[Exclusively for NBR] NVIDIA Kepler GK107 and Intel Ivy Bridge BENCHMARKS revealed by yknyong1!

Holy ..

Its about as fast a 460m in directx11(and that with ddr3 ram?).. what the hell kepler is really good!

 

Wow considering how the 3612QM is a slightly faster 35w 2630qm, that puts the Nvidia 640m at 20-25% faster than a stock AMD 6770m which scores around 1550 with a 2630qm according to notebookcheck and other users. It beats a 555m by around 45% (which scores ~1300) and even surpasses a stock 460m, source 2), which scores around 1800 (both were paired with a 2630qm). That's pretty damn good performance.

Though of course at higher resolutions the memory bandwidth advantage of the two 6770m and 460m will likely allow them to pull ahead. I'm guessing that The 640m will be 15-20% better than the 6770m and roughly the same the 460m at 900p due to the 6770m and 460m having double the memory bandwidth (50/60gbps respectively vs 28.8gbps).

Then again, at 28nm the 640m should have significantly better overclocking capability than the 6770m/460m so the full potential of the 640m may be a bit higher than the benchmarks actually show.

Kepler is actually shaping up to be pretty amazing. If the V3 costs as much as the 3830tg did (~$750), then it will be a very good little budget gaming laptop. I can't wait to see what HP puts out in the DV6 models (650m or 660m?) which usually have very good price/performance. Very exciting news, thanks yknyong1.

 

We will have to see about it, overclocking with kepler could be difficult or even impossible due to the new dynamic clocking which nvidia uses.. basicly it overclocks the GPU in intensive situations just like the sandy bridge or ivy bridge cpus do with turbo boost

 

True. Though I don't really see how turbo boost would work for gpus. Gpu intense programs will leverage all the cudas cores... so I dunno how the gpu "turbo boost" would be any different than the dynamic gpu clocks that exist already.

I really hope nvidia doesn't lock the gpus... that would be a lot of potential wasted.

 

why erase the IE address ? --

 

Delete that address off your post! There is a valid reason for that as per PM!

 

Fritz got 21.83...
It is as high as 2720qm.

 

NVIDIA GeForce GT 555M video card benchmark result - Intel Core i7-2630QM Processor,Alienware M14xR1 score: P1898 3DMarks


GT 555m 192bit overclocking to 825/1050 got the same score

 

I presume this is the GDDR5 version.

There is a huge reason why NVIDIA rebranded so many chips (especially those below 640M).

 

No its not, the m14x never had a GDDR5 version!! theres only a ddr3 variant

It is though the strongest 555m variant of all other variants.. It has a 192-bit bus and 144 cuda cores, and clocks relatively well (overclocks)

 

True. I wonder what happens to the GTX 660M. We do know that GT 650M is the exact same spec as the GT 640M.

 

bottleneck is the memory bandwidth

gtx 660M may be gk107 with gddr5 verion, at least 50% performance over gt 640M/650M

 

Yes, if the weakest Kepler is already at 460m level, how strong must the 660m be then (if its kepler)

The 660m would be probably stronger than the 570m then which doesnt make sense

So most likely only the GK 107 chips are avaible at this time (the weakest Kepler entry chips) But imagine them running GDDR5 with a higher bandwith, they would perfom really well

And the 660m is probably just a 560m rebrand (if the 670m and 675m are really rebrands of 570m and 580m, this would be very likely)

So it probably looks like the two variants of this article are the 640m and the stronger 650m?

GT 640M = GK107-200
GTX 650M = GK107-300

 

That was why I said NVIDIA had a reason to rebrand so many chips. The GT 640M is a good example of brute-forcing a graphics chip to run at an intended level.

 

To be honest the GK107 looks really really good, even while using DDR3 VRAM, im quite astonished!

But I have a few concerns which only time will tell how they turn out

1. While the idle temperatures are really good since the gpu is downclocked, what if the turbo boost from the Kepler chips is maintained over a longer time, what kind of heat does it generate?

2. Does it throttle, like the Sandy Bridge or Ivy Bridge (yes they too) CPUs?

3. How well does the NVIDIA Turbo Boost work with the Ivy Bridge Turbo boost together

Just to make this clear, throttling is a safety mechanism, but if a laptop, like the famous Acer TimelineX 3830TG has a bunch of power, but not the cooling to support this, the machines reaches the CPUs or this time even the GPUs throttling limit, and the CPU or GPU downclocks, which kills overall perfomance (after a certain time that is)..

28nm enables more transistor on less die space, the positive side is, higher perfomance at less voltage.. But if someone bruteforces the chip for the most power, it creates even more heat than a 40nm chip, since there are more transistors..

Before Sandy Bridge, throttling was not such a great concern for laptops.. Its a good example of gaining more power for the cost of heat(which in a Desktop world is acceptable, since adequate cooling is not a big problem, but cooling solutions for laptops havent changed and with laptops such as ultrabooks they get even worse)

 

The Acer V3 didn't seem to have throttled during the tests. I think the Kepler's hot clocks are only up to TDP.

 

The benchmark was only run for a short time, what I mean is extensive gaming over longer periods of time, like 1 hour Battlefield 3, which is a major difference to running a few benchmarks!

