explain the 7700 stats

I think this chart is really weird.

In reality, the 7700 is exactly like the 7600 except it has .08 micron and 12 pixel shaders.

Ok yet its scores are exactly the same.

Its core and memory are the same.

Chinese 3dmark scores are hard to compare because it uses the 7200 cpu vs the 1.83 t2400 and its score can actually be almost entirely the cpu difference.

What is the missing factor? According to the stats it is not underclocked or different in any way. So it should be about 150% faster, why isnt it?

 

ok Im not going to argue with you.

mostly because I have no idea what you are talking about.

define supply voltage gate length and and vdd

You are seeing ok, same core and clock speed running through 12 pixel shaders but its score is about lets call it 5% more than the 7600 or possibly the same were going on chinese numbers ill try to dig up to show you.

 

The whole point of moving to a smaller etching process is to decrease the VDD so the chip generates less heat and consumes less power -but I fail to see how that would decrease the performance of the part since the processing power of the GPU is determined by the number of pipelines and the frequency of the GPU. Memory clock rate can also influence performance if data can't move fast enough to keep up with the GPU -it's called "memory starving".

Gate length is only one of many factors that determine how quickly a transistor can switch on or off -and moving to a smaller etching process isn't going to make anything on the chip larger...

FWIW: The chart is wrong. The fill rate of the 7700Go is 5.4 billion pixels per second and not 3.6 (figures available here http://www.nvidia.com/page/go_7600.html ). I knew it was inaccurate before I looked it up, since the fill rate is clock speed X number of pipelines -the 7600Go is an 8 pipeline GPU, and the 7700Go has 12.

 

ok well that would definitely explain it then. the chart is incorrect.

ok well something is still wrong if 3dmark scores are in the 5% improvement range.
Why is the 7700 not competing with the 7600 gt where is it lacking if its stats are the same? Better even they are better than the 7600 gt.

 

Two things that popped into my head. The 7600GT is clocked faster and uses DDR3 RAM -so it processes pixels faster and has more bandwidth to its frame buffer to keep from getting memory starved. The 7700Go is clocked like the 7600Go and uses the same DDR2 RAM -so the only performance difference is the 4 extra pixel pipelines in the 7700Go. It's possible that a bench mark like 3Dmark won't be able to take advantage of the difference between the 7600 and the 7700, but I think you'll see a difference in games when you enable AA and AF -the 7700Go will take less of a performance hit. Keep in mind that 3Dmark is just a benchmark program and it cannot accurately reflect how well a GPU will do in a game...

 

The 7600GT has significantly faster core and memory clocks - 580Mhz and 1500Mhz.

 

Yup -and that's all it is (if anything that you wrote is even remotely true -you skipped over quite a bit, and some of what you wrote is, with all due respect, gibberish. You write like you are either trying to sound smart, or a first year EE student who needs to hit the books a little more...).

Only partially true. The length of the gate, and the speed at which the gate can forward bias (turn on), the resistance to the flow of current through the gate (some materials conduct better than others) also determines the maximum frequency that the circuit can handle. Since the etching process is smaller the length of all circuits is reduced. By your post there should be a maximum speed limit for a LSIC -and new etching processes and materials are developed every day. Intel will be moving to a 45nm process in 2007/2008 with CPUs that will clock well into the 2Ghz range. Maybe you should explain your theory to them -obviously they don't know...

FWIW: I think IBM developed a transistor that could turn on and off faster than an electron could move across the gate. Not worth much, since it's kinda hard to gate a signal that never makes it out of the transistor -one of the many things that IBM engineers did just because they could...

 

And there is the answer -and it didn't take an EE degree to figure it out...

 

ok

If it were the same core same memory

going through the 12 pipes.

Wouldnt it overclock exactly the same mhz with stability?

And wouldnt each 1 mhz make the speed go up as much as 1.5 mhz would in a 7600.

Can you explain a factor Im missing anyone?

 

You asked about the Go7700 vs the desktop 7600GT. The 7600GT desktop part has 12 pipelines just like the 7700 does.

 

oh no im talking about 3 mobile gpus the 7600 go, the 7600 gt and the 7700 go.

the 7600 gt is not a lot different than the desktop version though.

 

In an earlier post you said "Well, frequency is inversely proportional to period". So I went to Google and dug up this -I knew it before I looked it up, but I like to back up what I have to say with references:

From http://tf.nist.gov/general/enc-p.htm

"The period T is the reciprocal of a frequency, T = 1/f. The period of a waveform is the time required for one complete cycle of the wave to occur."

So frequency is not inversely proportional to period -so what were you saying about not knowing facts?

Now I'm the one who's done here -clearly you've added nothing to this thread...

 

My only guess here is that the 7700Go won't overclock to the performance level of the 7600GT because of the DDR2 memory that the 7700Go uses (the 7600GT uses DDR3). I think that the 7700Go reaches a point where it's sitting there, wasting clock cycles, waiting for data to move in and out of its frame buffer.

