Nvidia BigK GK110 Kepler Speculation Thread

The problem is AMD will inevitably be forced to follow suit so long as Nvidia is gaining this kind of advantage in benchmarks.
 
In my case boost 2.0 would give me higher clocks. My card doesn't go over 58 degrees and its dead silent. Yet nVidia's drivers limit clocks and voltages based on TDP. Since I don't have a temperature or noise problem I would much rather they allow higher clocks till temperatures are closer to 70/80 degrees.

Variability on 2.0 is certainly higher but you probably get higher clocks overall.

its benefit specially if you use H2o or other ( its the same with the PTE of AMD who allow faster clock and higher voltage if you was hit your limit as it take too temperature prediction for allow more TDP, but they dont forcibly play on temperature for limit the "clockspeed" as the temperature limit is the same as before ( so as high of 100+ °C ) ( basically they have not change the limit vs what it was before ) .

Now you understand why the control on temperature limit have been added on the overclocking tool, ( 670 have not one, and the limit was 70°C, after this the core speed start to go down: solution 1) you push the fan, "solution 2 ) you put a better cooler on it ), now you can too increase the temperature limit, but then you will quickly encounter the tdp limit this time. ( +6% )... Boost 2.0 is just using what AMD was using allready with PTE.. but on a different way,..
 
Last edited by a moderator:
Mmm for the GTX 600 this 70 °C limit you talk about is not the same thing. There are some unclear extra conditions to reach the last bins but this is not the same thing as a GPU Boost temperature target.

Still I don't think such a thing as Boost 2.0 exists. On the GTX 600 there seems to be a similar temperature target to the GTX Titan one. On the GTX 680 I can spot it at 95 °C. For some reasons, on the driver side, Nvidia decided to hide this temperature target parameter to the OC tools for the GTX 600 but not for the GTX Titan.

Boost targets :
GTX 680 : 170W / 95 °C
GTX Titan : 235W / 80 °C

The difference with Titan is the 80 °C target kicks in first. It's mostly impossible to hit the power target without tweaking the parameters. It means that if you play with those to increase the temperature target or the fan speed you'll find quite some room on the power target side.
 
Ther boost target for 680 is 195W , not 170W ( this is for the 670 ) .. temp was 95°C for the 680 and 70°C for the 670 ( reference ofc,as you can still find vbios for flash them for go further )

I was not saying it is the same thing, but the temperature on 670 was allready a limiting factor, more of the tdp itself.. ( i think Nvidia have add the exact same thing of have put AMD since the 7970GHZ, the PTE .. but they have choose to use it as a limiter factor instead of as AMD a tool for allow more clock speed and voltage... ). By using this limit, Nvidia limit too the TDP on usage, but if you increase the temperature limit you hit then the TDP limit depending ofc the Asic quality of your chip.

Like i was expect price here is out of reason... 1050CHF so 1178$ ... ( cards are not available too, 1 - 2 weeks minimum for delivery, but price will not go down then.. )
 
Last edited by a moderator:
Damn , that game is heavily CPU bound !

I just did a run on Crysis 2 and it looks like the CPU pressure is present there too so it's just a Cryengine 3 thing. At minimum settings it's hitting 70-75% cpu usage on a Q9550 @ 3.5Ghz. At maximum settings it increases slightly to 80-85%.

In both cases GPU usage on a GTX 680 @ 1228Mhz is relatively low at 50-60%.
 
Ther boost target for 680 is 195W , not 170W ( this is for the 670 ) .. temp was 95°C for the 680 and 70°C for the 670 ( reference ofc,as you can still find vbios for flash them for go further )

I was not saying it is the same thing, but the temperature on 670 was allready a limiting factor, more of the tdp itself.. ( i think Nvidia have add the exact same thing of have put AMD since the 7970GHZ, the PTE .. but they have choose to use it as a limiter factor instead of as AMD a tool for allow more clock speed and voltage... ). By using this limit, Nvidia limit too the TDP on usage, but if you increase the temperature limit you hit then the TDP limit depending ofc the Asic quality of your chip.

Like i was expect price here is out of reason... 1050CHF so 1178$ ... ( cards are not available too, 1 - 2 weeks minimum for delivery, but price will not go down then.. )

No, the boost target is 170W for the GTX 680, 195W is the TDP (= under 170W the GPU can Boost, but it won't go lower than the base clock till it reaches 195W). For the GTX 670 the boost target is 140W and the TDP is 170W.

