Jetway 8800GT uses Samsung 1.0ns GDDR3 !!!

Printable View

12-24-2007, 06:46 AM
alexio

Jetway 8800GT uses Samsung 1.0ns GDDR3 !!!

I bought a Jetway 8800GT 512mb 600/1500/1800 about a week ago and after removing the stock cooler 5 minutes ago I found some nice BJ1A (1.0ns) Samsung ICs to be installed on the card.

This card uses the Nvidia reference design, so It's very likely that more 8800GTs and 8800GTSs will carry these ICs in the near future.

I think I can bench this memory at +-DDR2200 which is much better than the average Qimonda ;)

I will post some screenies later :)
12-24-2007, 06:52 AM
Pacha

finally! pray for the other manufacturers to follow and put gaymonda in the rubbish bags...
12-24-2007, 07:41 AM
david901101

Good news!

Now if any other manufacturers starts to follow Jetway's example I will return my eVGA and get one of the new cards with samsung memory. :up:
12-24-2007, 08:02 AM
kiwi

I must have exceptional Qimonda since I can bench @ 2200Mhz on air and stock volts :)
12-24-2007, 10:37 AM
VETDRMS

Qimonda does not like the stock 2.000v on the 8800GT/GTS. At 1.900v my GTS does 1134 with Qimonda 1.4ns timings. I was a Qimonda hater too until I lowered vMEM. Now if it had -08 chips, yeah, that'd be cool.
12-24-2007, 11:31 AM
CERO

did u use a hardmod to lower your Mem v?
12-24-2007, 12:25 PM
Truckchase!

Quote:

Originally Posted by alexio

I bought a Jetway 8800GT 512mb 600/1500/1800 about a week ago and after removing the stock cooler 5 minutes ago I found some nice BJ1A (1.0ns) Samsung ICs to be installed on the card.

This card uses the Nvidia reference design, so It's very likely that more 8800GTs and 8800GTSs will carry these ICs in the near future.

I think I can bench this memory at +-DDR2200 which is much better than the average Qimonda ;)

I will post some screenies later :)

I've heard the 2k Mhz limitation is in the architecture, not the RAM. I guess people will find out if after running them higher than that for awhile they start to flake out. Let's hope all that is just BS.
12-24-2007, 01:28 PM
Jaco

Quote:

Originally Posted by VETDRMS

Qimonda does not like the stock 2.000v on the 8800GT/GTS. At 1.900v my GTS does 1134 with Qimonda 1.4ns timings. I was a Qimonda hater too until I lowered vMEM. Now if it had -08 chips, yeah, that'd be cool.

so , lower voltage and tighter timings gives higher Mhz . Why does the memory run 2.0v stock then :confused:
12-24-2007, 02:03 PM
LordEC911

Quote:

Originally Posted by [XC] Jaco

so , lower voltage and tighter timings gives higher Mhz . Why does the memory run 2.0v stock then :confused:

I believe 1.4ns is looser timings, though I might be wrong.
I assume that the lower the latency the tighter the timings, stock Qimonda-10 is 1.0ns.

If you are using the stock HS, I believe the full coverage HS gets too hot from the GPU core and the ICs don't like it too much.
12-24-2007, 02:33 PM
Pacha

let us know how your samsung chips perform!
12-24-2007, 02:40 PM
VETDRMS

Quote:

Originally Posted by Pacha

let us know how your samsung chips perform!

Yes I am using a hard mod to reduce vmem.

1.4ns timings are tighter than the stock 1.0ns timings. I am going to tighten the timings further to a theoretical 1.5ns spec and see what the results are. Give me 10 min and I'll post a loose/tight timing 3Dmark06/05 score.

I won't hi-jack this thread, but tighter timings netted a very minimal increase in 3DMark05. 23244 v 23173. 3DMark05 would not complete at 1144 with the default timings, so yeah, tightened timings --> more mhz.
12-24-2007, 11:56 PM
halo112358

1.4ns is not tighter than 1.0ns, it's looser - that's why you're able to get more mhz out of your ram. There could be more to it than that, if so I'd be happy to hear about it.
12-25-2007, 02:56 AM
hecktic

now if only they could add a 384bit memory busy...
12-25-2007, 02:56 AM
alexio

Quote:

Originally Posted by halo112358

1.4ns is not tighter than 1.0ns, it's looser - that's why you're able to get more mhz out of your ram. There could be more to it than that, if so I'd be happy to hear about it.

What gave him a boost in overclock was the reduced voltage. 1.4ns timings ARE tighter than 1.0ns timings (1.4ns = DDR1400, 1.0ns = DDR2000). Apparently the tighter timings didn't decrease his overclock by much, but they do increase performance/mhz.

Even when GDDR3 chips run cool (and probably GDDR4 chips too) they have an optimum voltage to run at. Sometimes, like in this case, it is below stock.

This is just a very quick test with the card (I don't have much time now):

http://img.photobucket.com/albums/v3...2Dec251151.jpg

This means +-DDR2200 should be benchable. Maybe I'll do some hard mods to the card and play with vmem :)
12-25-2007, 07:02 AM
david901101

Quote:

Originally Posted by alexio

What gave him a boost in overclock was the reduced voltage. 1.4ns timings ARE tighter than 1.0ns timings (1.4ns = DDR1400, 1.0ns = DDR2000). Apparently the tighter timings didn't decrease his overclock by much, but they do increase performance/mhz.

