How much faster is GDDR5 than GDDR3?

M

MyIllusion

n00b
Joined
May 14, 2008
Messages
39
Does anyone know any links w/ charts or written reviews that explain how much better/faster a card with 256-bus 512mb GDDR5 is compared to GDDR3?

I'm not sure if I should pick up 9800GTX+, GTX 260 or just go with 4870 that has the GDDR5.

I have an SLI motherbord. What to do?
 
You cant just look at the memory... The cards have to be taken as a complete package when you evaluate them.
Anyway, the two cards have about equal bandwidth. You can pretty much just pretend that DDR5 doubles the bus width from 256 to 512.
If you are thinking of SLI or Xfire you should just go ahead and go with a 4870X2 as it is cheaper than two cards and will work on your board.
 
Seeing as you have an SLI mobo, the 260 is probably the better choice.
GDDR5 + 256bus = GDDR3 + 512bus
 
veedubfreak said:
Seeing as you have an SLI mobo, the 260 is probably the better choice.
GDDR5 + 256bus = GDDR3 + 512bus
Click to expand...

So when the GDDR5 memory is traveling on the bus it's not 2x as fast speed wise as GDDR3, but just the bandwidth is doubled not the actual speed? This is so confusing to me.

Did that question make sense?
 
Jakalwarrior said:
You cant just look at the memory... The cards have to be taken as a complete package when you evaluate them.
Anyway, the two cards have about equal bandwidth. You can pretty much just pretend that DDR5 doubles the bus width from 256 to 512.
If you are thinking of SLI or Xfire you should just go ahead and go with a 4870X2 as it is cheaper than two cards and will work on your board.
Click to expand...

I don't think X2 is my style if my games will stutter then that defeats the purpose of spending alot of money on a card. Feel Me? :eek:
 
Then going SLI or CF won't matter to you, and AFAIK, stuttering isn't anywhere near as big an issue with the 4870X2 as before, in which case 4870 or GTX 260 is close to a wash.

And *speed* of memory isn't as big a deal as the effective bandwidth you get out if it. So ignore how fast the memory is actually working (and GDDR5 is twice as fast anyways), what matters in the end is how much bandwidth is allowed, in which case GDDR5 + 256-bit is the same as GDDR3 + 512-bit pretty much.
 
GDDR5 is basically GDDR3 with +2 to agility
Its a good increase, but definitely not epic.
 
GDDR5 is a significant improvement over GDDR3 in every fashion really. You can read the white paper if you want on it, but its more than just speed improvements
 
Well I found this on Wiki...

"The GDDR3 Interface transfers two 32 bit wide data words per clock cycle from the I/O pins."

"The GDDR5 interface transfers two 32 bit wide data words per WCK clock cycle to/from the I/O pins."

So that means GDDR5 transfers memory data 2x as fast creating smoother texture/pixel graphics drawing than GDDR3?

Maybe a smoother less jumpy overall framerate in your games, when comparing the same GPU's at same GPU core speed with GDDR5 vs GDDR3 being the only difference?

That's what I think their saying but hell I didn't design it lmao :p
 
All i know is that my 4870 crossfire set up rocks and maxes out every game i play (screw you crysis you suck anyway) at 1920x1200.
 
The speed of the memory isn't what deals with frames - the GPU core does that, memory is just for buffer but effective bandwidth is what matters

Actual speed isn't an issue
 
4870 is just under 8% faster than the 4850 at the same clocks. Since the only difference between the 4850 and 4870 (apart from the clockspeed) is the GDDR3 vs GDDR5 memory, the real world performance benefit of GDDR5 memory appears to be 8% on average. This is only one game, though. I wish someone would do a more thorough inverstigation of this.
http://www.xtremesystems.org/FORUMS/showthread.php?t=192690&page=4
 
JimmiG said:
4870 is just under 8% faster than the 4850 at the same clocks. Since the only difference between the 4850 and 4870 (apart from the clockspeed) is the GDDR3 vs GDDR5 memory, the real world performance benefit of GDDR5 memory appears to be 8% on average. This is only one game, though. I wish someone would do a more thorough inverstigation of this.
http://www.xtremesystems.org/FORUMS/showthread.php?t=192690&page=4
Click to expand...

