Kepler news and discussion

[RussK1]

Yes, what everyone fail to remember, not only can the GTX 680 be in the Gaming Gefore market but also professionals are trying it. Games use Direct Compute coding and such other features. the 680 i have seen do good and bad at Direct Compute Benches but that doesn't handycap it from being a great card

I'm one of those people that enjoys GPGPU performance. I considered the GTX 680 for my 2500K machine but the GPGPU performance is so dismal I dropped it from consideration.

I game and work with some picture and video stuff, and a professional GPU (Tesla, Quatro, ATI Firepro) are out of the question not only because they are so expensive but they simply aren't gamers so-to-speak.

Take the HD7970 and the GTX680... both cards gaming performance and pricing differences are negligible, but when you compare GPGPU performance the 7970 unequivocally beats it. People can dismiss this fact by lamenting - "it's a gamer card(!) yada, yada, yada..." this is no excuse for nVidia dropping the ball on single and double precision floating point performance.

 

[Rockdpm]

well i don't know if that would fall into the categorie but NVIDIA trying to focus on gamers for the GTX 680 could be a hint....they are going to make seperate tech for prefessionals

 

[Anik8]

Yes, what everyone fail to remember, not only can the GTX 680 be in the Gaming Gefore market but also professionals are trying it. Games use Direct Compute coding and such other features. the 680 i have seen do good and bad at Direct Compute Benches but that doesn't handycap it from being a great card

I'm having a similar opinion on this.I agree that the DP capability of the gtx 680 has been nerfed quite badly to reduce the power draw.As a result,an epic fail in OpenCL apps and bad news for those who do photoshop intensively.
But if one follows it closely,the gtx 680 gives the best performance as of yet when it comes to Direct Compute shading performance.So no matter,how much computational tasks a game throws at the 680,it'll easily rip through.
Moreover,I'm one of those optimistic people who thinks the price will be slashed by quite an extent in 3 months from now,so that people who do photoshop and boinc/community grid can have their desired BIG K @ the same $500 price tag. 😉

 

[I am a PC]

Yes, what everyone fail to remember, not only can the GTX 680 be in the Gaming Gefore market but also professionals are trying it. Games use Direct Compute coding and such other features. the 680 i have seen do good and bad at Direct Compute Benches but that doesn't handycap it from being a great card

GTX 680 has by a far far margin the lowest compute performance out of all the high end cards.

 

[Rockdpm]

GTX 680 has by a far far margin the lowest compute performance out of all the high end cards.

Like i said, i have seen it do Good and Bad. I didnt say its the greatest Compute card on the planet. But its not the worse either....

 

[I am a PC]

Like i said, i have seen it do Good and Bad. I didnt say its the greatest Compute card on the planet. But its not the worse either....

Like I say it's the lowest out of all the high end cards but good thing it does not count for gaming no matter how much people think as proven in the benchmarks.

 

[esrever]

Guru3d has their GTX 680 overclocking guide up.... 1300 mhz.
http://www.guru3d.com/article/geforce-gtx-680-overclock-guide/
Quote:
"It will be interesting to see custom cooled GeForce GTX 680 cards as I have no doubt a stable 1400 MHz is achievable on air cooling as well alongside a tiny bit of extra voltage."

I keep hearing that the 7970 can reach those speeds, but honestly I haven't seen it anywhere yet. For Guru3d, the best speed they got on the 7970 was the custom Asus Direct CU II at 1250 mhz.
http://www.guru3d.com/article/asus-radeon-hd-7970-directcu-ii-review/23

1250mhz on the 7970 is a 35% overclock

1300 on a 680 is a +250 offset which is a 25% overclock

here the kicker, OC scailing on the 680 only happens when there is headroom and also could never happen at all durring games. The scaling of an OC'ed 680 is all over the place, sometimes getting only a 3% increase in performance with a 15% OC. The 7970 gets almost 10% for a 15% OC most of the time in almost every game.

the reviewed 680 have been running at 1100-1200mhz at stock with the turbo and the offset sometimes doesn't do much. Meaning that the 1300mhz OC would only be about a 10% OC if it even hits it during gaming.