If you want to test throttling you have to run prime 95 (CPU Test) and furmark (GPU Stress Test) at the same time for atleast 10 minutes, to see if any throttling would occur (simulated) and also check temperatures

If the Acer V3 would have already throttled in the benchmark, the CPU/GPU would be no use at all

Also: Throttling is not the end of the world, atleast with the Intel CPUs you have the option to turn off turbo boost and even limit it in the different P-States

In the 3dmark06 benchmark youve posted you also recorded the temperatures, and if you watch the IVB cores you see they max out around 85 degrees and this only after running one (and a weak one that is) benchmark! Sadly you havent recorded the GPU temperature (because of optimus) you shouldve opened the HWINFO app after starting the benchmark..

 

There is no way I can do that testing. Not in that limited time anyway.

 

Went back today, tons of people there... I'm amazed you can get the laptop for the period of time you need to run the benchmarks

 

Thanks yknyong

At first glance it looks like the GT 640M perform like a GTX 460M. At 720p and 900p this GPU is going to be a beast. At 1080p it will suffer bad. So there it will perform much less than a GTX 460M.

If I should guess the difference between a 640M and 650M, I would say it can turbo boost higher than the 640M. I am guessing that the 650M is only going to get DDR3 like the 640M too, like 540M and 550M had. It will support GDDR5, but won`t be featured. The GDDR5 memory is saved for the lower end GTX version.

Now if only that GTX 660M could get a 192 bit bus.

 

I get the feel that NVIDIA has been holding back performance somehow. If that GK107 gets GDDR5, it will simply outclass the GTX570M. Given higher clocks it is the potential candidate for GTX 680M. Is it because AMD's GCN mobile is expected not to be as good?

 

Is the bus defined by the chip or the board?
Just for funs sake:

How would this theoretically perform reballed to a GTX580m board (2GB DDR5)?
I am looking at you DXG!

EDIT: Yes and you Saltius

 

By chip,
GK107 = 128bit bus max.(It could be cut down to 64bit but there seems no reason to do that)

 

if this GF107's 3dmark score are for sure ?

gt 640M only got half the gpu score?

 

First of all this was for the desktop GK 107 and more important this was proven as a fake because the driver that was used couldnt have any support for the GK107!! Its fake!! It turned out to be a 460m or something like that not sure

I guess its because of bad yields.. That would make the most sense, seems like they need the chips they have for desktop GPUs

 

3DMark programmes do not even recognise Kepler chips, so anything online with 3DMark mentioning a new name is fake for sure. In fact the GPU field is left blank.

 

If NVIDIA were to use Kepler in low end chips, with the 28nm production situation now, the volume required is never going to be reached. Do you agree?

 

Bad yields could mean low volume, but also a bad outcome of GPU chips.. Normally there is a top end chip like the 2820qm for example, chip production is analog and very complex.. Now the if the 2820qm is produced there are many bad outcomes, there are sold as 2720qm and so on, with deactivated parts on them, same as the dual core, which are basically downgraded 2820qm chips

Obviously the highest outcome are the low end chips, as in this case for example the GK107 in the two variants featured in the notebooks..

 

If you want to run a stress test to see if it's throttling, then you could probably set up an autorun script on a memory stick which launches both prime95 and furmark in the system tray.

If you use a really small memory stick (one of the ones that barely sticks out), then I doubt anyone would notice. That way you could put it in, walk off and do something else for twenty minutes, then come back and look at the results.

 

Create the autorun script for me... Better still, mail me the physical drive...

 

Agent yknyong1 in action

Dont forget, next time start the benchmark and then open HWINFO or such to see GPU temps, because optimus enables the discrete GPU only when in use, else HWINFO wont detect it or detect the hd4000 only

 

if he does this, small usb trick, I will seriously believe he is an agent

 

yknyong007

 

This guessing is ridiculous, score is P1898.

 

I'm guessing you'd want to know these things:

1. Temperatures for both the GPU and CPU
2. throttlestop logs
3. logs of GPU isn't throttling back

I'll try and look into it. Also, if you know the TJMAX of the cpu in question then that would be good, to prevent us getting any false readings.

EDIT: Damnit, throttlestop needs admin privelages.

 

any leaks on the GTX 680m?

 

yknyong1 Can you run 3dmark06 on the intel IGP graphics? I want to see the power of the intel integrated graphics 4000 series! PLEASE .

 

if yknyong1 could somehow get another computer and secretly swap it so he could take it in the bathroom then he could run all the tests he wants

 

Please post the 3DMark11 GPU score. Can we end the games please?

 

P1719 graphics score
P1898 total score

 

what about 3dmark vantage?

 

Tomorrow will be a good day

 

Agreed. Kepler is shaping up to be pretty good.

 

680m might be announced tomorrow, but I'm doubting anyone will be able to test a system with this part.

 

These lesser cards, 640M, 650M, don't excite me in the least. I haven't been waiting for months on this new generation of cards for minor upgrades to 555M level performance. I'd like to see what the 660M can do, but it has a mere 128 bus and they pegged it below the 570M rebadge, which we can already get, so I'm not expecting much. Then you have the 680M, which might be great, but will also be insanely expensive no doubt.

Aggghhhhh! Please bring out the mobile Cape Verde and Pitcairn parts.

 

Been following this thread and the other ones as well. So what's happening tomorrow exactly?

 

nvidia is supposed to announce kepler

 

Paper launch of Kepler

 

Final picture


What I feel is that finally GPU performance levels above GT 555M possible in Ultrabook form factor. That alone is a good reason to celebrate if you are looking for good graphics performance in 14" and below.

 

omg giddyup time!

I don't know, I can pay 1200 for 680m going nuts about this