Compal IFL90 8600GT 512mb benchmark

Hi all, Im just wondering how the 512mb version compares to the 256, wether or not theres any real difference in performance.

 

If it uses GDDR2 like the Zepto 6625WD or 6224W it will performe worse then the GDDR3 version with 256MB like in the Asus G1s.
If it has 512MB GDDR3 then it will be slightly better then the G1s.

Zepto said their IFL90 uses GDDR2.

 

wave, I think you will find that you cannot support the position that GDDR3 will give you any noticeable performance boost over GDDR2.

See http://forum.notebookreview.com/showpost.php?p=2014696&postcount=108 for example, and then do other searches to find that real world performance difference is simply insignificant.

 

The difference will depend on the game. In newer games, and games yet to be released you will definitely notice a difference with 512MB with the new DirectX 10 cards.

 

Well GDDR2 and GDDR3 doesnt matter but the clock speeds matters and the GDDR2 in the Zepto runs at 400mhz and the GDDR3 in the Asus at 700mhz. If the GDDR2 would run at 700mhz it wouldnt make a differance but since it is slower it does. And the real world benchmarks i saw so far do show it.

 

I agree 100% - Donald, do you think wave's comment fair? Also Donald, what would be really helpful is if you could suggest an objectively determinable 'real world' test so we can compare G1S (700mhz VRAM) performance against GDDR2-based laptops running memory at 400mhz?

Petrov.

 

Core/Clock speeds will definitely be more important than GDDR2 or GDDR3.

However I wish wave would have been more precise in his statement and could back up the Core/Clock speeds for each model with an authoritative link showing what both are for each model.

Now what will REALLY be important is the amount of on-die memory for each model. The IFL90 has 512MB on-die while the G1S has 256MB and only gets to 512MB by the pretty slow Turbocache using system memory for the other 256MB.

We really need to wait to see a comparison of synthetic benchmarks for both models, especially once the IFL90 is properly overclocked.

 

Oh, and "real world benchmarks" is an oxymoron

It is either "real world" or it is "synthetic benchmarks".

 

Petrov, you have been around these forums long enough to know this subject has been discussed NUMEROUS times and the conclusion is there is no way to do "real world" testing...that too is an oxymoron. What is true is that synthetic benchmarks are an indicator of performance, but one should not rely on them completely to judge real world performance.

People can get anal about these synthetic benchmark scores if they like, but the saavy geek knows too much to be that impressed with them.

 

Haha - thanks (I think?). From my perspective, the attraction of synthetic benchmarks is that they are one of the few ways us nerds can actually debate the pros/cons of some technologies - I'd love to reference real world performance, but its so hard to do, so we fall back onto synthetics.

I understand your desire to be precise about GDDR2 v GDDR3 being less at issue than 400mhz v 700mhz clocked memory - but so far we've only seen GDDR3 memory clocked at 700mhz and GDDR2 at 400mhz (see the Zepto and G1S as the two examples), so that's all people are going on. *If* we can get a 700mhz (or near enough) clocked GDDR2, then we can really understand what difference is dictated by clock speeds, and what by type (and no doubt we'll see that clock speed is dominant).

Your comment about seeing an IFL90 overclocked is also well taken - but I think we should compare stock speeds to stock speeds across laptops, and overclocked against overclocked speeds across laptops (ie, you can overclock the G1S GDDR as well, quite a lot, as Ken shows in his threads on the Asus boards, but the same exercise has not been undertaken afaik for GDDR2 based laptops, just due to supply).

Finally, I am interested to see (holding everything else constant), of 512mb vs 256 mb makes any difference to either synthetics or real world. What specifically makes you think it'll be a noticeable difference? On previous generation 7700's it didn't seem to make much difference?

Thanks again for your feedback Donald!

Petrov.

 

From what I've seen of Asus, they state their clock at the double-pumped effective rate, and not the actual clock rate, whereas Compal states the actual clock rate.

So a 700MHz GDDR3 clock for Asus is really GDDR3 clocked at 350MHz, and a 400MHz GDDR2 clock for Compal is really GDDR2 clocked at 400MHz, which when put into Asus' terms would be 800MHz GDDR2. Makes a lot more sense that way... there's no way Asus could release RAM that's almost double the speed of Compal's in a mobile form factor without making some SERIOUS power and heat concessions, as well as the fact that GDDR3 typically runs at a slower actual clock rate than GDDR2 anyway.

Am I wrong in this? Does anyone have a G1 or an older Asus that we can compare actual clock rates to advertised ones? I know that the A8JM was often advertised as having 700MHz GDDR2 RAM, but it really ran at 350MHz, the 700MHz is it's effective rate when doubled.

 

I'd rather have 512mb dedicated GDDR2 than 256mb dedicated GDDR3.

As we've seen with the 8-series memory gets eaten up like crazy (see 320mb 8800 vs 640mb 8800), moreso than the 7-series. Also the performance is relatively the same between the two so your not exactly taking a hit (both ~3900 3dmark06). The only difference is 512mb vs 256mb is ~7-8% faster and the higher memory speeds of GDDR3 relatively make up for the lesser dedicated memory.

So relatively similar performance yet twice the dedicated memory? It would be a no brainer for me if I was getting one.

Example - World in Conflict beta settings:
Despite running the game at very good FPS, the 256mb 8600 could only play medium while the 512mb 8600 could play high. Yes I know bad example because of poor memory optimization but its still the case. People with 7900gt and 7950gt had the same issue despite them being relatively similar cards (7900gt: 45fps medium, 12fps high ; 7950gt: 50fps medium, 39fps high)[drop of -33 compared to -11 = big diff for similar cards).

And its never fun seeing a memory buffer overload of 190mb in the corner of your screen believe me

 

I cant speak for the IFL90 because I have not seen it. Zepto said their IFL90 (with a 8600M GT and a 1280x800 screen) uses the same GPU, and clockspeeds as the Zepto 6224W and Zepto 6625WD, a 8600M GT with 512MB GDDR2 running at:
core ROP: 513mhz
core Shaders: 1026mhz
memory: 400mhz

This are the number the RV Tune shows. I am not sure if it is the actual or double pumped. But for the Macbook pro and the G1s it shows a memory speed of 700mhz in RV tune so I think it is comparable.

 

The early 8800 GTS 320MB benchmarks showed worse performance based on a driver bug, new drivers fixed most of those problems so unless you are running over 1600X1200 resolution your aren't going to see much of a performance difference between the 320 and 640MB cards. As far as the 8600s in the notebooks it all boils down to clock speed both core and memory as to who is faster, most game developers down program for super high texture memory usage cause then they would eliminate most of their customer base. Coupled with the relatively low resolution of 1280X800 you don't need alot of your video ram for frame buffer.

 

Interesting 'real world' feedback on impact of system RAM on gaming performance on the latest 8600GTs:

http://forum.notebookreview.com/showthread.php?t=125246&page=38

Petrov.

 

arent we talking about VRAM?

 

The link I posted above is about the impact of 2gb v 3gb *system* RAM (not VRAM).

Petrov.

 

Great! Now I need to rethink my plan of putting in 1+1 gig to start with

 

So how is that relevant here?

 

Well maybe it ran slower because the system was trying to use Turbo Memory.
Tim

 

In this particular thread? I thought it was relevant because it was suggesting that maybe system RAM might have a bigger impact in real world performance than previously thought - is this not interesting?

Petrov.