Tom's 680 OC review is one of the only ones where they used a consumer card instead of the cards provided by nvidia for review which shows the OC doesn't scale nearly as well as the 7970 or the review samples. Judging by that, if you go out to buy a 680 and a 7970 and OC both, the 7970 might be the faster card.

 

[Rockdpm]

Like I say it's the lowest out of all the high end cards but good thing it does not count for gaming no matter how much people think as proven in the benchmarks.

So?, it still would kick anything else in the arse that wasnt benched mark. but yes gaming wise it rips through games better than the cards benchmarks. making the matter irrelevant

 

[I am a PC]

So?, it still would kick anything else in the arse that wasnt benched mark. but yes gaming wise it rips through games better than the cards benchmarks. making the matter irrelevant

What I am saying and I am not bashing GTX 680 I am just saying people get all caught up into specs that mean nothing YES fact the GTX 680 compute performance sucks ass but in gaming which is the purpose of the GTX 680 it does excellent and thats what matters not business class professional compute performance LOL.

 

[Rockdpm]

What I am saying and I am not bashing GTX 680 I am just saying people get all caught up into specs that mean nothing YES fact the GTX 680 compute performance sucks ass but in gaming which is the purpose of the GTX 680 it does excellent and thats what matters not business class professional compute performance LOL.

Exactly what i was saying before.....

 

[-Fran-]

1250mhz on the 7970 is a 35% overclock

1300 on a 680 is a +250 offset which is a 25% overclock

here the kicker, OC scailing on the 680 only happens when there is headroom and also could never happen at all durring games. The scaling of an OC'ed 680 is all over the place, sometimes getting only a 3% increase in performance with a 15% OC. The 7970 gets almost 10% for a 15% OC most of the time in almost every game.

the reviewed 680 have been running at 1100-1200mhz at stock with the turbo and the offset sometimes doesn't do much. Meaning that the 1300mhz OC would only be about a 10% OC if it even hits it during gaming.

Tom's 680 OC review is one of the only ones where they used a consumer card instead of the cards provided by nvidia for review which shows the OC doesn't scale nearly as well as the 7970 or the review samples. Judging by that, if you go out to buy a 680 and a 7970 and OC both, the 7970 might be the faster card.

That caught my eye as well... And to add a lil' more to your argument, when OC'ed, both cards get so close to each other's consumption, they become pretty much a-like in P/W, with the caveat the the 7970 performs better in GPGPU (not all, but most of them).

They're both good cards... And they're both overpriced, hahaha.

Cheers!

 

[I am a PC]

Exactly what i was saying before.....

My rational if you are a professional in need of a GPU for designing, modeling ect get a professional working class GPU not a gaming entertainment one LOL. Oh but the people will say but the professional class GPUs are very expensive and a gaming GPU is relatively cheap well then I can kill two birds with one stone my answer to that is professional class GPUs are an "INVESTMENT" whereas gaming GPUs are not they will cost you money not make you money like an investment will. It's always the people that complain that are the people that really do not use or need the feature they just want the bragging rights of having it lol.

 

[bardacuda]

Has anyone heard anything (eg. rumours/"leaks") about the 670 or 660 branded products? Will they be a cut-down version of GK104? Will they be a beefy version of GK106? What will the memory capacity be?
I've seen "leaks" that say GK106 is going to have a 128-bit memory bus, but I would assume a 660/670 would have at least a 192-bit bus so if I were to guess I'd say they will be a cut-down GK104 with say 1152/1344 cores and 1536MB/1792MB memory with possibly a 224-bit bus.
This is all pure speculation obviously. Thoughts? Insights?

