Problems? See this thread at archive.org.
That is some impressive performance from a mid-range card. I don't think we've seen that sort of jump since G70/NV47 to G80, and certainly never without a die shrink. It's no wonder nV is sitting on their high-end chips at 20nm until later this year -- otherwise they'd obliterate their GK104 sales.
I can't wait to see what the top-end Maxwells can do.
-
900Mhz P5300 => +135MHz will give us P6000
This card is just a monster, i just regret a little the 128bit bus.. BUT with high voltage memory, some OC may cover up the nasty little bus
Cloudfire likes this. -
The good thing is that GTX 860M seems to have high voltage VRAM aka desktop VRAM so you should be able to overclock that VRAM to get higher bandwidth
Plus we must remember that GTX 860M also have 8x more L2 cache than GTX 660M, which according to Anandtech should give it similar bandwidth as 192bit.
Me too. We saw around 80% better performance with GTX 680M vs 580M. This GTX 860M looks to have around 120% better performance than 660M. In benchmarks, hopefully the same in games too
sasuke256 likes this. -
-
-
GTX 750 Ti: 640 cores running at 1085MHz scores around 5750 in 3DMark11
GTX 860M: 640 cores running at ?MHz scores around 5150 in 3DMark11.
It score less so I think its running at slower clocks than 750 Ti. I think its more like 900MHz to score 5150.
If it ran at 500MHz it would score way lower than 5150 in 3DMark11.
Do you mean the chip? Its a full GM107 without any parts cut (Same ROPs, bandwidth, cores etc).
Thanks. Its all over internet now. ooops, Nvidia is probably not happy
BlackSabs likes this. -
You guys are comparing the 860M to the 660M? Why?
Compare it to the 760M or the 765M at least.
The increase is another 25% over the 765M 3DMark 11 GPU score (4100ish). -
they are comparing 1st gen maxwell vs 1st gen Kepler
Cloudfire likes this. -
I agree with the two above.
GTX 660M was the first Kepler. GTX 860M is the first Maxwell.
We can compare GTX 760M with GTX 960M when it arrives.
Isnt that impressive? A 45W Maxwell beating a 70W Kepler by 20-30%?
I bet AORUS feel silly now...hailgod likes this. -
-
-
-
Because
A) GTX 770M/760M is midrange GK106, GTX 860M is low end GM107. GTX 660M is low end Kepler GK107.
B) GTX 770M TDP of 80W. GTX 760M 65-70W. GTX 860M TDP 45W. GTX 660M 45W.
We compared GTX 580M with GTX 680M because both had 100W TDP and both were high end.
I don`t know about the rest, but I compared GTX 680M and GTX 780M to figure out how much faster it was
But go ahead, nobody is stopping you from comparing 860M and GTX 760M. I`m more interested to see how much more efficient Maxwell is, and that is much easier when TDP is the same and they are both low
end. In the end you find same TDP GPUs in the same notebook size too. You may not find a 70W GTX 760M inside a superthin Asus notebook, but you could find both 45W GTX 660M and 45W GTX 860M there. ie replacement.
GT 840M - GM107
GT 845M - GM107
GTX 850M - GM107
GTX 860M - GM107
Like this time will be different
Low end is low end Meaker. Just because they couldnt build faster low end Kepler at that time, its still a highly valid comparison.
GTX 765M SLI score 8000 in 3DMark11. GTX 680M is at 6000.
If you overclock that GTX 860M to 1200MHz, you will actually score the same as a GTX 680M. Imagine that, low end Maxwell can reach high end Kepler with some overclocking
hailgod likes this. -
If this is true, think what a 100W Maxwell chip will perform like - Even the 75-80W Maxwell should crush the GTX 780M.
Cloudfire likes this.
-
I think HTWingNut gets it right.
TDP will probably be the limitation regarding what notebooks end up taking these GPUs and what characteristics they have (general size and thinness, for example). Maybe we'll see a healthy boost in graphics performance all across the board.
Cloudfire likes this. -
-
-
-
-
-
Anything new on the 840M/845M and their performance comparisons with a 660M/740M/750M/755M? Still wondering if I should be waiting for that 13-14 inch laptop with one of those (840/5) cards
ThePerfectStorm likes this. -
You realize that you can just dump the power settings from the VBIOS instead of doing all these calculations, right?
