Overclock.net banner

1 - 20 of 24 Posts

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #1
GDDR3 vs GDDR4
1) 7.2 vs 11.2 GBps per chip max data speed
2) 1.8 vs 2.8 GBps per pin max data rate
3) 1.8 vs 1.5 v minimum power consumption which results in a 20-25% energy savings
4) 2.1 vs 1.9 v maximum power for overclocking
5) 4 vs 8 bit burst length which allows more memory bandwidth and lower latency
6) no DBI vs DBI: Data Bus Inversion - it takes more energy to transfer 0s than 1s. When there are a lot of 0s to move, it will send them as 1s instead and then use DBI to signal that all those 1s needed to be flipped back to 0s. This is a power saving feature.
7) Three point voltage control via video card driver for fine tuning and overclocking

As a comparison, the X1900xt uses GDDR3 at 1.45GHz while the X1950xtx uses GDDR4 at 2GHz. This difference helps the the X1950xtx get 64GBps memory bandwidth over the 46.4GBps of the X1900xt. As for the future, GDDR5 is expected in 2008 and will use 1.2-1.5v. The next GDDR interation might be in 2010 and use 1.2v max.

Source: Computer Power User Magazine
 

·
Registered
Joined
·
3,113 Posts
Interesting.

Do you think those figures make a noticeable difference in todays games? Are there games that take advantage of the extra bandwith from DDR4?
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #4
Quote:


Originally Posted by Joeking78
View Post

Interesting.

Do you think those figures make a noticeable difference in todays games? Are there games that take advantage of the extra bandwith from DDR4?

Yes, GDDR3 is pretty much at its limit. The future is GDDR4!

@ATI83, the specs of future GDDR aren't set yet. However, engineers are working on it and probably do have a good idea of the voltages. It takes years to develop the technology.
 

·
Premium Member
Joined
·
5,395 Posts
Quote:


Originally Posted by DuckieHo
View Post

GDDR3 vs GDDR4
1) 7.2 vs 11.2 GBps per chip max data speed
2) 1.8 vs 2.8 GBps per pin max data rate
3) 1.8 vs 1.5 v minimum power consumption which results in a 20-25% energy savings
4) 2.1 vs 1.9 v maximum power for overclocking
5) 4 vs 8 bit burst length which allows more memory bandwidth and lower latency
6) no DBI vs DBI: Data Bus Inversion - it takes more energy to transfer 0s than 1s. When there are a lot of 0s to move, it will send them as 1s instead and then use DBI to signal that all those 1s needed to be flipped back to 0s. This is a power saving feature.
7) Three point voltage control via video card driver for fine tuning and overclocking

As a comparison, the X1900xt uses GDDR3 at 1.45GHz while the X1950xtx uses GDDR4 at 2GHz. This difference helps the the X1950xtx get 64GBps memory bandwidth over the 46.4GBps of the X1900xt. As for the future, GDDR5 is expected in 2008 and will use 1.2-1.5v. The next GDDR interation might be in 2010 and use 1.2v max.

Source: Computer Power User Magazine


I'm pretty sure they've got that backwards. 1's mean current, 0's mean no current. How could no current POSSIBLY take MORE energy to send current then NO current?

http://www.bookrags.com/Binary_numeral_system

Quote:


Operational characters 1 and 0 stand for 1 = on = closed circuit = true 0 = off = open circuit = false

 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #6
Quote:


Originally Posted by Ihatethedukes
View Post

I'm pretty sure they've got that backwards. 1's mean current, 0's mean no current. How could no current POSSIBLY take MORE energy to send current then NO current?

http://www.bookrags.com/Binary_numeral_system

Depends on what you define as 0 or 1. It you want high as 1, then fine. If you want high as 0, then fine. Not sure about specific implemenation.

Or... the constant signal is high. To create a 0 signal, they use transisitors to close a gate. Powering the transistors is the actual energy cost. Therefore, a 0 signal is more expensive.
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #9
Quote:


Originally Posted by ElMikeTheMike
View Post

whats with all these old threads getting dug up outta no where?

Thread is still pertinent and I did link it twice this week.
 

·
Premium Member
Joined
·
5,395 Posts
As far as I know, 1's are ALWAYS current. If this changed, please let me know. However, they're speaking about transferring of signal. In the transfer of signal there's only one way to expend energy throwing current across the circuit. They specifically talk about a singular bit that swaps the 1s and 0s. If you were to charge ONE 0 to a 1 but swapped it for 50 (of 64 bits) 1s changing to 0s you're gonna get a net power savings of a lot. If is takes 20 (number pulled from my ass) other 0s swapping to 1's to do it you've STILL got a net 'profit' of 14+20=34>50 current carrying bits.