Those tests were also run at fairly tame settings (1680x1050 with low AA), so at higher resolutions and amounts of AA chances are the benefits of the higher RAM bandwidth that GDDR5 provide would be much more significant.
 
Zero82z said:
Those tests were also run at fairly tame settings (1680x1050 with low AA), so at higher resolutions and amounts of AA chances are the benefits of the higher RAM bandwidth that GDDR5 provide would be much more significant.
Click to expand...

Still waiting for a [H] review of the 4850 and 4870 at the same clockspeeds :) The performance improvement from GDDR5 can't be that big however - even with both cards at their respective stock clock speeds, the 4870 is only 15-30% faster.
 
It seems like there should be some noticable benefit? I have only seen GDDR3 memory so I trying to get sum info on GDDR5.
 
MyIllusion said:
Does anyone know any links w/ charts or written reviews that explain how much better/faster a card with 256-bus 512mb GDDR5 is compared to GDDR3?

I'm not sure if I should pick up 9800GTX+, GTX 260 or just go with 4870 that has the GDDR5.

I have an SLI motherbord. What to do?
Click to expand...


if you go over to anandtech and check out the articles related to the gt260 and the ati cards 4870 and 4850, anand can explain it rather nicely for you.
 
JimmiG said:
4870 is just under 8% faster than the 4850 at the same clocks. Since the only difference between the 4850 and 4870 (apart from the clockspeed) is the GDDR3 vs GDDR5 memory, the real world performance benefit of GDDR5 memory appears to be 8% on average. This is only one game, though. I wish someone would do a more thorough inverstigation of this.
http://www.xtremesystems.org/FORUMS/showthread.php?t=192690&page=4
Click to expand...

theres one big difference you forgot, the 4850 has a tiny fan and cooler compared to the complete dual slot cooler of the 4870, that 60mm fan on the 4850 is inefficient and LOUD compared to the 4870. yeah the 4850 is cheaper but add a good cooling system to it and all of a sudden its price is not quite as attractive.
 
there was a review somewhere where they made the effective frequencies of the memory on the 4850 and 4870 the same and also kept the core clock the same. the result was that the 4850 was faster for latency reasons. so answer:
gddr3 is better per clock, but the clock speeds that can be reached with gddr5, are much, much, much higher (did i say much?).
its uncannily analogous to ddr2 vs ddr3 for system memory.
 
grimreeferx09 said:
its uncannily analogous to ddr2 vs ddr3 for system memory.
Click to expand...

Not at all. GDDR5's speed advantage over GDDR3 is far more significant than DDR3's advantage over DDR2. GDDR3 has the ability to potentially reach speeds of over 7GHz, as opposed to GDDR3's limit of slightly above 2GHz.
 
Zero82z said:
Not at all. GDDR5's speed advantage over GDDR3 is far more significant than DDR3's advantage over DDR2. GDDR3 has the ability to potentially reach speeds of over 7GHz, as opposed to GDDR3's limit of slightly above 2GHz.
Click to expand...

well, i meant the concept of higher frequency overcoming the performance loss from higher latency.

anyway, to the op, if you have an sli board, you should definitely pick up a gtx 260. i missed your little graphics card quandry and just responded to the title before lol. the gtx 260 has a 448 bit memory bus, so the memory bandwith is similiar to the 4870's with its 3.6ghz gddr5.
 
Some company should really make a graphics card benchmark, that simply measures Graphics Memory speed with a test where you can cycle 50mb, 100mb,250mb, 500mb, 1GB data blocks or a custome size through the program, to gauge graphics cards memory performances/speed/latency. This shit is to complex to figure out with these silly articles :p
 
Again, the memory speed itself doesn't work like a CPU because what matters in the end is that the OVERALL bandwidth that is available for the GPU to render frames, and the fact is that GDDR5 + 256-bit is equivalent to GDDR3 + 512-bit right now.
 