 

[RussK1]

Has anyone heard anything (eg. rumours/"leaks") about the 670 or 660 branded products? Will they be a cut-down version of GK104? Will they be a beefy version of GK106? What will the memory capacity be?
I've seen "leaks" that say GK106 is going to have a 128-bit memory bus, but I would assume a 660/670 would have at least a 192-bit bus so if I were to guess I'd say they will be a cut-down GK104 with say 1152/1344 cores and 1536MB/1792MB memory with possibly a 224-bit bus.
This is all pure speculation obviously. Thoughts? Insights?

GTX 660


MANUFACTURER: NVIDIA
SPECS STATUS:Rumoured
SERIES: GeForce 600
GPU MODEL:GK104
RELEASE DATE: Q1 2012
INTERFACE: PCI-E 3.0 x16
CORE CLOCK: 900 MHz
MEMORY CLOCK: 2900 (5800) MHz
MEMORY BANDWIDTH: 185 GB/sec
MAX POWER DRAW: 175 W
MEMORY SIZE: 2048 MB
MEMORY TYPE: GDDR5
MEMORY BUS TYPE: 64×4 (256 bit)
DIRECTX: 11.1
OPENGL: 4.1
PROCESS: 28 nm
FRAGMENT PIPELINES: 512
TMU: 64
ROP: 32

http://videocardz.com/nvidia/geforce-600/geforce-gtx-660

GTX 670 (gk110)

MANUFACTURER: NVIDIA
SPECS STATUS:Rumoured
SERIES: GeForce 600
GPU MODEL:GK110
RELEASE DATE: 2012-04-03
INTERFACE: PCI-E 3.0 x16
CORE CLOCK: 850 MHz
MEMORY CLOCK: 2500 (5000) MHz
MEMORY BANDWIDTH: 280 GB/sec
MAX POWER DRAW: 250 W
MEMORY SIZE:1792 MB
MEMORY TYPE: GDDR5
MEMORY BUS TYPE: 64×7 (448 bit)
DIRECTX: 11.1
OPENGL: 4.1
PROCESS: 28 nm
FRAGMENT PIPELINES: 896
TMU: 112
ROP: 56

http://videocardz.com/nvidia/geforce-600/geforce-gtx-670

GTX 670ti

MANUFACTURER: NVIDIA
SPECS STATUS:Rumoured
SERIES: GeForce 600
GPU MODEL:GK104
RELEASE DATE: 2012-04-03
INTERFACE: PCI-E 3.0 x16
CORE CLOCK: 950 MHz
MEMORY CLOCK: 2500 (5000) MHz
MEMORY BANDWIDTH: 160 GB/sec
MAX POWER DRAW: 255 W
MEMORY SIZE: 2 GB
MEMORY TYPE: GDDR5
MEMORY BUS TYPE: 64×4 (256 bit)
DIRECTX: 11.1
OPENGL: 4.1
PROCESS: 28 nm
FRAGMENT PIPELINES: 1536
TMU: 128
ROP: 32

http://videocardz.com/nvidia/geforce-600/geforce-gtx-670ti

 

[esrever]

^^well all that is basically wrong about the gk104 so I wouldn't hold any of that to be true.

 

[RussK1]

^^well all that is basically wrong about the gk104 so I wouldn't hold any of that to be true.

Have anything constructive or better?

...and if you're so enlightened why don't you share your inside knowledge...

 

[I am a PC]

Thumbs down ^ -1

 

[bardacuda]

doesn't make sense that the 670 would be GK110 and the 670ti would be GK104....also the 680 already has 256-bit/2GB memory and 32 ROPs so I can't see the 670 and 660 having exactly the same with only a slight decrease on the core and memory clocks

^^well all that is basically wrong about the gk104 so I wouldn't hold any of that to be true.

