Important: Ati 4850 Temps Too High?

[Angry_Games]

I can get more bandwidth out of GDDR3 than GDDR5 if my clocks are high enough. This is why it is limited to the memory interface bus, not the type of memory used. GDDR5 is lower latency, higher potential clocks, but again, I can "out-bandwidth" it with GDDR3 if the clocks are higher.

 

Same with DDR2 vs DDR3...I can pull massive amounts of bandwidth over DDR3 with my DDR2 depending on the clocks, because they are both running in 128-bit dual-channel mode.

 

The reason you see GDDR5 (or DDR3) with higher bandwidth is because by default the GDDR5 is a much higher clock speed.

 

On a secondary note, thinking anyone can actually use all that bandwidth is folly in the same sense that just because your hdd's claim 3.0Gb/s doesn't mean you'll get anywhere even close to it...it's just capable theoretically.

 

I've yet to see a card, hdd, memory subsystem, etc that is capable of using anything close to it's "theoretical bandwidth". Numbers are tricky, and marketing is even trickier. Getting fooled into a pure numbers game isn't for the experienced. Another example is the "contrast ratio" of LCD's. Yours say 10,000:1, yet mine only says 2500:1 and seems about 50% brighter and more colorful. So who is fibbing? Neither...it's all relative to who is presenting the numbers to you (or which program is calculating theoretical numbers based on class of hardware used...and as we all know, it isn't the size of your hardware, it's how you use it )

[Praz]

I read this a couple months back in print form but was able to find it online.

 

Bandwidth first: A system using GDDR3 memory on a 256-bit memory bus running at 1800MHz (effective DDR speed) would deliver 57.6 GB per second. Think of a GeForce 9600GT, for example. The same speed GDDR5 on the same bus would deliver 115.2 GB per second, or twice that amount. Take any GDDR3 bandwidth on a given clock rate and bus width and double it, and you get GDDR5's bandwidth.

 

http://www.semiconductor.net/articleXML/LN798369899.html

[watsonte]

I can get more bandwidth out of GDDR3 than GDDR5 if my clocks are high enough. This is why it is limited to the memory interface bus, not the type of memory used. GDDR5 is lower latency, higher potential clocks, but again, I can "out-bandwidth" it with GDDR3 if the clocks are higher.

 

Same with DDR2 vs DDR3...I can pull massive amounts of bandwidth over DDR3 with my DDR2 depending on the clocks, because they are both running in 128-bit dual-channel mode.

 

The reason you see GDDR5 (or DDR3) with higher bandwidth is because by default the GDDR5 is a much higher clock speed.

 

On a secondary note, thinking anyone can actually use all that bandwidth is folly in the same sense that just because your hdd's claim 3.0Gb/s doesn't mean you'll get anywhere even close to it...it's just capable theoretically.

 

I've yet to see a card, hdd, memory subsystem, etc that is capable of using anything close to it's "theoretical bandwidth". Numbers are tricky, and marketing is even trickier. Getting fooled into a pure numbers game isn't for the experienced. Another example is the "contrast ratio" of LCD's. Yours say 10,000:1, yet mine only says 2500:1 and seems about 50% brighter and more colorful. So who is fibbing? Neither...it's all relative to who is presenting the numbers to you (or which program is calculating theoretical numbers based on class of hardware used...and as we all know, it isn't the size of your hardware, it's how you use it )

 

 

While I understand you analogy, is it safe to say that for this specific case and application, the 4870 and its use of GDDR5 memory provides greater bandwidth and therefore greater performance, and that, based on Praz's information, could only be made up by essentially doubling (or drastically increasing) the current speeds of the 4850?

 

I mean generally speaking isn't the idea behind the use of GDDR5 to increase bandwidth and effectively double it compared to its GDDR3 counterpart?

 

More or less I am trying to understand the reasoning behind using GDDR5 memory, and why Nvidia has yet to introduce anything with GDDR5 or GDDR4 for that matter.

Edited July 18, 2008 by watsonte

[Praz]

More or less I am trying to understand the reasoning behind using GDDR5 memory, and why Nvidia has yet to introduce anything with GDDR5 or GDDR4 for that matter.

As long as nVidia is committed to their current die size there is no monetary nor performance gain to be had by moving from GDDR3.

[Puck]

As long as nVidia is committed to their current die size there is no monetary nor performance gain to be had by moving from GDDR3.

Agreed.

 

To put it in an easier way to understand it, mem bandwidth is not what is holding back their cards performance so increasing that will not give it any more grunt except in rare situations and the cherry picked situational benchmarks that we so often see in unreleased product slides.

 

This is why AMD couldnt take a first revision phenom, with its already INSANE mem bandwitdth, and add DDR3 to increase the mem bandwidth even more to try to catch up to the c2d processors from intel. The mem bandwidth was not why they were being outperformed, so more of it would not help their situation, so staying with DDR2 made perfect sense both logically and financially.

[Waco]

I can get more bandwidth out of GDDR3 than GDDR5 if my clocks are high enough.

Sure you can...but when you can get GDDR3 (which is essentially tweaked DDR2) running at 3.6 GHz+ let me know.

[Angry_Games]

Sure you can...but when you can get GDDR3 (which is essentially tweaked DDR2) running at 3.6 GHz+ let me know.

when you can get any video card's memory clocked to 3.6Ghz let ME know (and do it with proof and without doing anything out of the ordinary like having to hook it up to phase cooling etc...normal aftermarket cooling that any joe can install...ah...yeah...not really possible)

[Ste]

The 4xxx cards do not run hot at all, its only because people are afraid to turn up the fan speed that the heat issue was created..

 

I have Overclocked my X2 to 800/940 on both cores and memory, and while gamming with 80% fan speed Gpu's NEVER EVER Exceed 65C....

 

At idle, 50% fan speed, 40C..

 

Hardly a heat issue, people just don't like noise.... I don't know why...

Edited August 28, 2008 by Ste

[Puck]

The 4xxx cards do not run hot at all, its only because people are afraid to turn up the fan speed that the heat issue was created..

 

I have Overclocked my X2 to 800/940 on both cores and memory, and while gamming with 80% fan speed Gpu's NEVER EVER Exceed 65C....

 

At idle, 50% fan speed, 40C..

 

Hardly a heat issue, people just don't like noise.... I don't know why...

Huge +1!

[InCrYsIs]

the asus 4850 comes with a fan program, which i have mine set to %50. it idles at 50c

 

I have 2 Asus 4850's. Using the software for the fans works great. At 50% cards run at 55-60C playing COD4.

[Waco]

when you can get any video card's memory clocked to 3.6Ghz let ME know (and do it with proof and without doing anything out of the ordinary like having to hook it up to phase cooling etc...normal aftermarket cooling that any joe can install...ah...yeah...not really possible)

Any 4870 or 4870X2 with GDDR5 runs stock at 3.6 GHz effective. The GDDR3 4850 runs at 1.8 GHz effective on the same bus width. You'd literally have to DOUBLE the GDDR3 memory speed to get the same amount of bandwidth...

 

I'm not sure what point you're confused about?