NEW Asus G73JH - Page 2

qhn I am doing the same as you except I convert to bytes only so I get a number that is easier for me to comprehend and thus convert in my head. So I use your 1st equation with the bit to byte conversion (/8).

Here goes:

2000Mhz (1000Mhz DDR, memory frequency)x32 (256bit/8 to give bytes)=64000MB/s=64GB/s

The /8 can be done anytime so your exact 1st equation:

2000MhzX256=512000Mb. 512000Mb/8=64000MB=64GB.

I cannot plug 128bit into that equation in any way and get the correct ATI listed bandwidth, only 256bit works? And it works perfect. As I said I see marketing not understanding numbers? Unless I am the one who is missing?

It seems standard to list the memory core clock so 1000 is 2000 (DDR)? If it is not the equation is even more impossible. qhn we have to get 64GB/s memory bandwidth second link in my long post clearly states that as the bandwidth.

Any ideas? Can you make the equation work w/128? I can't. But I made it work with 256?

One can play with number of the memory clock speed (memory frequency) by using different DDR. I can take the same card packed with DDR5 and it will be more that the DDR in the example, to work out the same (256-bit) bandwidth number on the 128-bit. Interface and DDR Version and engine clock intertwine and determine the max bandwidth of the card, as I understand it.

And actually ATI HD5800 series show that with good memory clock speed (DDR5), it can push up to 153.6GB/sec of peak memory bandwidth.

Am I missing something?

cheers ...

qhn I did some reading. GDDR5 at same speed and bus has twice the bandwidth of GDDR3. So that says yes 128bit bus does seem correct.

OK that said I am still stuck on why? Why is it not GQDR (quad rate)? I know more about system memory. DDR, DDR2, DDR3 is a DRAM interface specification.

I do know why it is not called GQDR. it is still just DDR. But they have changed the naming convention as far as I can tell. DDR with 100Mhz memory clock I/O bus clock is 100Mhz X2 is DDR-200Mhz. The same 100Mhz memory clock DDR2 becomes I/O bus clock 200Mhz X2 DDR2-400Mhz. DDR3 100Mhz memory clock is 400Mhz I/O bus clock X2 DDR3-800Mhz. Same memory clock core speed is reported as the I/O bus clock then X2 to get rated speed.

Why does GDDR3 use the same standard but GDDR5 does not? Unless DDR3 has DRAM at say 1000Mhz? I do not think it does. Why does GDDR5 not report the I/O bus clock as the core? It in fact does in much marketing.

As near as I can tell that 1000Mhz core clock on GDDR5 is really 2000Mhz vs GDDR3 1000Mhz. Then you double the 2000Mhz so 4000Mhz effective.

4000MhzX128=512000Mb=64000MB=64GB.

It seems clear to me that the 1000Mhz in DDR3 and DDR5 not the same? It is one strange naming convention.

If someone can see the flaws in my thinking please teach me. I mean I can memorize but I like things to make sense and not be arbitrary. Is this just an example of techno geeks having issues with common sense or the English (or any other) language. Trust me it happens. Between marketing and the engineers they are less clear in writing their thoughts than even I am. Ever read an Intel technical document? Not an easy read and sometimes just bad English (or any other).

Please help a confused old man. And qhn how about a technical section, I have taken this thread so far off course. Sorry.

I personally feel that many users are getting used in seeing/reading that DDR2 is 2x over DDR, and DDR3 is 2x over DDR2 - right or wrong- to conclude that DDR5 "automatically" 2x over DDR3 ... or is it DDR4?

I don't see any flaws with the way you calculate, only that the factors in the equation no longer follow a "standard" pattern as one presume up until now.

DDR3 and DDR5 use the same 8n prefetch and DDR interface architecture. The difference lies in the density (mebibit) and the speed that can transport the mass.

Examples collected from the web:
- Qimonda DDR5 modules: 512MB density at
* 3.6Gbit/s
* 4.0Gbit/s and
* 4.5Gbit/s

- Samsung DDR5 modules:
* 512MB density rated at 6Gbit/s
* 1Gb density rated at 7Gbit/s (under certain configuration, this module can deliver up to 28.2 GByte/sec bandwidth, 4 times as much as it states)

- others?

Bits and pieces:
AMD was the first company to ship products using GDDR5 memory with its Radeon HD 4870 video card series, incorporating Qimonda's 512 Mb memory modules at 3.6 Gbit/s bandwidth.

cheers ...

qhn been reading and it comes down to not comparing apples to apples with the numbers they give. Why is that? I still don't know. Is it a dual DQ link? I don't know but I do know more than this morning.

