mercenar1e
Member
yeah lol.. i don't think consoles will have 8GB ram until 2020 =D
im going to make a guess.. next ps and xbox will have the graphics power of todays 560ti
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Nvidia Kepler - Geforce GTX680 Thread - Now with reviews
- Thread starter artist
- Start date
yeah lol.. i don't think consoles will have 8GB ram until 2020 =D
im going to make a guess.. next ps and xbox will have the graphics power of todays 560ti
evil solrac v3.0
Member
2 if they are lucky.
specialguy
Banned
Well then they're dumb, because it's eminently possible.
Wait a bit more...bigger and better things are on the near horizon.
mercenar1e
Member
I wouldn't spend $500 on a card with only 2GB of VRAM. I still don't know why they did that. I'm sure the reviews will have very good explanations, but 1.5GB 580s were hitting a VRAM wall. At least with the 7950 and 7970 getting passed by, early adopters can talk themselves into CFX down the line
piratepwnsninja
Member
Where?!
My mistake. NCIX yields a result of a single GTX 580 when you search GTX 680, heh.
Granted a lot of this is what I've pieced together from rather spirited partisan debates, but the 680 is supposed to be more energy efficient AND win the benches
bigtroyjon
Member
You need to send it to some random blog as a rumor, then all the usual suspects will not only believe that it's possible, they will consider it an iron clad fact.
lowrider007
Licorice-flavoured booze?
Why?, are these cards not a big jump over the 500 series? I have a 570 and I'm looking for a pretty big jump in performance, although I could probably only afford a 670 rather than the 680 tbh.
MoonLightWave
Member
and im selling my gfx card =P
Damn how long have you been saving gift cards ?
mercenar1e
Member
Zombie James
Banned
Sweet, love waking up to some new hardware launch.
I'd prefer the non-reference versions myself, hopefully they wouldn't take long to get to market. Unless you're talking about a better series, which I'd be interested in. Pricing is rather stiff on this thus far if rumors are true. Not $600-ouch, but $500-530-ouchie. Wonder how long it will take for other versions and/or models.
I'm currently running the following system and wanted some general advice and opinions on what to do in regards to upgrading in the next few months
Core i7 870
8GB DDR3
2x GTX 570 1280 in SLI
HP lp3065 @ 2560x1600
I'm thinking the best thing to do would be to wait for the 690, grab one and then a few months later grab another 690 and SLI again.
Is there a rough ETA on the 690 or whatever other cards NVIDIA are bringing out?
Thanks.
Core i7 870
8GB DDR3
2x GTX 570 1280 in SLI
HP lp3065 @ 2560x1600
I'm thinking the best thing to do would be to wait for the 690, grab one and then a few months later grab another 690 and SLI again.
Is there a rough ETA on the 690 or whatever other cards NVIDIA are bringing out?
Thanks.
That is NVENC, the thing that Charlie confused for dedicated PhysX block in Kepler.
It does encoding/transcoding
18-page review, but it's TweakTown...
If there won't be a dedicated review thread, these should all be added to the OP, and perhaps, a mod edit to the title for model clarification.
NVIDIA GEFORCE GTX 680 (Kepler) 2GB Reference Card Video Card Review
http://www.tweaktown.com/reviews/46...b_reference_card_video_card_review/index.html
If there won't be a dedicated review thread, these should all be added to the OP, and perhaps, a mod edit to the title for model clarification.
NVIDIA GEFORCE GTX 680 (Kepler) 2GB Reference Card Video Card Review
http://www.tweaktown.com/reviews/46...b_reference_card_video_card_review/index.html
Zombie James
Banned
Whoa, nearly the same power consumption as the 560 Ti. Pretty clear-cut victory with the benchmarks, too. Bring on the $200 variant, I like where this is going.
Slot it into my current aging rig (i7 720, 6gb drr3, X58mobo) and wait it out until Intel Ivy Bridge rolls out so I can build a new PC around it.
lowrider007
Licorice-flavoured booze?
I think after looking at those benches I may stay with my o/c'd 570 for this gen, the minimums aren't really a massive jump and I don't really care about frame-rates beyond 60 fps.
TheGuardian
Member
amstradcpc
Member
Minimums in general ( except for Battlefield 3 ) are better for HD 7970 which i consider a better architecture with more potential and much more elegant. This plus long idle tdp advantage that also is important if your leave your rig on at nights...
