Friday, March 19th 2010

GeForce GTX 400 Series Clock Speeds and Other Details Surface

Exactly a week ahead of releasing its GeForce GTX 400 series accelerators, NVIDIA held meetings with the press discussing the company's newest technologies, including GeForce GTX 400 series. Some lesser known details about the GeForce GTX 480 and GeForce GTX 470 surfaced, among more known and established ones. To begin with, the GeForce GTX 480 is confirmed to have a CUDA core (shader core) count of 480. The GF100 core operates at 700 MHz, its shader domain at 1401 MHz, and the memory operates at 924 MHz (actual, 1848 MHz DDR, 3700 MHz effective). With a GDDR5 memory bus width of 384-bit, the effective memory bandwidth would be 173.4 GB/s.

The GeForce GTX 470, on the other hand, has 448 CUDA cores, clock speeds of 607 MHz core, 1215 MHz shader domain, and 837 MHz memory (actual, 1674 MHz DDR, 3348 MHz effective). With a GDDR5 memory bus width of 320-bit, the effective memory bandwidth would be 130.7 GB/s. While the GTX 480 has a board power of 295W, the GTX 470 has a board power of 225W. Another piece of information the source reveals is that internal testing by NVIDIA showed that the performance level to expect from the GeForce GTX 470 should be 5-10% higher than that of the ATI Radeon HD 5850. The GeForce GTX 480 should be expected to be just that much faster than the ATI Radeon HD 5870. It is also expected that the target price of the GeForce GTX 480 should be typically US $499, while the GTX 470 should go typically for US $349. Detailed reviews of the two should be up by this time, next week.Source: VR-Zone
Add your own comment

82 Comments on GeForce GTX 400 Series Clock Speeds and Other Details Surface

#1
jamsbong
Fermi does not look all that impressive considering they are 6+months late. At a ridiculous 295watt TDP, it is highly doubtful that it can be overclocked. I won't be surprise if it is pre-overclock to the fullest just to overtake ATI by 10%.

If I want DX11, I would go for a decent ATI card. If I want CUDA or Physx, I would rather buy some mid-end nvidia card.

Tessellation is largely unseen at the moment. by the time it is available in every PC game, the price of these DX11 card would be so much cheaper than they are now.

One thing FERMI do not have is Eyefinity. ATI multi-monitor tech is hardware based. That is why you can pump 6 monitors with 1 card. NV's tech is software based which allows 2 or 3 monitors per card. if you want more output you need 2+ cards.

Another note, AMD has launch their "gaming evolved" initiative. For a long time Nvidia always have great relationship with developers. Now AMD is finally doing something like it. This is great news for ATI users.
Posted on Reply
#2
Super XP
jamsbong said:
Fermi does not look all that impressive considering they are 6+months late. At a ridiculous 295watt TDP, it is highly doubtful that it can be overclocked. I won't be surprise if it is pre-overclock to the fullest just to overtake ATI by 10%.

If I want DX11, I would go for a decent ATI card. If I want CUDA or Physx, I would rather buy some mid-end nvidia card.

Tessellation is largely unseen at the moment. by the time it is available in every PC game, the price of these DX11 card would be so much cheaper than they are now.

One thing FERMI do not have is Eyefinity. ATI multi-monitor tech is hardware based. That is why you can pump 6 monitors with 1 card. NV's tech is software based which allows 2 or 3 monitors per card. if you want more output you need 2+ cards.

Another note, AMD has launch their "gaming evolved" initiative. For a long time Nvidia always have great relationship with developers. Now AMD is finally doing something like it. This is great news for ATI users.
Rumours have it Fermi was late because of design defects. I love this title, its funny like hell :laugh:
Nvidia's Fermi GTX480 is broken and unfixable
Hot, slow, late and unmanufacturable
http://www.semiaccurate.com/2010/02/17/nvidias-fermigtx480-broken-and-unfixable/

Obviously NVIDIA cannot admit that something went wrong with Fermi or NVIDIA's stock price would go way down into the dirt. All NVIDIA wants to do IMO right now is release what they have, try to get away with the heat issues, then work hard in getting Fermi 2 up and running IMO
In the mean time, I'll enjoy my ATI Radeons thank U :D
Posted on Reply
#3
Grings
What i dont get is why the ram is so slow?

I know they are using a higher bit interface and all, but considering even ATI's semi-budget cards have modules rated for 5000mhz effective, it seem odd to see a number as low as 3700mhz bandied around.

Saving it for the 485 maybe?, the cooler inadvertently heats the ram too much to clock it any higher? very odd either way...
Posted on Reply
#4
Wile E
Power User
jamsbong said:
Fermi does not look all that impressive considering they are 6+months late. At a ridiculous 295watt TDP, it is highly doubtful that it can be overclocked. I won't be surprise if it is pre-overclock to the fullest just to overtake ATI by 10%.

If I want DX11, I would go for a decent ATI card. If I want CUDA or Physx, I would rather buy some mid-end nvidia card.

Tessellation is largely unseen at the moment. by the time it is available in every PC game, the price of these DX11 card would be so much cheaper than they are now.

One thing FERMI do not have is Eyefinity. ATI multi-monitor tech is hardware based. That is why you can pump 6 monitors with 1 card. NV's tech is software based which allows 2 or 3 monitors per card. if you want more output you need 2+ cards.

Another note, AMD has launch their "gaming evolved" initiative. For a long time Nvidia always have great relationship with developers. Now AMD is finally doing something like it. This is great news for ATI users.
If it's anything like TWIMTBP, it is great news. ATI should've had a similar program in place years ago. Now maybe we'll see better ATI support out of the box on new titles, instead of having to wait for ATI to patch things up in drivers.

As for the 400 series specs, what can I say that hasn't been said already. Power consumption is far too high for the hinted at performance level, and this is way too late to the party.

That said, I'm dying to know the real numbers. nVidia has walked all over ATI in terms of GPGPU performance. My 8800GT is faster than my 4870X2 in Folding, for example. I want to see what Fermi can do in folding and encoding.
Posted on Reply
#5
Hayder_Master
why memory size not like normal 1024,2048, is there is a point to pick this numbers
Posted on Reply
#6
btarunr
Editor & Senior Moderator
hayder.master said:
why memory size not like normal 1024,2048, is there is a point to pick this numbers
1536 is 1.5 x 1024, so 1.5 GB. 1280 is 1.25 x 1024, so 1.25 GB. Since each memory chip has a 32-bit wide path, and the GPU has 384-bit / 320-bit memory interfaces (for GTX 480 / GTX 470, respectively), there are 12 or 10 memory chips. Each chip is 1 Gbit (128 MB), so 1536 MB for 12 chips, or 1280 MB for 10 chips. AMD HD 5800 use 256-bit memory interfaces, so 8 chips, 1024 MB. To get 2048 MB, they make two chips share a 32-bit path.
Posted on Reply
#7
Hayder_Master
btarunr said:
1536 is 1.5 x 1024, so 1.5 GB. 1280 is 1.25 x 1024, so 1.25 GB. Since each memory chip has a 32-bit wide path, and the GPU has 384-bit / 320-bit memory interfaces (for GTX 480 / GTX 470, respectively), there are 12 or 10 memory chips. Each chip is 1 Gbit (128 MB), so 1536 MB for 12 chips, or 1280 MB for 10 chips. AMD HD 5800 use 256-bit memory interfaces, so 8 chips, 1024 MB. To get 2048 MB, they make two chips share a 32-bit path.
thanx bro that's very useful info form me :toast:
Posted on Reply