Test of the Real Nvidia Gtx Titan, single-sli-tri Sli, 4 WAY SLI
-
And within the same GTX 680 series, it would give better results with more VRAM, it would be comparing between a 2 GB and a 4 GB one.
In some factor it should be noticed, less FPS drop, better FPS minimums, etc... I mean it should be different in some Synthetic Test that takes advantage of the 4 GB of VRAM.
Of the Titan if it's because of the bus that is wider it can't be compared with the GTX 680 because it uses a different architecture and it would be to know the improvement or evolution between both technologies, GF104 and GF110, but for now I stick with the GF104 until I see how things are coming out for the GF110.
Let's see, both architectures are the same, what happens is that the Titans use the GK110 which is a high-end chip, and it has more of everything, starting with more bandwidth thanks to its larger bus, followed by many more Cuda Cores, 5 GPCs instead of 4, more TMUs, more ROPs, larger caches...
It has more of everything, it's a fattened up GK104 in all aspects so it's normal that it's smoother simply because it's more powerful, it doesn't have much mystery either.
I suppose that all the GFs you mention you refer to GK ¿no? ;D.
-
So, both architectures are the same, the thing is that the Titans use the GK110 which is a high-end chip, and it has more of everything, starting with more bandwidth thanks to its larger bus, followed by many more Cuda Cores, 5 GPCs instead of 4, more TMUs, more ROPs, larger caches...
It has more of everything, it's a fattened up GK104 in all aspects so it's normal that it's smoother simply because it's more powerful, there's not much mystery to it either.
I assume that all the GFs you mention you mean GK, right? ;D.
Ok, understood.
But the questions I've thrown out there have gone unanswered.
Basically, if you run the Synthetic Tests with two GTX 680s, one with 2 GB and another with 4 GB with the same overclocked clocks, which one will give you the better result?
As for whether it was the GK the mnemonic of the chip, I was hesitating between those two and it was the one I didn't put, the GK.;D
-
Ok, understood.
But my questions have not been answered.
Basically, if you pass the Synthetic Tests with two GTX 680s, one with 2 GB and the other with 4 GB with the same overclocked clocks, which one will give you the best result?
As for whether GK was the technician of the chip, I was in doubt between those two and it was the one I didn't put, the GK.;D
It will depend on whether the synthetic test you pass uses more than 2GB of VRAM or not, because if it doesn't use them I already told you that the result will be THE SAME on both, and in case there is some use of VRAM above 2GB, the 4 GB version will go from slightly to much better (depending on the amount of VRAM that the application really uses, and beware, DO NOT confuse this VRAM in use with what the graphics memory meter shows in Afterburner, not because it doesn't measure the graphics memory used by an application X, but because even if it did, it's not the same to measure the VRAM used in the purest style of a "victim cache" in graphics with a lot of VRAM to the authentic need for more VRAM to render an image.
-
By modifying the bios you can increase the fans, limit Power Target and in theory the voltage but this does not significantly improve the OC on the GTX 680 it's about 40-50mhz nothing important, because nvidia has limited the power consumption of its gpu with a chip independent of the gpu that keeps the consumption and temperatures within the limits they consider safe... without the possibility of modifying it.
-
With a standard OC, it almost matches the performance of my GTX 690 in Fire Strike.
My result with stock clocks. NVIDIA GeForce GTX 690 video card benchmark result - Intel Core i7-3930K Processor,ASUSTeK COMPUTER INC. RAMPAGE IV EXTREME score: P10550 3DMarks
As you rightly say, with some maturity, if they refine the drivers, firmware, and a good stable OC at 1250, it would reach its height. I also think that Nvidia knows very well the potential of the graphics card and "maybe they didn't want to or couldn't get everything out at once". By refining some things, they can awaken the beast.
Best regards
What I think is that they didn't want to leave their dual orphaned, if the monogpu reaches the dual, it would have to lower that dual quite a bit and as long as there is no competition, they won't do it.