Actually the specs he posted (except maybe TMUs and fragment pipelines which I have no clue about) do all match up to GK104

 

[RussK1]

doesn't make sense that the 670 would be GK110 and the 670ti would be GK104....also the 680 already has 256-bit/2GB memory and 32 ROPs so I can't see the 670 and 660 having exactly the same with only a slight decrease on the core and memory clocks



Actually the specs he posted (except maybe TMUs and fragment pipelines which I have no clue about) do all match up to GK104

http://www.legitreviews.com/images/reviews/1881/gtx680-gpuz.jpg

They look pretty good to me, the veracity of the rumored specs for the 670ti look questionable but I wouldn't put it past nVidia to use a "dumbed down" gk110 in the 670 and 690. And use the big dog gk110 in the 685 (if rumors are real and looking back at the 560ti 448core, 570, 580 and 590) with more pipelines, TMU's, higher clocks and more memory.

 

[esrever]

clock, naming, tmus, memory clock and pipelines are all wrong...

the only things they got right was the obvious things like pci 3.0 and 32 rops.

what possible useful information can you get from that?

 

[esrever]

http://www.fudzilla.com/home/item/26459-geforce-kepler-gk110-basic-specs-leaked

 

[bardacuda]

clock, naming, tmus, memory clock and pipelines are all wrong...

the only things they got right was the obvious things like pci 3.0 and 32 rops.

what possible useful information can you get from that?

Clocks can be whatever nVidia wants to set them at. They are lower than the 680 so it could make sense. Also the memory bus width, capacity, and ROPs are also correct for GK104. Again though it wouldn't make sense to release the same product branded as a 660 or 670 with only slightly lower clock speeds, which could be easily overclocked back to 680 levels anyway, for way less money than a 680. Chances are....if they will be based on GK104 at all and not GK106....that the number of shaders, memory capacity, and bus width would all be reduced judging from how the 5xx series went.

 

[RussK1]

clock, naming, tmus, memory clock and pipelines are all wrong...

the only things they got right was the obvious things like pci 3.0 and 32 rops.

what possible useful information can you get from that?

Maybe I'm obtuse but these look pretty good~

GTX 680

MANUFACTURER: NVIDIA
SPECS STATUS:Official
SERIES: GeForce 600
GPU MODEL:GK104
RELEASE DATE: 2012-03-22
INTERFACE: PCI-E 3.0 x16
CORE CLOCK: 706 MHz
MEMORY CLOCK: 1502 MHz (6 GHz)
MEMORY BANDWIDTH: 192.3 GB/sec
MAX POWER DRAW: 195 W
MEMORY SIZE: 2GB
MEMORY TYPE: GDDR5
MEMORY BUS TYPE: 64×4 (256 bit)
DIRECTX: 11.1
OPENGL: 4.1
PROCESS: 28 nm
FRAGMENT PIPELINES: 1536
TMU: 128
ROP: 32

The only thing I see wrong is the core clock...am I missing something?

 

[bardacuda]

Since the price on the 580 has come down to ~$400 I would guess that the 670 will either be a little bit cheaper for roughly the same performance as a 580....or will be about the same price for slightly higher performance.

 

[JAYDEEJOHN]

According to Gupta, the second Kepler implementation will include a lot of capability not present in these first gaming-oriented products. In particular, it will have a lot more double-precision capability (which is not required for most graphics applications) and include new compute-specific features. And of course the raw power of these chips will be quite a bit higher than the mid-range graphics version introduced this week.

Although the company is not yet giving any of the speeds and feeds on the second Kepler, one would expect the core count and peak double precision performance to be two to three times higher, and memory bandwidth to get at least a 50 percent bump. Clock speed will almost certainly be whittled down from the current 1.3 GHz on the Tesla M2090, but perhaps not so aggressively as in these first Kepler gaming parts.

http://www.hpcwire.com/hpcwire/2012-03-22/nvidia_launches_first_kepler_gpus_at_gamers_hpc_version_waiting_in_the_wings.html?page=2
So, how big is the perf hit going to be on the big K, and how many watts will it have to pull to smash the 680?
Ive a feeling itll be close to 20% in perf, tho itll be 60% larger