Above and beyond that, what you're saying is akin to claiming it might take more energy for you to flip the light switch off on the lamp than just leave it on. That better be one damned efficient lamp if switching it off takes more energy than using it. Logically, if things were that efficient, we wouldn't even be looking for power savings anymore, would we?

I'm fairly certain it's just a big typo.

You'd think that with a 64bit bit pathway
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #11
Quote:


Originally Posted by Ihatethedukes
View Post

As far as I know, 1's are ALWAYS current. If this changed, please let me know. However, they're speaking about transferring of signal. They specifically talk about a singular bit that swaps the 1s and 0s. If you were to charge ONE 0 to a 1 but swapped it for 50 (of 64 bits) 1s changing to 0s you're gonna get a net power savings of a lot. If is takes 20 (number pulled from my ass) other 0s swapping to 1's to do it you've STILL got a net 'profit' of 14+20=34>50 current carrying bits.

I'm fairly certain it's just a big typo.

You'd think that with a 64bit bit pathway

Write them a letter to confirm? Either way... it works.
 

·
Premium Member
Joined
·
5,395 Posts
Quote:


Originally Posted by DuckieHo
View Post

Write them a letter to confirm? Either way... it works.


It's all good, I'm not arguing with you, man. I'm just trying to clarify why I said what I did. (Can't afford looking like a COMPLETE idiot, hehe.)
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #13
Quote:


Originally Posted by Ihatethedukes
View Post

It's all good, I'm not arguing with you, man. I'm just trying to clarify why I said what I did. (Can't afford looking like a COMPLETE idiot, hehe.)

No doubt, I was just playing Devil's Advocate.
 

·
Registered
Joined
·
1,079 Posts
lets drop the DDR and go QDR...lol Quad Data Rate FTW lol
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #15
Quote:

Originally Posted by deadagain6591 View Post
lets drop the DDR and go QDR...lol Quad Data Rate FTW lol
It requires more leads per memory module. It is currently easier to just shrink parts and tighen specs than redesign PCB and memory access.
 

·
Registered
Joined
·
1 Posts
So far GDDR 3 is only using 32% of its Capability and GDDR 4 is running on 67% and the difference isnt that big . You cant fell the difference (By Performance) And Gamers wouldnt feel any difference at all.
The upcomming 9900 series gonna run with 1GB GDDR3 (the smallest ones) and will outrun. almost every GDDR4 Cards comming. (in games)

So final words is

GDDR3 = Best Performance for Games
and
GDDR4 = Best Performance for Programs.

But again Spend your money where you want. ill still smile when people calls in for support
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #17
Quote:


Originally Posted by Talent
View Post

So far GDDR 3 is only using 32% of its Capability and GDDR 4 is running on 67% and the difference isnt that big . You cant fell the difference (By Performance) And Gamers wouldnt feel any difference at all.
The upcomming 9900 series gonna run with 1GB GDDR3 (the smallest ones) and will outrun. almost every GDDR4 Cards comming. (in games)

So final words is

GDDR3 = Best Performance for Games
and
GDDR4 = Best Performance for Programs.

But again Spend your money where you want. ill still smile when people calls in for support


What are you talking about....

Their is no 9900 series. There is a performance difference. GDDR4 would benefit gamers. There are no upcoming GDDR4 cards at this point since it looks like GDDR5 is coming out soon and Qimonda isn't developing GDDR4

GDDR3 is maxed out already at around 2GHz. A lot of overclocking testing show that memory is a big limitation on the 8800GT.
 
  • Rep+
Reactions: noobdown

·
Premium Member
Joined
·
5,395 Posts
That's true Duckie but that is more of a bitwidth problem over a memory refresh problem (although that's one way to circumvent a narrow bitwidth problem)
 

·
Registered
Joined
·
357 Posts
is this talent guy a angry spammer from an other forum or is there a secret 9900 series? maybe he's a topguy at nvidia or maybe he's an alien!
 

·
Premium Member
Joined
·
65,162 Posts
Discussion Starter #20
Quote:


Originally Posted by celebro
View Post

is this talent guy a angry spammer from an other forum or is there a secret 9900 series? maybe he's a topguy at nvidia or maybe he's an alien!


read his sig....

Quote:


Bjørn Christian Andersen
Support Engineer
Asus Personal Support Denmark

 
  • Rep+
Reactions: LAKEINTEL
1 - 20 of 24 Posts
Top