This may or may not get covered in the reviews but here goes ..
How did Nvidia manage to triple core count in such a small die? They ditched the hotclocks! Yes, probably half of the 3d geek world already knows this answer (hello dr_rus) but what else did Nvidia change? For starters they increased the density of the SMs. Second, in the instruction issue pipeline, they replaced the several Fermi blocks dedicated for scheduling by a single more intelligent scheduler. Third, each core was optimized, both in terms of area and timing.
The seed of hotclocks was laid with G80. Back then, on 90nm Nvidia needed more than 1GHz clocks to achieve their performance goals. So they decided to lengthen the pipelines, a trick they learned from G70 (7800GTX) (G70 had some logic dedicated (~20M transistors) for pipeline lengthening to hit its clock). So instead they could architect the G80 with 128 cores @ 1.35GHz rather than going for 256 cores at ~700MHz. 256 cores wouldnt have been feasible on 90nm either, so hotclock was a outcome of bypassing a constraint. They continued the hotclock approach with GT200 (GTX280) and GF100/GF110 (GTX480/GTX580) simply because the process nodes could not achieve the clock targets Nvidia had in mind. However their clock target was getting closer and closer with each new node. With 28nm, they knew that even in the worst case scenario target clocks were missed, they could work on leakage, timing, wiring lengths and other aspects to almost get there.
With Kepler, Nvidia hit all their desired goals - minimize the die area by aggressively tuning each unit of the core, remove the hot clocks, invest the gained area in increasing the core count & reduce power consumption (removing hotclocks reduces power consumption).
So why 1536 and not 768? GF114 (560Ti) normalized i.e without hotclocks would be 768CC. With new architecture on a new node, the expectation is to double the raw performance, 768x2=1536.
How did Nvidia manage to triple core count in such a small die? They ditched the hotclocks! Yes, probably half of the 3d geek world already knows this answer (hello dr_rus) but what else did Nvidia change? For starters they increased the density of the SMs. Second, in the instruction issue pipeline, they replaced the several Fermi blocks dedicated for scheduling by a single more intelligent scheduler. Third, each core was optimized, both in terms of area and timing.
The seed of hotclocks was laid with G80. Back then, on 90nm Nvidia needed more than 1GHz clocks to achieve their performance goals. So they decided to lengthen the pipelines, a trick they learned from G70 (7800GTX) (G70 had some logic dedicated (~20M transistors) for pipeline lengthening to hit its clock). So instead they could architect the G80 with 128 cores @ 1.35GHz rather than going for 256 cores at ~700MHz. 256 cores wouldnt have been feasible on 90nm either, so hotclock was a outcome of bypassing a constraint. They continued the hotclock approach with GT200 (GTX280) and GF100/GF110 (GTX480/GTX580) simply because the process nodes could not achieve the clock targets Nvidia had in mind. However their clock target was getting closer and closer with each new node. With 28nm, they knew that even in the worst case scenario target clocks were missed, they could work on leakage, timing, wiring lengths and other aspects to almost get there.
With Kepler, Nvidia hit all their desired goals - minimize the die area by aggressively tuning each unit of the core, remove the hot clocks, invest the gained area in increasing the core count & reduce power consumption (removing hotclocks reduces power consumption).
So why 1536 and not 768? GF114 (560Ti) normalized i.e without hotclocks would be 768CC. With new architecture on a new node, the expectation is to double the raw performance, 768x2=1536.
brain_stew
Member
Finally, a real antialiasing solution for real time rendering. I expect many torches and pitchforks to land at Nvidia's door from those that like an over sharpened aliased mess but in time most should recognise the benefits.
LabouredSubterfuge
Member
As long as there is none of the ghosting prevalent in older temporal AA solutions I'm all for it.
BoobPhysics101
Banned
God damn. 680 is a BEAST. And with Nvidia's superior antialiasing, it's a no brainer. I want one. GTX 570, you must go.
The Dutch Slayer
Member
I have a AMD HD 5850.
I know 500 is a lot of money but man I think it would be a great upgrade.
I know 500 is a lot of money but man I think it would be a great upgrade.
DeFiBkIlLeR
Banned
When are the 660 cards(or the equivalent to this gens 560 range)being launched..?
BoobPhysics101
Banned
Why the fuck not? This thing OC'd vs my GTX 570 OC'd = 40-60% faster. Considering I got my GTX 570 for free, selling and getting this seems like a no brainer.