follows. I believe that CPUs are not going to scale efficiently or even at all past 4 cores for most applications. Only embarrassingly parallel problem domains will scale (graphics, video transcoding etc) There are a whole lot of inherently serial processes that are going to get in the way of this dream of continuing to benefit from the transistor growth of Moore's law with multi-core instead of clock speed increases. But what to do with all those extra transistors and die space when the process goes to 32nm? Well AMD could just put a couple of 5870 GPUs on the rest of the die. Of course you would need proper motherboard support in the form of GDDR7 or whatever on the motherboard itself. But I could easily see this as the wave of the future. Does it screw with the upgrade cycle by requiring a CPU upgrade along with a GPU upgrade? Yup. But I think we are going to be living in a world of HUGE diminishing returns in the CPU world post-Nehalem. So CPUs dies may be cheaper. At least without some fundamental advances in CPU manufacturing which would allow us to continue to ramp up clock speed. If I'm right about this it could spell trouble for Nvidia and Intel. Cue the rumors of an Nvidia acquisition by Intel in a year or two. Unless Larrabee works out, they may have no choice but to at least try to acquire Nvidia.
Edit: Also if CPU progress is about to grind to a near halt (in a sense this has already happened), it will give AMD more than enough time to catch up with Intel in terms of CPU tech.
Well, AMD hassnt performed great recently. nvidia has had a more gradual decline over the past month.
Heres the 1 month chart.
Remember Intel has always been interested in Nvidia. It has always been the price of entry that has tempered their thoughts.
Reasons could include:
Intellectual property desired
Personnel desired
Simply beating someone by acquisition (the Microsoft method, which Intel has more than enough resources to manage)
Driver expertise and experience, or IP rights
The knowledge and desire to integrate Nvidia's tech onto their die, not just SLI on their chipsets
Since AMD set the precedent, it is unlikely this would get halted on monopoly or market grounds. If that were going to happen, the AMD acquisition would have failed, because you can bet Intel wanted that acquisition reviewed thoroughly.
So it isn't as far fetched a concept as you'd think. The likelihood is that Intel has set some kind of an internal "price target" at which they'd be likely to consider taking action. So you'd probably want to know what Nvidia's market cap was at the price drop.
Intel + NV = Blasphemy on a chip.
One is trying to give longevity to the CPU, the other is trying to destroy the need for the CPU in almost every application.
Perkam
Buth if they can make it so one chip does do it all, why not enjoy the benefits?Quote:
Originally Posted by perkam
I wouldn't mind an all in solution, as long as there still is a broad choiche off products to choose from.
That just shows how much they need each other. Of course Nvidia has an elaborate theory as to why a CPU/GPU chip well never be bleeding edge tech, but I think that theory hinges on the idea of continued progress on the CPU front. If CPU tech stalls for a while until some new tech is invented to increase clock speeds again, we are going to have extra die space for GPU type processing. It will be especially compelling if you have something like 2 CPU cores and 2 GPU cores all integrated on the same substrate with power saving tech to shut down whichever cores are not being used. It would just be a win on so many levels.Quote:
Originally Posted by perkam
Exactly but on the other hand I think he's afraid that with such an solution the current market will only get smaller and smaller untill their only a few companies left offering products which in the end will mean lesser inovation and higher prices. That's why I said I wouldn't mind the all in sollution as long as it's not the only option.Quote:
Originally Posted by gojirasan
i mentioned that in my previous post.. im not a technical guy working at the RnD of Fusion... a few years back... no one ever figured Intel would come out with Conroe.. everyone was harping about how AMD 64 was the future and all that.. that there was nothing intel could do to get their performance up.. but we all got slapped on our faces when conroe was announced and released... and many many years before that.. no one knew semi-conductors would exist.. and a few few few many years before that.. edison invented the lightbulb... end of rant :D so see my point is.. im sure theyre workin on something.. it mite come to be or not.. im just speculating.. as we all are.. end of rant again.. now ok ill try and put my thoughts on how this is technically possible.. i mite be right or wrong.. feel free to correct me..Quote:
Originally Posted by Shintai
wat does a discrete card have.. ?? a discrete graphics chip, a pcb, chokes, memory, ?? say you have an on die or on cpu discrete core.. say of the 6 cores 1 is a discrete graphics core.. wouldnt that help.. coz the pci express bridge latency is gone.. memory bandiwdth is much more.. gpu to cpu speak is the fastest possible.. much much faster than a discrete connection... am i missing out one piece of the puzzle ?
i wonder what the current chart will show.Quote:
Originally Posted by ghost101
haha amen to that.. however.. u never know with these guys. suddenly intel buys nvidia and jen sung says i thaw my chinese mathter thold me inthel is sacred be good to intel tho i thot inthel andh we become one.. its holy sacrimony.. i mean u get the point.. lol :shrug:Quote:
Originally Posted by perkam
from what I see Nvidia dropped the most in that chart, AMD about 12% intel around 7% and nvidia over 20%Quote:
Originally Posted by ghost101
The Colors shown may have confused you since the three companies have their own colors and the chart ignores it.
You make absolutely no sense...so because of some random inventions that got nothing in common. Fusion will be able to host a high performance GPU?Quote:
Originally Posted by saveus222
Dont drink and post...or do drugs...
I don't think you understand the situation :)Quote:
Originally Posted by gojirasan
What you're talking about is already in the works, it's called Fusion from AMD, and Nvidia has already stated it is not going down that road ;)
So you see, Nvidia doesn't want to go down the CPU route, but doesn't want to go down the CPU/GPU route either, and by the end of 2009 when Larrabee is out, Intel will no longer have a need for buying a $11B graphics behemoth.
Perkam
You seem confident in your posts, you must work with ASICs yourself. Can you tell us more about why it shouldnt work? I am interested in CPU/GPU architecture myself and are currently studying for a bachelor degree of electrical and computer engineering.Quote:
Originally Posted by Shintai