It seems that they don't want to cannibalize any of their cards, although in the end, since both cost around the same, they have them as exclusive products, if the Titan cost €750 or less, the sales of 690 would drop a lot. Even now, the 690 should start to drop in price, at least until they run out of stock.
Congratulations ELP3 again among the top 10, if I remember correctly, this is the best position you've achieved or at least since I've known you, impressive!!!!
With so many Titans in your possession, is there any project for a 4K TV (as far as I understand, they are limited to 30 fps) or a monitor like the Sharp PN-K321 during this year.
Obviously their prices are high between U$S 5000 to U$S 10000 but those who have tried them say that their quality is monumental.
Salu2.
I don't think this is their best position, I believe that with the 680 or the 580, I can't remember, they came in first, in extreme and performance, although then they started to drop according to how more people entered with LN2 or better drivers
!0000$ don't look at Argentina as it is, we are all luxury to have a Titan there, immoral prices, they already are here, I don't even want to imagine what they are like there.
I've flashed them, the power limit leaves it at 120 and you can set the fan to 100%, but in synthetic tests I haven't seen improvements, it ends up more or less the same, maybe in games…., but you never know, maybe we'll see what they modify, voltage and temperature limit will be important, but for now...
If it doesn't do anything by putting in another bios, maybe Nvidia gave them well tied, so that they don't do a significant OC and they can cannibalize the 690, anyway, the Titan would have to improve more by driver than a 690, but it seems that Nvidia wants them staggered, but this thing about Boost 2 doesn't seem like any progress compared to the first boost, it seems more like a further twist to try to cap them.
Looking at this, I think these graphics cards come capped so that they don't consume and aren't noisy, surely if they have a high consumption peak, they cut voltage and frequency.
Saludos
-
I don't think this is their best position, I think with the 680 or the 580 I can't remember it came first, in extreme and performance, although then it started to go down as more people entered with LN2 or better drivers
!0000$ don't see Argentina as it is, we are all a luxury to have a Titan there, immoral prices, if they are here then I don't want to imagine there.
Regards
Hi Fjavi so if ELP3 was #1 I must have missed it.
The Titan if it arrives in Argentina will surely be for April although the graphics here at the moment have an additional price of 80-90% so they will be U$S 1900, although it's not like 4 years ago here every year our salary increases by 20-25%, so that cost ratio with Spain is no longer 3 to 1 but rather 1.5/2 to 1 but still most can't afford it.-
Actually the query for ELP3 was if he had any intention of trying the Titans at 2160p although as he referenced the TVs and monitors will be expensive and seeing the results he is getting it seems very feasible that he can move the games between 30-40 fps and as Titan is the first Nvidia graphics card with 4096x2160 resolution I got curious.
Regards.
-
Hello Fjavi, so if ELP3 was #1, I must have missed it.
The Titan, if it reaches Argentina, will surely be in April, although the graphics here at the moment have an additional price of 80-90%, so they will be U$S 1900, although it's not like 4 years ago, here every year our salary increases by 20-25%, so that cost relationship with Spain is no longer 3 to 1 but rather 1.5/2 to 1, but still most people can't afford it.
Actually, the question for ELP3 was if he had any intention of trying the Titans at 2160p, although as he referenced the TVs and monitors will be expensive, and seeing the results he's getting, it seems very feasible that he could run games at 30-40 fps, and as the Titan is the first Nvidia card with 4096x2160 resolution, I got curious.
Regards.
Well then, better, while you guys are advancing, we have Mr. Scissors, anyway, I love the Titan as a card, I think it's a great card, although it's beyond what I'm willing to pay, if they release a cheaper 3Gb one, I'll try to get at least one.
Anyway, it was said that there would be very few, but I see them everywhere, plus the price isn't making them not sell, I think it's quite the opposite, people with multiple monitors will surely try to get their hands on them, like the one in the video.
Then, about 4k, I don't think ELP3 wants it now, let him say it, I think he wants to play at 1600p with everything possible activated and play well of course at constant 60 fps.
