Friday, February 10th 2012
NVIDIA GeForce Kepler Packs Radically Different Number Crunching Machinery
NVIDIA is bound to kickstart its competitive graphics processor lineup to AMD's Southern Islands Radeon HD 7000 series with GeForce Kepler 104 (GK104). We are learning through reliable sources that NVIDIA will implement a radically different design (by NVIDIA's standards anyway) for its CUDA core machinery, while retaining the basic hierarchy of components in its GPU similar to Fermi. The new design would ensure greater parallelism. The latest version of GK104's specifications looks like this:
SIMD Hierarchy
Source:
3DCenter.org
SIMD Hierarchy
- 4 Graphics Processing Clusters (GPC)
- 4 Streaming Multiprocessors (SM) per GPC = 16 SM
- 96 Stream Processors (SP) per SM = 1536 CUDA cores
- 8 Texture Units (TMU) per SM = 128 TMUs
- 32 Raster OPeration Units (ROPs)
- 256-bit wide GDDR5 memory interface
- 2048 MB (2 GB) memory amount standard
- 950 MHz core/CUDA core (no hot-clocks)
- 1250 MHz actual (5.00 GHz effective) memory, 160 GB/s memory bandwidth
- 2.9 TFLOP/s single-precision floating point compute power
- 486 GFLOP/s double-precision floating point compute power
- Estimated die-area 340mm²
139 Comments on NVIDIA GeForce Kepler Packs Radically Different Number Crunching Machinery
What is there to comprehend? Benetanegia replies to your posts with a well thought out reply and starting with post #64 you basically do nothing more than call him a fanboy.
1-gk104 was meant to be the mid/high range card and not the high end
2-usually cards like the gk104 might end up in the mobile segment therefore must be built with both worlds in mind(tho im not sure whether nvidia uses the second fastest desktop chip for mobile or the third fastest so correct me if im wrong)
3- considering the fact it was designed to be the med/high there are 2 reasons on why nvidia would choose 256bit, the first reason is to purposely limit the performance to allow for faster and more expensive cards to be released later or to place the card in the preferred place they desire in the market in terms of performance and price (similar to what they did with gtx460 768mb and gtx460 1gb) the second explanation could be that the card doesnt benefit from more bandwidth and would only make it more expensive for minor gains.
4- we dont know whether nvidia will call the gk104 gtx660 or 680, and if they do call it gtx680 is it because they failed to release gk110 due to yield issues or is it because the gk104 is sufficient to compete?
if it does end up to be gtx680 it would probably be the first time nvidia would have smaller die sizes than amd, tho after all the issues they had with fermi and manufacturing such a change in methodology is not all that shocking
but overall nvidia seems to have learned alot from amds strength(i wish amd would do the same from nvidia) as amd(the graphics division strictly) had more experience with facing manufacturing difficulties and knowing what to expect, and used to work according to due dates,i read an article by one of the chief engineers at ati where he explains how things work at ati(i will try to find the article and post it up, it was about rv670 and how amd jumped the wagon first for 55nm and how the process of releasing products works)
Don't get it twisted... It's Nvidia's design that AMD copied, tweeked it, called GCN and then said "look we win" lol .... Or you could say Nvidia came out with the Big Mac and over one year later AMD came out with Le' Big Mic! lol Now suddenly, all the AMD/ATI noobs that don't know the history or any better say "look a new sandwich, we win!"... lol It's Nvidia's game... always has been and always will be. AMD is just playing at it. And to prove it, the 700 series is gonna come out and steal AMDs thunder (again) till they catch up ANOTHER year or 2 later. :bounce:
The ONLY time ATI/AMD had any real self-ingenutiy advantage over NVIDIA was way back when Doom3 (@2004 over 8 years ago) came out because AMD had onboard decoding and encoding and Nvidia's driver based codeing had a serious problem whith DOOM3 when it first came out. That small 6 month to a year hic-up was AMD's ONLY shinning moment over Nvidia and Nvidia has been knocking ATI/AMD in the dirt ever since. AMD is still trying to catch up and you're just catching one hop of the leap frog that happens to be AMD's turn... But again, the only jump forward AMD is making is that they got smart, ditched the single stream VLIW4 6000 series architecture and copied to a similar multistream/reading Fermi style architecture. It's just clever marketing that they are describing their newest chip design in reviews like it's revolutionary... well it WAS revolutionary... over a year ago when Nvidia CREATED IT and called it FERMI!