Boom, power target range. Nominal (i.e. stock) TDP is 60000mW = 60W. Go find a 765m vbios and check it yourself if you don't believe me. Needless to say, it is very likely that the 760m has a lower TDP.
But go ahead, ignore the differences in binning and board power and keep making up numbers.
"The cores are only running 65MHz lower than GTX 650 Ti. Same amount of cores.
The GTX 660M got a 30% reduction in TDP compared to the desktop version. It was 115Mhz lower, not 65Mhz lower.
But to make it easier, say it get a 30% reduction here as well."
What you're doing is pretty much the reason why nobody pays attention to wccftech and similar sites. Because they just 'logic' some random numbers out of previously known facts and post things that seem like they might be true.
I'm not even going to go into the "People, you need to be more critical about what you guys read." statement, because that is just the definition of irony.
Anyway, you were right about one thing -- TPU does have misleading TDP numbers, since they listed the 765m with a higher TDP than it should be.Karamazovmm and octiceps like this. -
-
-
Actually I am spot on with my TDP information.
First up, GTX 770M.
You will notice that it have a power target of max 80W. Unlike your 765M vbios dump, it only list "100%".
Here is a GTX 765M Vbios from a Clevo card. Notice it goes all the way up to 80W? Unlike the Dell dump you posted, this have a power cap at 80W. I think one should mention that we are talking POWER here and not TDP.
Now to fully discredit your rant, here is GTX 760M.
According to your logic, based on this screenshot, GTX 760M have a TDP of 43W.
GTX 660M: 384 cores @ 719MHz = 45W
GTX 760M: 768 cores @ 657MHz = 43W
Whatever you are smoking, you should mail me some because obviously you are tripping bigtime.
Your magical binning and mature 28nm process is one thing, which you even cant back up with anything, but DOUBLING the cores and going down in TDP with the exact same architecture is just funny.
I however think that the vbios dumps that shows 43W as 100% but also list 100W as "233%", means it may allow the card to draw up to 100W. But of course in the end the clocks control the TDP, so the real TDP is somewhere between 43W and 100W. I may be in deep water here, but that is what I get from all of this.
For good measure, here is GTX 750 Ti. It does NOT have a TDP of 70W. It is 60W.
tl;dr:
GTX 770M: Still TDP around 80W
GTX 760M: Way higher than 43W. lol
GTX 765M: Somewhere between 60-100W, more toward 80W like the 765M I posted. Say 70W.
Wether you like it or not, my assumptions based on calculations of the desktop cards may be spot on.ThePerfectStorm and hailgod like this. -
-
-
-
You seem to have a couple misconceptions, so I'll try to clear them up:
1. A Nvidia GPU will not draw more power than the power limit (not power target, my mistake for confusing the two). From the Nvidia SMI documentation:
The min/default/max definitions correspond to the same labels as in the Kepler BIOS Tweaker. Note that much of the API is not exposed to consumer GPUs (I checked with my 770m), but the Kepler BIOS Tweaker parses through the vbios dump so it doesn't need to be.
As you can see, the max power limit is something that has to be applied by the user and is not something that the GPU will reach on its own. Also, if you check against GPUs with known TDPs (e.g. GTX780ti, GTX770, etc.), you'll see that the TDPs match up almost perfectly with the default power limits. So kudos on finding a 760m vbios, the TDP of that card (or that particular variant) is 43W give or take a few watts. Maybe it doesn't make any intuitive sense to you, but the card will not go past the power limit, as per the Nvidia documentation.
Same goes for the 765m. I suspect that the "Clevo" (actually a Dell MXM card) 765m that you found is 80W because it's MXM instead of being soldered like the 60W ASUS, MSI and non-MXM Dell models. But I'll concede that I jumped the gun in concluding that the 765m is 60W.
Also, the Kepler BIOS Tweaker probably doesn't read the Maxwell 750ti correctly because, well, it's the Kepler BIOS Tweaker.
2. I can very well back up the "magical binning and mature 28nm process". But don't listen to me, listen to Nvidia. From Anandtech:
There you go. Maxwell is on a better optimized 28nm process than the first Kepler chips. Not that this is anything unusual-- Intel regularly releases higher clocked versions of their chips on the same node as their process gets better over time.
Don't get me wrong, I appreciate all the news that you post. But sometimes your analysis just makes no sense to me.
New details about Nvidia`s Maxwell
Discussion in 'Gaming (Software and Graphics Cards)' started by Cloudfire, Feb 12, 2014.