Regards
-
Well then better, while you guys are advancing we have Mr. Scissors, anyway I love the Titan as a card, I think it's a great card, although it goes beyond what I'm willing to pay, if they release a cheaper one with 3Gb then I'll try to get at least one.
Anyway they said there would be very few but I see them everywhere, plus the price isn't making them not sell, I think it's the opposite, people with multimonitor will surely try to get their hands on them, like the one in the video.
Then the 4k thing I don't think I want it now ELP3, let him say it, I think he wants to play at 1600p with everything possible activated and play well of course at constant 60 fps.
Regards
If I also love this graphics card and if they release one for $600 with 3Gb and with 20% less performance there will be several of us who will fall for the temptation.
The video is very good, it's a blast to play like that.-
Salu2.-
-
It will depend on whether the synthetic test you pass uses more than 2GB of VRAM or not, because if it doesn't, I tell you that the result will be THE SAME in both, and in case there is some use of VRAM above 2GB, the 4GB version will go from slightly to much better (depending on the amount of VRAM that the application really uses, and beware, DO NOT confuse this VRAM in use with what the graphics memory meter says in Afterburner, not because it doesn't measure the graphics memory used by an application X, but because even if it did, it's not the same to measure the VRAM memory used in the purest style of a "victim cache" in graphics with a lot of VRAM to the authentic need for more VRAM to render an image.
So the 4GB will be important or necessary as soon as we use a large number of filters etc ….., right?
Well I can see that it's not that much of a difference.
-
So then, the 4 GB will be important or necessary when we use a large number of filters, etc..., right?
Well, I can see that the difference is not that big.
Only if the use of those filters really makes the application exceed the 2GB of graphics memory in use.
If not, it's exactly the same.
By modifying the BIOS, you can increase the fans, Power Target limit, and in theory, the voltage, but this doesn't significantly improve OC on GTX 680s, it's about 40-50MHz, not very important, because NVIDIA has limited the power consumption of its GPUs with a chip independent of the GPU. This chip maintains consumption and temperatures within the limits they consider safe... without the possibility of modifying it.
I can confirm that you can change all that you mentioned at the beginning, more so, of course, the frequencies. But the OC limit, which will depend on each GPU, gained with overvolting (which in my case confirms those extra 40-50 MHz), depends on each unit, and is not at all limited by that chip that you mentioned, independent of the GPU. That little gain (although "little" is relative) is due to the fact that the allowed overvolting is up to 1.2125V, and there seems to be a hard limit imposed there (it's said that this doesn't happen in all Kepler boards, some can go further), a voltage that, compared to the stock 1.175V, is 3.2% higher. So, in addition to the fact that the voltage point of 1.175V is probably the inflection point of the frequency/voltage curve in terms of scalability, the overvolting we're talking about is only 3.2%. So, it's a pretty modest overvolting and therefore one can't expect miracles to take the GPU much beyond what it achieves at 1.175V. It's pretty logical, in fact, what happens. Without the need for explanations of "external chips".
Consumption is indeed modified, and temperature, well, except that we're talking about exceeding safe temperature limits, for which control doesn't need any chip on the board, but rather the driver itself can cut "gas" by reading the internal sensors of the GPU, but not because it's an "artificial limitation", but because it's necessary to prevent the GPU from being damaged.
Any modern CPU or GPU will protect itself by lowering frequencies or even blocking/turning off if it exceeds a certain temperature "trigger" considered dangerous, but this happens not only with NVIDIA CPUs/GPUs, but also with AMD or Intel, etc.
Regarding consumption, my graphics card has easily exceeded the "maximum" allowed consumption limit by default (+122%) thanks to the BIOS modification, so that part of what you said is not true, because I've seen just the opposite by making a BIOS modification. Modern graphics cards all carry TDP control chips and so on, but that doesn't mean that the TDP limit is a "hard" limit and not configurable.
Best regards.
-
I was referring to the AA filter with SuperSampling, do the 4 GB affect having them, no?