www.nvidia.com/object/fermi_architecture.html?ClickID=azzwwsat59szk05t0aa0sll0zsrttknlzsks (<that is actually over 1-2 years old despite the 2012 copyright because it states the 512 cuda cores which has already been in exsistance since the GTX 580)
www.anandtech.com/show/5261/amd-radeon-hd-7970-review (<and there is AMDs copy cat GCN version)
Don't the GNC core pictures (page 3) for the Radeon 7970 look familar ?? lol They should, similar architecture profile pictures were in the previous link's FERMI architecture press release over a year ago! lol The GCN architecture = a.k.a Le' Big Mic lol - over a year later lol
... so stick those theorectical copy cat numbers up your GPU. lol I doubt you'd believe anything but the crud AMD Radeon is selling you (cause they're second best) anyways. People are so ignorantly sold on the "drama" that Nvidia is the big selfish Giant and AMD the independent hero that people don't actually research or know the facts... If it weren't for the Fermi architecture with better Dx11 FPS, better Tessallations and multi-reading streams, we wouldn't have the FPS or the Dx11 Games that we do. Hardware always spurrs on new games and software that take advantage of the hardware technology and Nvidia's Fermi architecture spurred Dx11 and tessallations GREATLY in games like Skyrm, BF3, Batman AC, Witcher 2 and more ... you should be thanking them, honestly. I know AMD secretly is because without coping FERMI AMD would be AM-DONE. lol rolfloflmaoftwpwn!
AMD could e easily built a die that big nbfit more transisters and would still match gtx580 and at the same power consumption or even lower, if you truly did ur research you would know that, but I agree Fermi did introduce revolutionary tech in the same way bulldozer did, tech that will only mature in time but wasn't all refine at release
I'm also very sure that based on the fact that nvidia pioneered the fermi architecture and that it's still fairly new and fresh that it has a long way to go and more possibilities to hit because what's the sense of creating a whole new architechture if it doesn't have headroom to accomidate the next several years ? And is that not an architecture that Nvidia created and AMD is merely copy catting one step at a time?... I'm sure that Nvidia has years of plans for Fermi and limitless possibilites THEY CREATED IT.. I'm sure they know full well all it's avenus and applications; that it's leaps and bounds beyond AMD in a similar architecture. That the 700 series will infact crush the current wannabee attempt from AMD because currently (in the $200-$300.00 market where the real market war is won and lost) AMDs barely released 7850 had referrence clock benches BARELY beating the Nvidia's 560 Ti (which has been a phenomenol price point card!) in Dx11 and failed misserably in Dx9 which Nvidia still does well. Not to mention, that the 7850 has 1gib more than the 560 Ti and the reference board only performed marginly better than the referrence 560 Ti. Now what's the big deal with Dx9? Not much really except the actually the most demading game for video rendering and drawing right now is actually SC II in 4v4 mode which is Dx9 and the 560 ti still killed the 7850 by 20FPS. So AMD is releaseing a barely better than version that does worse on the old stuff ? noooicccce. Again, no fear whatsoever that Nvidia is going to knock AMD back in the dirt!
Do you work in any technology manufacturing field ? I do, I work with years apon years of laser technology. I know what stages are involved in many avenues of product development from design to deploy to da money! I know that something you see tomorrow from Nvidia was probably on a drawing board 2-3 years before you saw it, that they probably had a mock up model 1 year ago and a functioning test model atleast 6months to 1 year ago...lol And I certainly know how to do my homework but thanks for presuming I don't just because you do and I didn't mention something off topic. Learn not to presume so much about what's not said lol :nutkick:
PS... I've even done field trips lol .. I've been (on more than one occasion) to one of ATI's engineering facilities and talked first hand with PCB, CPU and RAM engineers on Radeon cards AND have worked for a major gaming company's coporate center and have worked with unrestricted models of the Nvidia 8000 series... have you ? lol
Pss.. Bulldozer is a joke and an extreme dissapointment at the time to any true gamer with half the sense to read benchmarks because single threading was horrible and @98% of the games out there are single threaded....Windows had to come out with a patch just to set it right. Plus, all they did was create hyperthreading pipes so large you could fit a truck through them. Lol a lot of good that did for gaming (which is pertinent to this discussion because you mentioned a gaming graphics card and the Bulldozer in the same paragraph pressumably as having linked innovation) because the only games I know of for hyperthreading are Civ V and Oblivion... PLUS, That's like putting 22's on a Bonneville and calling it badass and innovative! rofl And the Bulldozers still didn't top i7s; and, the first batch of FX 6100 that came to my local CompUSA were all BAD and wouldn't work with the 990Fx chipsets lol. I stood there watching as my good tech friend behind the counter put every FX6100 chip they had left on a chip tester/meter and one after another were bad! This was only after 2 different customers came in to swap FX6100s they just bought 2-3 times over! If you're going to use an example like that then YOU should deffenitely DO YOUR RESEARCH and maybe a little real-world expierence first! LoL And, pick a better example lol :nutkick: :nutkick: :nutkick: And before you start with the 2nd gen Iseries B2-B3 problem let me stop you there. That was every bit a Motherboard bridge problem effecting Sata III and affected motherboards and MBoard manufactures not the fault of the CPU itself that they didn't get the chipset right.