ZerazaX said:
Again, the memory speed itself doesn't work like a CPU because what matters in the end is that the OVERALL bandwidth that is available for the GPU to render frames, and the fact is that GDDR5 + 256-bit is equivalent to GDDR3 + 512-bit right now.
Click to expand...

Yeah but if you read up, there is alot more than just memory and megahertz operating speeds as well. There is undisclosed amounts of GPU L1 & L2 cache on sum cards. With a CPU that information is readily available with a GPU they can get away with not telling apparently.

Example Anandtech Quote , Taken from the bottom of the page. http://www.anandtech.com/video/showdoc.aspx?i=3341&p=4

"Even though AMD wouldn't tell us L1 cache sizes, we had enough info left over from the R600 time frame to combine with some hints and extract the data. We have determined that RV770 has 10 distinct 16k caches. This is as opposed to the single shared 32k L1 cache on R600 and gives us a total of 160k of L1 cache. We know R600's L2 cache was 256k, and AMD told us RV770 has a larger L2 cache, but they wouldn't give us any hints to help out."

The only thing average customers know is GPU core clock, Memory Speed, and Stream Processor count if even that, lmao.:p

I'm trying to read this whole article to get a better grasp on how a Card really operates. It's not easy to understand though. :D :rolleyes:
 
DDR5 is like a race car
DDR3 is like a normal car

but what happens - 256bit is like a 2 lane highway while 512bit has 4 lanes.
so you have a race car with 2 lanes vs a normal car and 4 lanes.
either way the car gets to its destination. :p
 
Do all of the hidden details make any friggen difference? You can see the end result via FPS at given image quality settings! who cares if X card has faster memory if Y card works better? Yeah it may be fun to learn that kind of stuff to understand what is happening but it shouldnt make a lick of difference when picking a card to buy.
 
Well, let's see...

GDDR5
-GDDR3
------------
00002

2, your answer is 2.
 
I like this analogy:

4 people have to get from point A to point B, they have two choices:

The GDDR5 on a 256 bit bus is a car that can carry 2 people at 100mph.
The GDDR3 on a 512 bit bus is a car that can carry 4 people at 50mph.

Either way the 4 people get to the destination at the same time.

Just replace MPH with effective memory clock speed and you're good to go. The effective clock for GDDR5 is 4 times the base clock, GDDR3 is 2 times base clock.

Example: 4870 with GDDR5 base is 900, for 3600 effective. GTX 280 with GDDR3 base is 1107, for 2214 effective

So for a 256 bit bus GDDR5 setup to equal a 512bit bus GDDR3 setup in bandwidth you will need the same base memory clock.
 
gmkmay said:
I like this analogy:

4 people have to get from point A to point B, they have two choices:

The GDDR5 on a 256 bit bus is a car that can carry 2 people at 100mph.
The GDDR3 on a 512 bit bus is a car that can carry 4 people at 50mph.

Either way the 4 people get to the destination at the same time.

Just replace MPH with effective memory clock speed and you're good to go. The effective clock for GDDR5 is 4 times the base clock, GDDR3 is 2 times base clock.

Example: 4870 with GDDR5 base is 900, for 3600 effective. GTX 280 with GDDR3 base is 1107, for 2214 effective

So for a 256 bit bus GDDR5 setup to equal a 512bit bus GDDR3 setup in bandwidth you will need the same base memory clock.
Click to expand...

If a 4870 left Chicago at 2AM and was going at 3600MHz with GDDR5 and a GTX280 left New York going at 2400MHz with GDDR3 how long before we didn't give a shit anymore and just googled the answer?
 
gmkmay said:
I like this analogy:

4 people have to get from point A to point B, they have two choices:

The GDDR5 on a 256 bit bus is a car that can carry 2 people at 100mph.
The GDDR3 on a 512 bit bus is a car that can carry 4 people at 50mph.