Whatever the asic quality is, at a given temperature / fan speed, the power consumption will be about the same. A "better" asic will reach a higher clock for the same power/temperature/fan speed.
 
I didn't like boost 1.0 (don't like AMD following suit) and boost 2.0 just seems to be even more of a train wreck.

While I don't like boost myself, it's mainly because it more than doubles the amount of work needed for a non-misleading review. Boost in and by itself is fine, because it can provide extra performance headroom. The only problem I have is that most reviews and methods will be showing peak performance.

I don't think that Nvidia deliberately engineered boost to artificially inflate benchmark scores, but here reviewers need to take up responsibility and put the necessary extra amount of work into their articles in order to not mislead their customers.
 
Last edited by a moderator:
http://www.anandtech.com/show/6774/nvidias-geforce-gtx-titan-part-2-titans-performance-unveiled/4
as part of our 2013 GPU benchmark suite we put together a larger number of compute benchmarks to try to cover real world usage, including the old standards of gaming usage (Civilization V) and ray tracing (LuxMark), along with several new tests. Unfortunately that got cut short when we discovered that OpenCL support is currently broken in the press drivers, which prevents us from using several of our tests.


How convenient.... :rolleyes:
That launch practice could also work for games - if there is a benchmark game you know you perform poorly in, well, just break the drivers, you can always "react fast" and fix them again a week later :)
 
It is not really broken. Some things like the popular Luxmark crash. But OpenCL per se works - in a basic fashion at least.
 
While I don't like boost myself, it's mainly because it more than doubles the amount of work needed for a non-misleading review. Boost in and by itself is fine, because it can provide extra performance headroom. The only problem I have is that most reviews and methods will be showing peak performance.

I don't think that Nvidia deliberately engineered boost to artificially inflate benchmark scores, but here reviewers need to take up responsibility and put the necessary extra amount of work into their articles in order to not mislead their customers.

I think this extra work puts too much emphasis on stock settings. The 80c temperature ceiling can be raised. If this wasn't available the throttling would probably happen at a higher temp. Now with the stock settings it's really tightly controlled, but doesn't really translate well to real life usage imo, except maybe allowing this card to exist in smaller cases and less forgiving environments with these strict throttling protocols. Focusing too much on what it can do at 80c is nearly pointless imo.
 
IMHO what you need as a baseline is a stock comparison. Fine-tuned settings can only be an extra - if not for other reason just for pure matters of sample size =1.

Nvidia choose a temp target of 80°C for power, noise and longevity reasons.
 
IMHO what you need as a baseline is a stock comparison. Fine-tuned settings can only be an extra - if not for other reason just for pure matters of sample size =1.

Nvidia choose a temp target of 80°C for power, noise and longevity reasons.

I think nVidia is conservative with those setttings and they are taking into account all the possible scenarios where these cards will be placed, however I think that when reviewing a product, a case by case methods can be applied. Running these cards throttled does not serve many of the enthusiasts who are buying these. I mean it's great to know what happens at stock settings, but you should also put emphasis on how to avoid the throttling. Changing the settings is not exactly rocket science and the majority of the target audience should be capable of doing so.

reviewing a family car or a sports car should be different.
 
I think nVidia is conservative with those setttings and they are taking into account all the possible scenarios where these cards will be placed, however I think that when reviewing a product, a case by case methods can be applied. Running these cards throttled does not serve many of the enthusiasts who are buying these. I mean it's great to know what happens at stock settings, but you should also put emphasis on how to avoid the throttling. Changing the settings is not exactly rocket science and the majority of the target audience should be capable of doing so.

reviewing a family car or a sports car should be different.

Evidently Nvidia feels there is a significant risk with having the temperature of GK110 higher than 80 degree's for a prolonged basis for enough chips that make it through qualification for it to be at that point. While it is certainly possible to increase that, it's always at user risk.

It doesn't matter what the environment is. 80 degrees on the GPU is 80 degrees on the GPU.

If the card fails after 3-4 months when someone raises the ceiling to 95 or 100 degrees or whatever, 1000 USD down the drain is going to hurt a lot more than 500 USD down the drain on a cheaper card.

