So are you assuming that the GT200 die on 65nm process is +500mm^2?
NVIDIA GT200 Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
So are you assuming that the GT200 die on 65nm process is +500mm^2?
No. This comes from some decent sources. (and no I don't have "sources" other than respectable people posting across the web)
Originally I was told 520-600mm2, my assumption was 530-560mm2.
Then someone came up with a 575mm2 number, which I don't entirely believe since I doubt his "source."
Berek
Regular
Someone have Turkish to English translation ability?: http://forum.donanimhaber.com/m_23295300/tm.htm
I only got gibberish throwing in through an online translator...
HAL posted that pic a page ago. The numbers dotting the Turkish seem to indicate a summary of the current rumors, no? Or is this Turkish site the source of that pic?
If you Google Turkish-English translation the first site does a half-decent (-assed?) job, but one paragraph at a time. Someone patient could piece together some buzzwords for us to puzzle over.
If you Google Turkish-English translation the first site does a half-decent (-assed?) job, but one paragraph at a time. Someone patient could piece together some buzzwords for us to puzzle over.
Well i dont know if theres any other translators out there that does Turkish to English, but this is the best it could do.
Now decipher away!
They keep mentioning 2nd generation architecture. Possibly 2nd generation "unified" architecture meaning they've tweaked or improved areas where its first generation "unified" architecture was lacking in, like its triangle setup. They could have changed the MADD+MUL configuration to Dual MADD. But im still surprised no word on DX10.1.
Installation Knowledge olarak gene one each personal knowledge with karşınızdayız. Nvidia'nın GeForce GTX serisine gravitation olarak important some technique data of elaborate elimize communication. Kartların official lansman historic possibility 18 June as much as one each monthly long one each period of time okay in now official dökümanlara hour rapidly reflection unprecedented even if memory data yolu , memory capacity , parallel sum birimlerinin innumerable and beam data technology ensure. This haberimizde sizlere Nvidia'nın GeForce GTX 200 quick yada more widespread olarak declared adıyla GeForce serisinin detaylarından bahsedeceğiz.
Bilindii at the point of GeForce 8800GT'nin G80 sole serisinden more hzl unprecedented and GeForce 8800GTS 512 with GeForce 9800GTX'in technique anlamdaki yaknl end kullanclarda head karklna for what reason all right Nvidia aspect even if modellerin snflandrlmas noktasnda problem yaratabiliyordu. New screen postcard with modellendirmeyi simple indirgeyerek this problem amay hedefleyen Nvidia'nn new name with open market sunulacak first modelleri GeForce GTX 200 quick which will happen.
note - Basically, they are saying that the 8800GTS 512 is similiar to 9800GTX in terms of performance, and that the new video card or screenpostcardGeforce GTX200 gets rid of this naming problem
simlendirme konusunda yaplan deiiklii ksaca hatrladktan afterwards GeForce GTX serisinin technique özelliklerine deinmeye honeymoon. AMD ATi'nin Radeon HD serisinde 256- louse memory data yolu and GDDR5 memory use plan on the contrary GTX 200 quick with together Nvidia memory data yolunu arttryor. Elimize ulaan official informed according to GeForce GTX 280 512- louse GeForce GTX 260 however bit'lik memory data yolu with future. Print beam data memory shipment parallel olarak memory kapasiteleri even if deikenlik gösterecek.
note - something about the memory interface, and maybe talking about GTX260 being 448bit.
1GB GDDR3 memory ( probably 0.8ns) with future possibility GeForce GTX 280'nin on the contrary GeForce GTX 260 however 896MB GDDR3 memory with donatlacak. Mind Nvidia geçtiimiz year use porch G80 sole GeForce 8800GTX and arasndaki different stratejisini GeForce GTX serisine even if yanstyor. Print curiosity passive one each dier point however parallel ilem birimlerinin says. Estimate edebileceiniz at the point of GeForce GTX 280 and GeForce GTX 260 parallel ilem birimlerinin says açsndan even if different gate.
Elimize ulaan official dökümanlara according to GeForce GTX 280 240 number parallel ilem unit GeForce GTX 260 however 192 number parallel ilem unit owner which will happen. PCIe 2.0 melodious possibility screen postcard Nvidia'nn SLI teknolojisinin side sra high tanml videolara gravitation olarak fleet hzlandrma silly PureVideoHD teknolojisini desteklemekteler. 2. jenerasyon unified architecture about kurgulanan GeForce GTX 200 quick screen postcard , Steering-wheel 10 handle presentation even if continuation ediyorlar.
note - soemthing about 2nd generation and could the last sentence mean 10 clusters?
240 number parallel ilem biriminin verdii difficult and 2. jenerasyon architectural mind addition with oyunlarda %50'ye leaf performance increase salayaca claim passive GeForce GTX 200 quick Nvidia'nn AGEIA'y satn alp ürünlerine adapte do bee PhysX teknolojisini even if prop up. The farther side sideways CUDA teknolojisine even if beam sunan new screen postcard hazrlanacak derleyiciler with hedeflenen practical in chart ilemci upon fleet hzlandrma even if yapabilecekler.
note - 2nd generation architecture.. 50% faster in performance over something..