Even when GDDR3 chips run cool (and probably GDDR4 chips too) they have an optimum voltage to run at. Sometimes, like in this case, it is below stock.

This is just a very quick test with the card (I don't have much time now):

http://img.photobucket.com/albums/v3...2Dec251151.jpg

This means +-DDR2200 should be benchable. Maybe I'll do some hard mods to the card and play with vmem :)

Looks great so far :up:.
12-25-2007, 09:31 AM
VETDRMS

The tighter timings yield a 10mhz increase in clock speed with no errors in ATITool with NO VOLTAGE CHANGE. And yes, 1.4ns timings are tighter as they are used in a lower speed-binned part (700mhz).

Here is the spec sheet: http://www.qimonda.com/download.jsp?...rev110_www.pdf
12-25-2007, 10:49 AM
Jochenp

I don't get it? 1ns looser than 1.4? What am I missing?
12-25-2007, 11:02 AM
sam400

with my 8800 gt jetway i can play at UT 3 up to 2240 i don't test over
12-25-2007, 11:31 AM
alexio

Quote:

Originally Posted by Jochenp

I don't get it? 1ns looser than 1.4? What am I missing?

1.0ns and 1.4ns stand for 1Hz (clock cycle) every 1.0 and 1.4 nano seconds repectively. 1MHz = 1 million Herz

Just taking away the same amount of zeros on both sides you can say that 1.0ns means the chips can do 1000MHz in one second; this is DDR2000. 1.4ns means it can do 1000MHz in 1.4s. 1000/1.4 = 714mhz aka DDR1429 (rounded correctly).

In plain language: 1.4ns chips are rated to run slower than 1.0ns chips. Because the chips are of the same design (just a different binning) the slower chips get tighter timing ratings from the manufacturer. It's just like with desktop RAM were companies sell DDR3-1333 at tighter timings than DDR3-1800.

The fact that the guy manages to overclock further with the tighter timings is rather special. It's very hard to explain really, but there are many things that could be causing this. One scenario (a pessimistic one) is that with the tigher timings more artifacts apair making the card start to correct itself (thus lowerring performance) while the viewer can't observe any artifacts. Another scenario is that the memory controller just handles the tighter timings better for whatever reason.

These things can only be explained by the people at Samsung that designed these chips or the people at Nvidia that designed the memory controller.

God I need another beer :rofl:
12-25-2007, 12:52 PM
VETDRMS

Thanks for explaining that alexio. The tighter timings net 70-100 points in 3DMark05, which is not much.
12-25-2007, 01:36 PM
Frank M

Quote:

Originally Posted by alexio

This is just a very quick test with the card (I don't have much time now):

This means +-DDR2200 should be benchable. Maybe I'll do some hard mods to the card and play with vmem :)

Only tests will tell.
E.g. my 7900GS with 1.4 Hynix is artifact-free in AtiTool even over 800MHz.
3dmarks run without a problem at over 750.
However, it's game stable only at 725. At 750 it artifacts and locks.

I know that we all know this, but this is a good example that AtiTool artifact-
free != 3dmark stable != game stable. (Unfortunately...)
12-25-2007, 01:53 PM
alexio

Quote:

Originally Posted by Frank M

Only tests will tell.
E.g. my 7900GS with 1.4 Hynix is artifact-free in AtiTool even over 800MHz.
3dmarks run without a problem at over 750.
However, it's game stable only at 725. At 750 it artifacts and locks.

I know that we all know this, but this is a good example that AtiTool artifact-
free != 3dmark stable != game stable. (Unfortunately...)

I've played Crysis for some hours at 1052mhz, so it's pretty good :)
12-26-2007, 12:28 PM
jonnyGURU

Quote:

Originally Posted by Truckchase!

I've heard the 2k Mhz limitation is in the architecture, not the RAM. I guess people will find out if after running them higher than that for awhile they start to flake out. Let's hope all that is just BS.

Yup.
12-27-2007, 08:13 AM
WeStSiDePLaYa

where are the 1.4ns timings?
12-27-2007, 08:59 AM
largon

Quote:

Originally Posted by WeStSiDePLaYa

where are the 1.4ns timings?

->

Quote:

Originally Posted by VETDRMS

Here is the spec sheet: http://www.qimonda.com/download.jsp?...rev110_www.pdf

See pages 34-36.
12-29-2007, 08:02 AM
alexio

I'll be doing some work to identify more timings in the bios than Nibitor supports. I've discovered the CAS latency bit before on another card and will do some work to find it again together with some more timings.

Manual HEX editing is needed for changing these timings though ;)
01-06-2008, 11:07 AM
lowfat

I pulled the HSF from my Rev1 Galaxy 8800GT, it is using samsung ram. I am 24/7 with a memory clock of 1110. :) Don't know high it will bench to though.
01-14-2008, 08:07 AM
jae215

Quote:

Originally Posted by VETDRMS

Qimonda does not like the stock 2.000v on the 8800GT/GTS. At 1.900v my GTS does 1134 with Qimonda 1.4ns timings. I was a Qimonda hater too until I lowered vMEM. Now if it had -08 chips, yeah, that'd be cool.

Do you have a link for the mod to lower vMEM, rather than increase it?:confused: