Welcome to techngaming.com
Register now to gain access to all of our features. Once registered and logged in, you will be able to create topics, post replies to existing threads, give reputation to your fellow members, get your own private messenger, post status updates, manage your profile and so much more. This message will be removed once you have signed in.Login to Account Create an Account
Rumour: NVIDIA GTX Titan to have 915MHz GPU Clock, 512-Bit Memory Bus?
Feb 11 2013 03:00 AM |
qubit
in News
As we approach the February 18 launch date of the NVIDIA GTX Titan flagship graphics card, the product rumours and leaks start coming in thick and fast. This latest one is from Australian retailer Austin Computers and what makes it especially interesting are the supposed GPU clock speed and memory bus width.
Their product page briefly listed the Asus version before it was taken down. Of course, said product page was dutifully snapped by our friends over at VideoCardz before this happened, preserving their leak forever. Interestingly, the product listing shows the card as having a 915MHz GPU clock speed with a boost clock of 1019MHz, memory clock of 6008MHz effective, a memory size of 6GB and a 512-bit memory bus width.

This is interesting for several reasons. Firstly, the GPU clock and memory clock are the same speed as the current GTX 690 and this GPU speed is much faster than the previously rumoured 732MHz. Then, we have the memory bus width of 512 bits which would be considerably better than the previously reported 384-bit one. However, it's a strange one given that the memory size is given as 6GB and the GPU only has 15 SMX units max for the uncut version (see article picture). This looks like a typo, because a power of two bus width would require 16 SMX units and either 4GB or 8GB of RAM to be attached, which is also a power of two.
This is why 6GB is an odd figure, so something is wrong here. I'm inclined to believe that the 512-bit spec is a typo, especially as the GeForce version is likely to have only 14 SMX units, rather than the full 15. This gives rise to 2688 Cuda Cores for 14 SMX units as opposed to the 2880 Cuda Cores for 15 SMX units in the Tesla K20 HPC card which is based on the full, uncut GK110 GPU. Of course, if the full GK110 had 16 SMX units, then we would see a massive 3072 Cuda Cores and a 512-bit memory bus. But we'll just have to continue dreaming about such a chip since NVIDIA didn't design this version, no doubt due to transistor, power and heat budget constraints.
Other details listed about this card include two dual DVI-I ports, one DVI-D port and one mini-DisplayPort. There's no mention of a HDMI port, but missing out such an obvious and widely used port that other recent NVIDIA cards use seems unlikely. Then again, one of the other ports could work with a suitable HDMI adapter plug or adapter cable. Finally the price is set at a wallet-busting AUD 1599. Of course, this is pre-release pricing, so should be considered as a guide only. Some of the details on Austin Computers' product page are likely to be correct however, since this is the third leaked retailer's website to show such information, giving some weight to it. As ever, these details should be taken with a large pinch of salt until the official launch.
GeForce Titan specs compared to other NVIDIA cards and to AMD's HD 7970 GE:

For us enthusiasts who like to know these things, here are the bus widths for recent generations of NVIDIA graphics cards:
GTX 580 - 384-bit
GTX 480 - 384-bit
GTX 285 - 512-bit
GTX 285 - 512-bit
Note that there are no true 300 series cards, since these were all low end OEM cards which were simple rebrands of the 200 series designed to fool uneducated buyers of prebuilt PCs, so are therefore completely uninteresting to enthusiasts.
Source: VideoCardz
Finally, we have included NVIDIA's official GK110 white paper below for those who want to check out the architecture and capabilities of this GPU. While it doesn't state the GPU's memory bus width, it does show it as having only 15 SMX units, rather than a "perfect" 16.
NVIDIA-Kepler-GK110-Architecture-Whitepaper.pdf 1.67MB
46 downloads
Their product page briefly listed the Asus version before it was taken down. Of course, said product page was dutifully snapped by our friends over at VideoCardz before this happened, preserving their leak forever. Interestingly, the product listing shows the card as having a 915MHz GPU clock speed with a boost clock of 1019MHz, memory clock of 6008MHz effective, a memory size of 6GB and a 512-bit memory bus width.

This is interesting for several reasons. Firstly, the GPU clock and memory clock are the same speed as the current GTX 690 and this GPU speed is much faster than the previously rumoured 732MHz. Then, we have the memory bus width of 512 bits which would be considerably better than the previously reported 384-bit one. However, it's a strange one given that the memory size is given as 6GB and the GPU only has 15 SMX units max for the uncut version (see article picture). This looks like a typo, because a power of two bus width would require 16 SMX units and either 4GB or 8GB of RAM to be attached, which is also a power of two.
This is why 6GB is an odd figure, so something is wrong here. I'm inclined to believe that the 512-bit spec is a typo, especially as the GeForce version is likely to have only 14 SMX units, rather than the full 15. This gives rise to 2688 Cuda Cores for 14 SMX units as opposed to the 2880 Cuda Cores for 15 SMX units in the Tesla K20 HPC card which is based on the full, uncut GK110 GPU. Of course, if the full GK110 had 16 SMX units, then we would see a massive 3072 Cuda Cores and a 512-bit memory bus. But we'll just have to continue dreaming about such a chip since NVIDIA didn't design this version, no doubt due to transistor, power and heat budget constraints.
Other details listed about this card include two dual DVI-I ports, one DVI-D port and one mini-DisplayPort. There's no mention of a HDMI port, but missing out such an obvious and widely used port that other recent NVIDIA cards use seems unlikely. Then again, one of the other ports could work with a suitable HDMI adapter plug or adapter cable. Finally the price is set at a wallet-busting AUD 1599. Of course, this is pre-release pricing, so should be considered as a guide only. Some of the details on Austin Computers' product page are likely to be correct however, since this is the third leaked retailer's website to show such information, giving some weight to it. As ever, these details should be taken with a large pinch of salt until the official launch.
GeForce Titan specs compared to other NVIDIA cards and to AMD's HD 7970 GE:
For us enthusiasts who like to know these things, here are the bus widths for recent generations of NVIDIA graphics cards:
GTX 580 - 384-bit
GTX 480 - 384-bit
GTX 285 - 512-bit
GTX 285 - 512-bit
Note that there are no true 300 series cards, since these were all low end OEM cards which were simple rebrands of the 200 series designed to fool uneducated buyers of prebuilt PCs, so are therefore completely uninteresting to enthusiasts.
Source: VideoCardz
Finally, we have included NVIDIA's official GK110 white paper below for those who want to check out the architecture and capabilities of this GPU. While it doesn't state the GPU's memory bus width, it does show it as having only 15 SMX units, rather than a "perfect" 16.
NVIDIA-Kepler-GK110-Architecture-Whitepaper.pdf 1.67MB
46 downloads




Sign In
Create Account














3 Comments
I'll bet NVIDIA release a revised Titan within a year that has all the SMX units enabled. This will be a similar situation between the GTX 480 and GTX 580, where the latter was just an optimised Fermi, allowing the disabled cluster to be enabled and higher clocks to be used. This significantly improved the performance of the card.
Therefore, this revised Titan will be the one I'm interested in. In the meantime, my GTX 580 works fine, especially as I game at 1080 @ 120Hz refresh.
As someone who bought two 680's...
I'll never pay more than $400 per card anymore.
I'll bet you do.
You're an addict, go on admit it! 