So to recap: before you start a rebuttle and start spitting out stuff you clearly need to reasearch or look at yourself better
1. make sure your points are on topic or relevant because last series 480s and 470s clearly are not.
2. Make sure you pick better examples than the Bulldozer LOL fail
and finally 3. Make sure you read clearer, do your own research and know who you're talking to and what their expierence is before you open your crap trap and give advice to anyone to do homework because mine is way done and you've been taken to school :) lol :nutkick: :)
vliw has the edge in graphical tasks, Fermi has more compute capabilities, AMD. This round improved compute while keeping the edge in graphical tasks, that's what and does, as for copying, um NVIDIA was all about big cores with hot clocks, now I can say the same thing that NVIDIA copied and by dropping not clocks and fitting more cores but I'm not gonna get down to that level, idk what on earth males GCN anything like Fermi lmao,like its a freaking 2000core chip, NVIDIA never built an architecture like that until now sorta. As for dx9 one game doesn't tell it all, much more factors might be the cause but it doesn't matter really so I don't think I need to bring this point
As for bulldozer well AMD had yield issues which were apperant with llano too, NVIDIA had that issue with fermi, it almost always happens when moving to a new node, bulldozer is a revolutionary core in theory, AMD. Just failed to deliver this time around same way fermi failed with gtx400, this is y I mentioned it, now if ur opinion sais otherwise then good for u, cuz I simply happen to disagree as I've done enough research on the matter
Nvidia came out with a idea and design that was late, hot and badly implemented, they refined it and won.
AMD was building the same for many years, a GPU doesn't happen in a few months time, they saw the writing on the wall that computing was becoming more generalized and with the purchase of ATI they managed to get a very very good foothold in this area.
Nvidia is countering their early inefficiency with experience brought from their mobile departments. A smaller more efficient cip that uses tech brought from more power efficient designed Tegra chips.
Bulldozer belongs nowhere in this discussion. Back on topic.
the only time nvidia's design "lost" was Fermi vs the 5xxx series, but then they still had performance crown, just too late to market. That's one design out of 5.
Where are you getting you info that this is tegra tech?
Actually 2.3333 months is hardly enought time to rest on. Kepler is presumably 3 days away. If Kepler owns then 2 months 10 days is barely on top long enough to call a victory for the 7970. It's more like treading water till the sharks arrive! lol Especially, since it takes way longer than 2months and 13 days to produce a series which means Nvidia has been working on 700s for a while and is only "fashionably" late... A.K.A while AMD is stuffing orderves and drinking champaigne in hollow victory... Nvidia's gonna show up via VIP private entrance and F$#! the Prom queen... lol -pimpslap- :pimp:
ps - perhaps you were thinking 6970 ? :confused:
This card is listed as GK104, which means it was originally designed to replace cards like the GTX460/560/560Ti. There was information suggesting there was a GK100 and even a GK110 in development, but that disappeared. Couple this with Nvidia PR people and CEO's saying they expected more from AMD, and it would appear that Nvidia just took their intended mid-range offering, tweaked it, and relabeled it a GTX680.
Just look at the specs for the GTX680 that we know, 2GB VRAM, 256-bit memory bus, and GK104 chip. It has all the markings of what should have been a GTX660. The actual model number is second to the chip
I would appreciate some maturity and some logical reasoning if you're going to start calling me an "AMD fanboy" which in itself is just being used to slur and discredit what I have to say which is insulting and uncalled for.
Also I'm a System Admin and I have a degree in computer science, what are you doing? Don't pander to me and try to tell me what I know and who I am.
Edit: As you're a new user, I highly recommend that you read the rules.
Listen man I learned along time ago not to argue with newer members as it spins out of control fast as they are not familiar with the culture of TPU. They have "teething" issues for a lack of a better term. Relax man. Don't take it personal. I see where you are coming from with the market placement and I agree. However its all up in the air until we see some real benches from W1zz.
Edit: Just saw your a new member also. lol.