I am actually more interested in seen the cards performance. I know actual clocks are better than theoretical (bus width) but GDDR5 how will the high latency affect actual performance? Vs DDR3 with half the actual clocks but twice the bus so same theoretical bandwidth and lower latency. That has my interest. I know GDDR5 will win but more a question of by how much?

I asked last week. Someday they will get us something so hang on to them. At this time they don't but just wait and save them is what I was told. every four posts you get one if you write a review a few thousand seems like I get 10 every day? I'll get you to zero!

Minus zero? If I had known I would of given one more.

Right now they are for nothing like I said. Someday they might be for something? Not joking they have not figured out what they are for. Cool it does not say -0 anymore.

Credits are planned to be used to "buy" badges and stuff for your account. You get them when you make posts and threads, and when you submit news stories and reviews. There was a glitch when we first set them up where many NBF members had negative credits. We try to fix the situation whenever we notice someone with negative credits. I just gave you a thousand credits to get you nicely back in the positive.

You can give credits to other users via your User Control Panel (UserCP) if you think they were especially helpful or posted a good question or useful info, or really for pretty much any reason.

cheers ...

Add me to the pre-order list as of this morning.

OK guys I wanna talk/ponder the 128bit bus and what it means. So I guess I have to throw out some thoughts and you guys agree or disagree and give me some insight.

Memory bandwidth is a combination of speed and bus. On system memory I know that running 128bit bus (dual channel) theoretically doubles bandwidth but in reality only increases around 10%.

So what does that mean with 5870? Well with system memory speed is more "real", bus is more "theoretical". I suspect that GPU's might make better use of the bus to get closer to that theoretical. I suspect that because of the ability to do greater parallel processing. But that said let us get back to the one core reality. Bandwidth is speed and bus. The 5870m has 64GB the GTX 280m has 60.8GB. GTX would have 30.4GB on 128bit bus. If the bus on the GPU suffers as the system does from bus being more theoretical than actual the 5870m has major advantage vs GTX. Even if GPU bus is 100% efficient (it can't be) the 5870m still has advantage as 64GB vs 60.8GB.

So I do not see the 128bit as being an issue vs current GPU's even with 256bit bus. Yes a 256bit would improve performance vs itself and 128bit. But it has a greater bandwidth with a 128bit than the GTX 280m with a 256bit bus.

Hypothetical here. I am almost certain that if two GPU's have the same bandwidth and one is 128bit the other is 256bit the 128bit will outperform the 256bit. The reason is more of the bandwidth in the 128bit comes from "real" speed not "theoretical" bus.

Edit: No one throw in the increased latency of GDDR5 vs GDDR3 as I have not accounted for that yet. And that could be a very important thing to account for?

GDDR5 is a huge thing needs to be accounted for. I think Asus or ATI came out with a statement regarding this I need to find it. By the way a single 5870 scored 7K GPU P in Vantage while a single 4870 scored 6K.

Not sure I understand the first sentence? I mean the GDDR5 at least as far as bandwidth is accounted for? I mean it's effective clocks and bus give it 64GB? Desktop faster clocked GDDR5 but if it was the same with 256bit bus would be 128GB bandwidth.

It is clear your HD4870 hangs as close as it does because the GDDR3 while slower has much less latency. And does have 256bit bus.

The mobile HD4870m has 56.8 bandwidth. 888Mhz memory (effective 1776 GDDR3))X8 (bytes) X4 (256 (4X64))=56.8GB.

HD5870m is only 64GB.

If you do the math on the Vantage scores you gave, 6000 vs 7000. The bandwidth increase is 12% (rounded) but the performance increase Vantage is 17% (rounded)?

What does this mean? It means I don't have a clue but I can do basic math.

It's also possible that the 5870 hardware is better optimized for the benchmark or that the scores were taken with different driver versions. There are a lot of potential variables.

As long as my basic math skills are not called into question I am good. The long and the short is while I would like the 5870m to have 256bit bus for reasons stated. It does not matter. It is what it is. But further at least in my recent research and thoughts. The predetermined idea that 256bit is somehow better than 128bit is bunk. Many like to say the bus is somehow critical, almost magical? It is not and the 5870 basically proves it. The bus is just a means to an end same as clocks are. I think a lot of people that I have read need to rethink their common sense. And these are not people on this site per se.

Edit: T-21:07