Elimizdeki informed according to GeForce GTX 200 quick screen postcard upon pair DVI and HDTV support tv çk earth money owing. Ayrca pair link HDCP handle even if sunulacak. OpenGL 2.1 with even if melodious possibility new screen postcard with Nvidia moment nominal predominant situation okay 3D leadership pekitirerek yoluna continuation do aim. Trip SLI teknolojisine possibility handle new pattern with continuation do beklenirken next week inside kartlarn bog hour hzlar become at the point of dier technique detaylarnn even if netlemesi be expected.
note - OpenGL 2.1, HDCP blah blah and something about a possibility of a new rendering method for Triple SLi
Now decipher away!
They keep mentioning 2nd generation architecture. Possibly 2nd generation "unified" architecture meaning they've tweaked or improved areas where its first generation "unified" architecture was lacking in, like its triangle setup. They could have changed the MADD+MUL configuration to Dual MADD. But im still surprised no word on DX10.1.
Because when TSMC starts manufacturing its first commercial wafers with a new process, it usually takes some time before ATi or nVidia can use it for their chips. I don't know why exactly is that, but usually, low-power optimized processes (for making less complex chips for mobile devices etc.) come first. Then it takes another few months (at least) before TSMC can manufacture GPUs using the new process. The very procedure of manufacturing, packaging and testing the chips also takes weeks (not mentioning the boards themselves), so in the end. So if TSMC starts offering the 45nm node in Q4'08, it doesn't mean that ATi or nV can launch their 45nm powered Radeons and GeForces in Q4'08.
Basically, it is a reworked G8x/G9x. By the way, these rumours all lack one important piece of information about the chip that I'm sure will surprise you
The main reason is that handheld devices have much longer design times; it takes ~18 months right now from the time a chip is sampling and got a mobile phone design win to the time the end-consumer device is actually available in stores. So by the time mass production starts, yields have already improved and wafer costs have already gone down. Furthermore, the power benefits of new process nodes are a lot more important in that business.
At 65nm, a huge part of it was that TSMC prioritized the handheld process by consumer demand. However, on 45/40nm, things have changed a lot: http://notforidiots.com/TSMC.gif
In fact, 40G (aka 45GS on that chart, exact same thing, marketing ftw) was made available in early/mid Q2, while 40LP was/will be available in late Q2. So it's still a fair bit behind 45LP, but 40G is now actually ahead of 40LP. Interestingly, on 32nm, it appears that 32G will come out in 4Q09 while 32LP has been delayed to 1Q10.
The reason for this is simple: process transition in the handheld industry is slowing down slightly, in part due higher levels of non-digital integration, while it's as important as ever in the GPU industry - and NVIDIA is TSMC's largest customer right now, so if they want to prioritize a process *and* ATI agrees while the LP guys don't really care, then it sure as hell is going to happen. It's worth pointing out that Qualcomm is bigger in terms of overall fabless revenue, but they're more diversified at multiple fabs than NV so I'm pretty sure they're behind them at TSMC.
Well obviously it's not like the process didn't exist before; those dates represent 'risk production' schedules. On LP that tends to be Qualcomm, while on GP it tends to be FPGAs. Anyway, for 40G, as I said, that date was early/mid Q2.
However I firmly believe traditional models for when 40nm chips will be available simply aren't viable this time, because the leap in perf/mm² is so incredibly massive and GPU redundancy mechanisms have improved so much in recent years. And the current GPU line-ups are extensive and don't need to be replaced for API reasons, so there aren't many reasons to play it safe. I think I said this already, but I wouldn't be surprised at all if we saw the first 40nm GPU in mid/late Q4 or early Q1.
Pixie dust? TBDR? eDRAM? MIMD microcontroller? FP80 support? Combination of all of the above?Basically, it is a reworked G8x/G9x. By the way, these rumours all lack one important piece of information about the chip that I'm sure will surprise you
Right Anyhow, this is highly anecdotal, but regarding 40/45nm: May I point out that NVIDIA announced their design flow decision at pretty much the same time this year for 45nm as they did for 65nm last year?
http://edageek.com/2007/01/24/nvidia-gpu-magma-quartz/
http://files.shareholder.com/downlo...4824/LAVA_News_2008_1_30_General_Releases.pdf
Anyhow, this is highly anecdotal, but regarding 40/45nm: May I point out that NVIDIA announced their design flow decision at pretty much the same time this year for 45nm as they did for 65nm last year?http://edageek.com/2007/01/24/nvidia-gpu-magma-quartz/
http://files.shareholder.com/downlo...4824/LAVA_News_2008_1_30_General_Releases.pdf
leoneazzurro
Regular
Pixie dust? TBDR? eDRAM? MIMD microcontroller? FP80 support? Combination of all of the above?
Clocks?
Thanks for the comprehensive update
Ah well, it probably won't surprise the likes of you. It is a "step back" in a certain way - not that GT200 would be any worse than it's predecessors, but I'm intrigued why nVidia made such a design decision.Pixie dust? TBDR? eDRAM? MIMD microcontroller? FP80 support? Combination of all of the above?
By the way, isn't pixie dust some sort of hard drive related technology from the 90's?
Thanks for the comprehensive update
Ah well, it probably won't surprise the likes of you. It is a "step back" in a certain way - not that GT200 would be any worse than it's predecessors, but I'm intrigued why nVidia made such a design decision.
By the way, isn't pixie dust some sort of hard drive related technology from the 90's?
No double shader clocks?
Jawed
Legend
That would be an extraordinarily stupid design decision for an ultra-high-end part that'll often be benchmarked at 2560x1600 where bilinear performance matters a lot... I would be more willing to bet that some texture formats have become more expensive or something along those lines.
Jawed
Legend
Jawed
Legend
Jawed
Legend
- Status
