Kepler Nvidia GeForce GTX 780

[paulbagz]

Holy...

http://videocardz.com/41916/gigabyte...mance-unveiled

-PB

[Tha Last Meal]

Gtx 780= 700$
gtx 770= 500/550$

[Sushi Warrior]

Holy...

http://videocardz.com/41916/gigabyte...mance-unveiled

-PB

No way that is right. 15% less shaders and TMU's, 1-2% better clocks, half the memory, and only a 3% performance hit? They aren't using the same drivers (314 vs 320). I'd say it's a driver difference closing the gap.

[GoldenTiger]

No way that is right. 15% less shaders and TMU's, 1-2% better clocks, half the memory, and only a 3% performance hit? They aren't using the same drivers (314 vs 320). I'd say it's a driver difference closing the gap.

As I told you on [H ]... it's hardly unbelievable. I've posted a followup for you to explain the pieces you're missing, if you are interested in that thread.

[Sushi Warrior]

As I told you on [H ]... it's hardly unbelievable. I've posted a followup for you to explain the pieces you're missing, if you are interested in that thread.

As I told you in [H], "architecturally imbalanced" is the biggest load of bull I've heard. What I DO believe is they used DIFFERENT DRIVERS for the testing. That is HARD, SOLID FACT. 40-50mHz clock differences (5% difference at most) do not balance out with 15+% architectural differences to give results that are 3-4% off. Why on earth would Nvidia release a Titan which has "more shaders than it needs"? They have some of the most brilliant engineers in the field, who just so happened to make the mistake of giving a card "too many shaders". Put away your green flag and look at the facts.

[GoldenTiger]

As I told you in [H], "architecturally imbalanced" is the biggest load of bull I've heard. What I DO believe is they used DIFFERENT DRIVERS for the testing. That is HARD, SOLID FACT. 40-50mHz clock differences (5% difference at most) do not balance out with 15+% architectural differences to give results that are 3-4% off. Why on earth would Nvidia release a Titan which has "more shaders than it needs"? They have some of the most brilliant engineers in the field, who just so happened to make the mistake of giving a card "too many shaders". Put away your green flag and look at the facts.

So if there's a bottleneck in a working chip, what do you think it should be called? I'd say architectural imbalance is a good summation of it. For example, in most circumstances, GK104 benefits far more from a high memory OC than a high core OC. This indicates an imbalance with there not being enough memory bandwidth for those cases. It has nothing to do with a "green flag", it's just basics of how these kinds of electronics work (GPU, CPU, heck even memory simply with timings, etc.). I'm not sure where the anger is coming from...

[Sushi Warrior]

So if there's a bottleneck in a working chip, what do you think it should be called? I'd say architectural imbalance is a good summation of it. For example, in most circumstances, GK104 benefits far more from a high memory OC than a high core OC. This indicates an imbalance with there not being enough memory bandwidth for those cases. It has nothing to do with a "green flag", it's just basics of how these kinds of electronics work (GPU, CPU, heck even memory simply with timings, etc.). I'm not sure where the anger is coming from...

Sure, GK104 could have a memory bottleneck. That makes sense. Memory bottlenecks are a real thing. "Shader bottlenecks"? What on earth is a "shader" bottleneck? Do you understand how a GPU works? You give it more shaders, it does more work in parallel quicker. There are 200-something extra "cores" doing work. And that isn't like a CPU core where some apps aren't multi-threaded. GPU shaders would always get used as long as there is work to do (which, in a game, there always is). Read up on how GPU architecture works maybe? It's a parallel system. Each SMX has the hardware necessary to do both instruction fetching, decoding, scheduling and dispatch. The notion that "those shaders can't get work to do" is ridiculous at best.

So, that's my "snarkiness" and "indignant attitude" for you. I think you're just getting it confused with "knowledge", "understanding", and "observational ability", all of which you apparently lack.

[Tha Last Meal]

http://www.techpowerup.com/184311/ge...-revealed.html

[LordEC911]

Sure, GK104 could have a memory bottleneck. That makes sense. Memory bottlenecks are a real thing. "Shader bottlenecks"? What on earth is a "shader" bottleneck? Do you understand how a GPU works? You give it more shaders, it does more work in parallel quicker. There are 200-something extra "cores" doing work. And that isn't like a CPU core where some apps aren't multi-threaded. GPU shaders would always get used as long as there is work to do (which, in a game, there always is). Read up on how GPU architecture works maybe? It's a parallel system. Each SMX has the hardware necessary to do both instruction fetching, decoding, scheduling and dispatch. The notion that "those shaders can't get work to do" is ridiculous at best.

So, that's my "snarkiness" and "indignant attitude" for you. I think you're just getting it confused with "knowledge", "understanding", and "observational ability", all of which you apparently lack.

Do you know how GPUs work?
To call you out, how do you explain 5870->6970?

FYI- Synthetics aren't the best way to compare the difference between Titan and GTX780. GTX780 is ~10-15% slower.

[LordEC911]

Sure, GK104 could have a memory bottleneck. That makes sense. Memory bottlenecks are a real thing. "Shader bottlenecks"? What on earth is a "shader" bottleneck? Do you understand how a GPU works? You give it more shaders, it does more work in parallel quicker. There are 200-something extra "cores" doing work. And that isn't like a CPU core where some apps aren't multi-threaded. GPU shaders would always get used as long as there is work to do (which, in a game, there always is). Read up on how GPU architecture works maybe? It's a parallel system. Each SMX has the hardware necessary to do both instruction fetching, decoding, scheduling and dispatch. The notion that "those shaders can't get work to do" is ridiculous at best.

So, that's my "snarkiness" and "indignant attitude" for you. I think you're just getting it confused with "knowledge", "understanding", and "observational ability", all of which you apparently lack.

Do you know how GPUs work?
To call you out, how do you explain 5870->6970?
Might want to read up on SMXs again. Scheduling is highly dependent on software.

FYI- Synthetics aren't the best way to compare the difference between Titan and GTX780. GTX780 is ~10-15% slower.

[BababooeyHTJ]

No way that is right. 15% less shaders and TMU's, 1-2% better clocks, half the memory, and only a 3% performance hit? They aren't using the same drivers (314 vs 320). I'd say it's a driver difference closing the gap.

Have you seen some of the 7950 vs 7970 clock for clock reviews or the 6950 before and after shader unlock comparisons or the gtx 670 and gtx 680 clock for clock comparisons? A lot of apps don't scale linearly with extra shader cores, most don't seem to.