Sure, show some results for higher overclocks, but stock settings should always be the focus.

Regards,
SB
 
I think that when reviewing a product, a case by case methods can be applied. Running these cards throttled does not serve many of the enthusiasts who are buying these. I mean it's great to know what happens at stock settings, but you should also put emphasis on how to avoid the throttling. Changing the settings is not exactly rocket science and the majority of the target audience should be capable of doing so.

Correct, but yet most reviewers did not seem to pay attention to the matter.
 
Evidently Nvidia feels there is a significant risk with having the temperature of GK110 higher than 80 degree's for a prolonged basis for enough chips that make it through qualification for it to be at that point. While it is certainly possible to increase that, it's always at user risk.

It doesn't matter what the environment is. 80 degrees on the GPU is 80 degrees on the GPU.

If the card fails after 3-4 months when someone raises the ceiling to 95 or 100 degrees or whatever, 1000 USD down the drain is going to hurt a lot more than 500 USD down the drain on a cheaper card.

Sure, show some results for higher overclocks, but stock settings should always be the focus.

Regards,
SB

I doubt there is a high risk in a typical usage scenario if you raise the limit to say 85-90c, if you run it 24/7 then perhaps. Raising the temp ceiling or even overclocking with software doesn't void the warranty of these cards.

Temperature limit will make it easy to use this in small cases though.

Correct, but yet most reviewers did not seem to pay attention to the matter.

Yes it definitely is important to make that point known. More info is always welcome, but also provide results without throttling and information on how to avoid it.
 
Yes it definitely is important to make that point known. More info is always welcome, but also provide results without throttling and information on how to avoid it.

We did. That's why I wrote that it doubles the amount of work for a reviewer when doing it properly:

• Free Boost @open test bench
• OC-Setting (+100 MHz GPU-Offset, 85°C temp-target, 105% power target, unlinked)
• Fixed Boost @876 MHz (nv-semi-guaranteed frequency)
• Individual clock rates per game & resolution forced after doing real gaming sessions of at least 30 minutes with no loading pauses etc. at a constant air-temperature of 28°C in front of the blower fan.

You see, I did not get much sleep since saturday.
 
So low TDP ceiling is nonsense. According to TPU, 208 W on average, and 238 W at peak. Add 6% on top of this, you will end around the 250 W mark. Should have been around 290-295 W.
Water cooling should have been stock because I doubt anyone sane who spends 1000 $ will keep it on air.
Also, this ridiculous price should be around 650- 700 $ max with water.

Of course environment matters- 80 degrees is 80 degrees but under most circumstances you will never reach it.

If the card fails after 3-4 months when someone raises the ceiling to 95 or 100 degrees or whatever, 1000 USD down the drain is going to hurt a lot more than 500 USD down the drain on a cheaper card.

I doubt anyone will allow this.
 
We did. That's why I wrote that it doubles the amount of work for a reviewer when doing it properly:

• Free Boost @open test bench
• OC-Setting (+100 MHz GPU-Offset, 85°C temp-target, 105% power target, unlinked)
• Fixed Boost @876 MHz (nv-semi-guaranteed frequency)
• Individual clock rates per game & resolution forced after doing real gaming sessions of at least 30 minutes with no loading pauses etc. at a constant air-temperature of 28°C in front of the blower fan.

You see, I did not get much sleep since saturday.

Yeah I was saying it in general, not targeted to you. It seems you have done very good work.
 
I think nVidia is conservative with those setttings and they are taking into account all the possible scenarios where these cards will be placed, however I think that when reviewing a product, a case by case methods can be applied. Running these cards throttled does not serve many of the enthusiasts who are buying these. I mean it's great to know what happens at stock settings, but you should also put emphasis on how to avoid the throttling. Changing the settings is not exactly rocket science and the majority of the target audience should be capable of doing so.

reviewing a family car or a sports car should be different.

It's not that simple, because if you lift that ceiling, not only are you operating what essentially becomes an overclocked card, but you also change power consumption, temperature (obviously), noise and naturally, performance.

Basically, every metric is altered. You can do that, but you also have to measure the card at 80°C to give a sense of what buyers can expect at stock settings. I mean, at $1000, not everyone is going to want to have to bother fiddling with settings they might not even fully understand.
 
Back
Top