Better to have them in that case for most PC games and synthetic benchmarks.
-
Congratulations ELPE, well done. I'm glad you added a Spanish touch to the top ten.
I thought the cards were given to you, but I see that you bought them. That's more merito, but tell me what you work at and I'll make a note of it!!!;D -
The truth is that you don't stop man, you're a great one, let's see if I can find time and do a small update myself too because from what I see they throw the dice well these ones aren't ELP?, I have this a bit abandoned to be honest, don't let your guard down for a second look every day in futuremark... don't let a guy from Murcia come and kick you in the ass... jajaja :mad:, it's a joy to read us all again, with enthusiasm and good atmosphere which is what was needed, well as I said a pleasure as always and we'll be around here, great job as always makina¡¡¡ ;D
-
The truth is that you never stop, man, you're a great one. Let's see if I can find some time and do a small update myself, because from what I see, these chicks are doing well, not ELP? I've kind of abandoned this, to be honest. Don't let your guard down for a second, check in on Futuremark every day... don't let a guy from Murcia come and kick your ass... jajaja :mad:, it's a joy to read all of us again, with enthusiasm and good vibes, which is what was needed. Well, as I said, a pleasure as always, and we'll be around here, great work as always, makina!!! ;D
Well pakitoesp, it makes me happy to read you again, you don't know how many times I've thought of you, and even talked about you with buddies, and had no idea about you…...
I hope you come back to it soon, we want to see those beautiful rigs you used to build again.
Welcome back and as I said, it's a joy to "see" you again.....
A hug...!!
-
¡Esta publicación está eliminada! -
It is indeed a joy to see pakito back in the forums. Keep up the good work.
-
I was referring to the AA filter with SuperSampling, do they affect the 4 GB of having them, right?
Better to have them in that case for the majority of PC games and synthetic benchmarks.
No, it's a filter like any other with an X consumption of extra VRAM, it doesn't mean that it needs more than 2GB of VRAM by obligation because it's supersampling, what SSAA does need is bandwidth. Which, in any case, is the same for a 2 GB version as for a 4 GB.
It depends on the game you look at, but the vast majority, with supersampling or not, perform the same. Obviously you can always force things so that an application consumes much more VRAM than it should, perhaps by forcing MSAA on top of SSAA or something like that and using three monitors, but then don't expect adequate performance with 2 or 4 GB of VRAM either.
This type of graphics card pairs well with 2 GB of VRAM, 4 GB is more of a marketing point for the model than for its real performance.
-
Congratulations ELP3 again among the top 10, if I remember correctly this is the best position you have achieved or at least since I know you, impressive!!!!
With so many Titans in your power is there any project for a 4K TV (as I understand they are limited to 30 fps) or monitor like the Sharp PN-K321 during this year.
Obviously their prices are high between U$S 5000 to U$S 10000 but those who have tried them say that their quality is monumental.-
Salu2.
Thanks patagonico, and also welcome..;)
It's not my best position. I was first in both xtreme and perfomace at the launch of the GTX 680. Before of course, the arrival of professional overclokers, however I was always within the Top HOF with that score.
I have no intention at the moment of a 4K panel... its price must be astronomical, and as you say they would be capped. I already have enough with the Dell 30" at 2560X1600p or multinimonitor.La the joy, is that you can put EVERYTHING you want, to EVERYTHING that the game gives..it doesn't matter 4 that 8X, that SSAA, that MSAA etc..it's a fucking joy...and 4 WAY is not that it scales well.Es that it's the best 4 way I've had.
A hug.
The truth is that you don't stop man, you're a crack, let's see if I can find time and do a small update I also have these chicks that are doing well, not ELP?, I have this a bit abandoned the truth, don't let your guard down for a second look every day in futuremark …. don't let a Murcian come and kick you in the ass.... jajaja :mad:, a joy to read us all again, with enthusiasm and good atmosphere that's what was needed, well as I said a pleasure as always and we'll be around here, great job as always makina¡¡¡ ;D
Don pako..what a joy to have you here..jeje..