 

CMIIW... if T = 1/f, then f = 1/T, and does it mean that the inverse of T is f?

 

That seems to be the common consensus from people with the go7600. Overclocking core yields little to no benefit, whereas overclocking memory give the required boost. The 7600/7700 is a nice GPU but it's memory bandwidth limited.

 

Yes it does -but I was hoping match000 would come back with the answer. From his post I wasn't sure if he knew what he was talking about and I wanted to see if he was going to correct my post.

His explanation for the performance of the 7700Go is still incorrect. The performance of the GPU is limited by the speed of its frame buffer -VDD has absolutely nothing to do with it...

 

Here's the part that puzzles me: Is the 7700Go a new GPU, or is it a 7600GT with DDR2 RAM?...

 

OK... sorry.. don't mean to interrupt anything..

 

No harm, no foul

 

...and people who try to impress others with what little they know, and who try to win arguments through character assassination, aren't worth talking to...

/Ignore

P.S. What's really funny is that I had a chance to quote your post before you editted out calling me stupid...

 

um

well lol what it is exactly is the 7600 go with 12 pipelines.

I guess that is also like the 7600 gt with slower core and memory. they are all alike. I wish I could say this has been helpful but it hasnt.

I cant wait to see the actual performance of the thing though. I am looking more for the 5000 3dmark 05 range not the 4200 4300 that i saw in the chinese review.

 

3Dmark may be the problem stamar: I'm not sure you're going to see a big jump in performance with the standard test suite. But you might see the difference that those 4 extra pipelines can make if you can run the benchmark with AA and AF enabled...

One thing I'm going to do is run a game with and without AA and AF enabled and look at the difference in frame rates. My guess is that the performance difference will be negligible -in fact I think I read somewhere that it has already been done and the guy who did it said that there wasn't much difference. I'll have to see it for myself though...

 

So is there a real advantage for the 7700 in terms of performance? Besides AA and AF ...

 

I don't know. Maybe less heat and power consumption if it's running at a lower voltage than the 7600. I've seen some 3Dmark05 scores that do place the 7700Go at about 30% faster than the 7600Go -but I still need to get the A8Js and test it for myself.

Edit: I just Googled for benchmarks and got to this site: http://www.notebookcheck.net/Mobile-Graphics-Cards-Benchmark-List.844.0.html


3DMark06 results
7600Go 1900
7700Go 2700

And then in this post: http://www.notebookforums.com/post2426897.html the A8Js scores 2668 in 3Dmark06 -the scores are close enough for statistical error (meaning that they are probably accurate).

 

If I remember it correctly, one member here has the A8Js, someone from China.

 

Dalantech, there's nothing you posted that I disagree with (besides your little trap ), but I think you should give match a little slack. I don't doubt the accuracy of his equations as I'm sure he either is currently taking a course in idealized circuit or referenced a textbook, but I doubt he fully understands the ramifications of using an idealized equation in practice (no offense match).
Anyway, I think he was just trying to help...even if he was trying to sound smart...btw, I think he's more like a 2nd or 3rd year EE student.

Match, while your equations are probably accurate, your assumptions in this case are obviously wrong because the specifications of the gpu say that it runs at the same clock speed as the 7600. In this case, clock speed is dictated by a clock generator on the chip and influenced by VDD. In other words, the thing will be clocked at a certain speed and if the voltage isn't high enough, it won't operate properly.
Your equations are for idealized circuits with fixed resistance and power consumption but the fact of the matter is that the ideal and the real world are different beasts altogether. Remember that resistance in a circuit is influenced by temperature...not to mention tunneling (current leakage for you EE people), impurities in the silicon, and a ton more variables that I'm sure I've left out.
IC manufacturers overvolt their products to raise yeilds, which is why many people can undervolt their processors to reduce heat and save power. The basic V=IR and P=I^2R equations can be applied just to understand what's going on but you can't draw any concrete numbers without expanding those equations out to include margins of error...and even then it's only an approximation.

Anyway, my 2 cents...

 

that most recent chart, it greatly underestimates the scores of the 7600 go.

it also uses the exact numbers from the chinese article meaning it just copied over the scores, nm the cpu used.

Anyhow, its actually entirely worthless. I mean if you consider the 3dmark scores are worthless to start with, how useful is it to compare them and then use systems with different cpus, ram etc.

Thank you for the chart though. But its actual data is barely worthwhile enough to actually rank the cards in order.

I mean it is actually worthwhile to rank the cards plus or minus about 3 spots either way. It lists the core and pipeline information though.

kateflics,
if the chinese data score is correct which it cant be, the 3dmark score 06 is a huge improvement. I think it will be a improvement but how I cant say. If I learned anything from reviewing the information I have had, I learned that I know absolutely nothing about its speed and all data is totally unreliable to this date.