Either way the 4 people get to the destination at the same time.
Click to expand...

So the first bus would be more effective as the first two people can be hard at work (or play) while the other two are on their way. Also less time on the bus too. :)
 
Jim-H said:
So the first bus would be more effective as the first two people can be hard at work (or play) while the other two are on their way. Also less time on the bus too. :)
Click to expand...

Most likely they're just waiting for their ride doing nothing productive, lazy bastards.
 
gmkmay said:
I like this analogy:

4 people have to get from point A to point B, they have two choices:

The GDDR5 on a 256 bit bus is a car that can carry 2 people at 100mph.
The GDDR3 on a 512 bit bus is a car that can carry 4 people at 50mph.

Either way the 4 people get to the destination at the same time.

Just replace MPH with effective memory clock speed and you're good to go. The effective clock for GDDR5 is 4 times the base clock, GDDR3 is 2 times base clock.

Example: 4870 with GDDR5 base is 900, for 3600 effective. GTX 280 with GDDR3 base is 1107, for 2214 effective

So for a 256 bit bus GDDR5 setup to equal a 512bit bus GDDR3 setup in bandwidth you will need the same base memory clock.
Click to expand...

but wouldn't the 256bit "car" have to make the trip back to pick up the other two people, losing precious time in the process...? the 4 people wouldn't arrive at the same time, so wouldn't the 512bit "car" then be the winner...? what if one of the cars gets a flat...? I'm confused...
 
wol-va-rine said:
but wouldn't the 256bit "car" have to make the trip back to pick up the other two people, losing precious time in the process...? the 4 people wouldn't arrive at the same time, so wouldn't the 512bit "car" then be the winner...? what if one of the cars gets a flat...? I'm confused...
Click to expand...

dude seriously you need to smoke up a bit and stop thinking too much :cool:
 
As already mentioned, memory bandwidth is the most important factor in determining performance. I'll show you how to calculate memory bandwidth. It's really easy!

Bandwidth (GB/s) = bus width (bits) * memory speed (GHz) * (1 byte / 8 bits)

GDDR5 is really great. Among other things, it can go approximately twice as fast as GDDR3. If the HD4870 and GTX260 had the same bus width, you'd be getting about twice as much memory bandwidth with the HD4870. But the GTX260 has almost twice the bus width. So the end result is that they have about the same bandwidth.
 
Kaldskryke said:
As already mentioned, memory bandwidth is the most important factor in determining performance. I'll show you how to calculate memory bandwidth. It's really easy!

Bandwidth (GB/s) = bus width (bits) * memory speed (GHz) * (1 byte / 8 bits)

GDDR5 is really great. Among other things, it can go approximately twice as fast as GDDR3. If the HD4870 and GTX260 had the same bus width, you'd be getting about twice as much memory bandwidth with the HD4870. But the GTX260 has almost twice the bus width. So the end result is that they have about the same bandwidth.
Click to expand...

DO you or anyone else know if GDDR5 has less latency than GDDR3?:confused:
 
wol-va-rine said:
but wouldn't the 256bit "car" have to make the trip back to pick up the other two people, losing precious time in the process...? the 4 people wouldn't arrive at the same time, so wouldn't the 512bit "car" then be the winner...? what if one of the cars gets a flat...? I'm confused...
Click to expand...

Assume for these purposes that the 256-bit car instantly reappears at the start to pick up the other two passengers and doesn't expend any time coming back
 
MyIllusion said:
DO you or anyone else know if GDDR5 has less latency than GDDR3?:confused:
Click to expand...

Did you check out the computerbase.de review?

The 4870 running at 4850 speeds for both core and memory was 12% slower than the 4850. Even at 750 Mhz core the 4870 with memory underclocked was STILL slower than the stock 4850. So that suggests there is significantly more latency on GDDR5, given the same data rate -- remember you have to reduce GDDR5 clock by 50% to match GDDR3's data rate.
 
You must log in or register to reply here.
Back
Top