Titan VERY good..in the order of 50-60% better at my resolution and filters than a 680…I think that has never been seen before within the same node and generation.Whenever you want Murcian,I'll be ready to dispute my position with you...even if it's by pouring liters of water over it...jeje..
Yes,the truth is that posting here is another story.Everything is based on respect for others and the desire to learn,so there shouldn't be problems..
I'm SO HAPPY to hear from you again.
A big hug Paco.
-
No, it is a filter like any other with an X consumption of extra VRAM, it doesn't mean that it needs more than 2GB of VRAM by obligation because it is supersampling, what SSAA does need is bandwidth. Which, by the way, is the same for a 2 GB version as for a 4 GB.
It depends on the game you look at, but the vast majority, with supersampling or not, perform the same. Obviously you can always force things to consume much more VRAM from an application that it's due, perhaps forcing MSAA on top of SSAA or something like that and using three monitors, but then don't expect adequate performance with 2 or 4 GB of VRAM either.
This type of graphics card pairs well with 2 GB of VRAM, 4 GB is more of a marketing point for the model than for its real performance.
If it's the bandwidth that SSAA depends on then I don't care if it's the 2 GB card or 4 GB what is needed is a superior chip in this case the TITAN.
And since I don't have three monitors, nor do I play at resolutions over 1080p, I don't need the 4 GB of VRAM.
It's getting very clear to me, thanks for the explanation.
Best regards
PD: When improved versions of the TITAN come out I can already see myself buying at least one.
-
If SSAA depends on bandwidth, then I don't care if it's the 2GB or 4GB card, what's needed is a superior chip, in this case the TITAN.
And since I don't have three monitors, or play at resolutions higher than 1080p, I don't need the 4GB of VRAM.
It's becoming very clear to me, thanks for the explanation.
Best regards
P.D.: When improved versions of the TITAN come out, I can already see myself buying at least one.
Don't be crazy... ;D
It's not like that either, the GTX 680 use their bandwidth very well and the truth is that they move quite well with SSAA, another matter is when a game, in addition to being able to use SSAA, can activate other bandwidth-consuming effects, as an example the last Tomb Raider that we have already discussed here:
The game runs very smoothly at 2xSSAA, and in a decent way, although not perfect at 4xSSAA and everything else at full except for two details:
DoF via DirectCompute, AMD's implementation of the effect in addition. Very expensive both in processing and bandwidth, although in equivalent implementations of nvidia they also depended a lot on bandwidth (Metro 2033), AMD's implementation makes you lose even more performance with nvidia.
TressFX, the effect is totally bugged, it's not normal or reasonable that suddenly when approaching a light source (and not all of them, just "here" and "there" apparently without any clear reason) the fps rate drops from about 55 fps to just under 20 fps, that can happen with this implementation, and the cost of the effect is approximately the same as the DoF above, I don't know if the bandwidth consumption is important or not with this effect, but what I do know is that it's irregular and negatively affects performance. In addition to the huge amount of glitches and other associated artifacts.
These two effects if you combine them with SSAA may not be manageable on a GTX 680, except that you limit yourself to something like 2xSSAA. Which goes pretty well, although being so unstable and bugged the mentioned effects (TressFX especially), it's usually not a good idea.
But anyway, the game can really be used with SSAA, and like him others, I played SniperV2 with 2.5xSSAA, and I turned off AMD's DirectCompute effects because these were the ones that most affected performance (recently with DirectCompute effects activated, nvidia cards have gained a +25% in fps, verified by myself). SSAA is more bearable than people think with the small Keplers.
In the same way that DirectCompute works pretty well on these cards IF the effects are well implemented.
The Titan seems to me an overkill to play at 1080p, but well, you decide. I'm also not sure if there will be a cheaper Titan clearly, if the 7 series comes out in a few months (they would step on the more advanced 7s and the discounted Titans).
