It's just a characteristic of this IC...Quote:
Originally Posted by G.S Technician
Printable View
It's just a characteristic of this IC...Quote:
Originally Posted by G.S Technician
i am going to ask G.Skill about playing games during my office hour since it is some test to see the memory is stable or not. or add into RMA process. Man i can only do 1 RMA per day..... :)
back to topic, if all kinds test passed but why games won't.... becasue more transfer between CPU and RAM, or why? there must be something, let discover this together
Apart from the obvious possible cause of more heat (due to graphics cards active), there are two more possible reasons:Quote:
Originally Posted by G.S Technician
- WIth tight timings it might make a difference whether the CPU reads/write data for its own consumption or just for transfer via AGP or PCIe. Obviously, the overall timing of these transfers is fundamentally different.
- Prime95 and SuperPi only test a small part of the RAM. Memtest86 tests all RAM but doesn't load the CPU. It is possible that CE-6 sticks tend to have larger areas where they go higher with loaded CPU, but other areas, not tested by Prime, don't go as high.
Theory number two could be verified by using Prime95 under different OSes. There are versions running on Linux and FreeBSD. These should end up loading different areas.
I would say the dropping of 3dmark scores as they near instability is the exact same behavior I am seeing, very similar to a marginal GPU OC. Possibly a result of indiviual chips on the stick dropping off. The mushkin however runs siginificantly faster and has a very clear wall where all the chips max out together. I would guess this is indicitive of more stringent testing and binning at the component level for these higher priced sticks.Quote:
Originally Posted by Manol
However, as a generalization, both sets seem to fail in 3D before anything else. Quake 4 in ultra quality, BF2, FEAR, set to go way over the 1 GB boundry seem to be good tests. I've seen a marginal speed low score 3dmark05 run fail instantly in Q4 ultra quality, and run fine in high quality.
Perhaps it's a flaw in the module PCB layout that is exploited in particular with these infineon chips and very large processes.
I think 2nd one is more reasonable, :)Quote:
Originally Posted by Martin Cracauer
for prime95/sp2004 if using blend which loaded memory and test both CPU
and memory, well that test help?
actually for windows memtest, both CPU and memory load i believe. i remember if CPU is not stable then memtest fail either.
Unfortunately there seems to be no way to make prime95 work over all of physical RAM.
Even if you changed prime95's source code, the OS will prevent you from nailing down to certain locations in physical memory because userlevel programs only have virtual memory. And even if you forced it to do that, then you still face areas where physical memory is permanently allocated to non-movable data such as OS-internal data - and sections that are used for intermediate storage of data exchanged with the video card.
However, if you prime95 under Windoze, Linux and FreeBSD you are almost guaranteed to test at least three different sections.
Best would be to make a self-booting prime95 CD based on MS-DOS. Then you only have physical memory and you can do each iteration of the prime95 tests on a different area.
It might be possible to re-write prime95 as a Linux or FreeBSD kernel module and do a self-booting CD with that. That way it could access physical memory, too, but you wouldn't have the hassle of dealing with MS-DOS' braindead memory allocation.
thx, i learn a lot today!Quote:
Originally Posted by Martin Cracauer
but if we open more prim95/sp2004, two blend or custom large test with custom phyical memory size, will that possible to run most the memory. but just like you said, heap, stack, and OS area will not be touch. it is hard to do but if 3 OS prime95 testing... well..
but i am just startig learning Linux and FreeBSD now, noob lol. but prime95 under linux and xBSD will testing different area? i will try it later, since install linux is not like windows, xBSD takes even more time for me... :)
With 2GB physical memory, P95 or SP2004 and 32-bit Windows you can allocate and test up to:Quote:
Originally Posted by Martin Cracauer
(a) ~1.80GB under Server2003,
(b) ~1.50GB under Windows XP (P95 v24.6, but no SP2004 based on P95 v23.8)
But that is virtual memory. You have no way to know what physical memory the OS chooses to back up the pages that the userland program uses.Quote:
Originally Posted by bachus_anonym
So in the end? Whats the deal with the chips being everything else stable but not 3D anything stable?
Is it a case of bad luck, something we can change? Or can we get the chips replaced?
Thanks
Manol
I was swiching memory around and forgot what I had in. I set it to 2.5-3-2 and had to back down the MHz to 240. That is the highest my best set of zx will do at 2.5. Not worth it though. 3-3-2@268 beats the pants off of 2.5-3-2@240.:D
I bought the 2GBZX kit for my new gaming rig, did I make a big mistake? Or is it possible to clock my Opteron 165 to a reasonable speed with this kit, and still have stability in games?
I started shaking when I read about the instability in 3D/games :(.
Running closer to spec with a divider should be just fine and won't hurt performance noticably.
I can't even boot into Windows running this RAM at 200mhz 1:1 at 2-3-2, it will boot ok at 3-4-4 though, I'm feeling seriously ripped off to say the least!
I have Patriot TCCD that does >267mhz at 2.5-3-3
My system is Epox 9nda3+, Antec Neopower 480 PSU, Opteron 165
I'm putting in for an RMA request. Hey GS Tech, will they just send my RAM back to me if it works at stock on their test computer?
From the looks of recent results on this forum, including mine, I think the RAM in these chips has been rediculously unpredictable, where early performance was awesome, now people are barely running it over stock, probably they switched from using different chips but the labels on all the products are the same.
Did you try a full CMOS reset?Quote:
Originally Posted by swiharta
But please note that the Mushkin set with the same chips has precisely the same problem. Early ones were doing well, later ones don't.Quote:
From the looks of recent results on this forum, including mine, I think the RAM in these chips has been rediculously unpredictable, where early performance was awesome, now people are barely running it over stock, probably they switched from using different chips but the labels on all the products are the same.
Can't say for OCZ because they were "lucky" not to get any chips and the EB kit is sold out everywhere for a long time now. But they seem to have missed a bad shipment of these chips.
Or OCZ received a shipment and refused the chips after testing :)
lol, if OCZ really did that, much propz to them hehe.Quote:
Originally Posted by Martin Cracauer
Perhaps that is why OCZ platinum EB CE-6 are so rare and the other modules containing CE-6's are so plentiful?
actually, how many ppl in here has the 3D unstable problem? i had ask around ppl who has 2GBZX and non of them had found this problem. however, they don't really overclock to all the way max, some of them not even overclock lol. over all 2GBZX is a great product to me. if you guys experience 3D unstable, try to lower ZX's speed. let me know i am always here.
It seems the chip type is encoded into the part number?
F1-3200PHU2-2GBZX = Hynix
F1-3200BIU2-2GBHX = Infineon
F1-3200USU2-2GBHS = Samsung
No, HZ and HS are UCCC.Quote:
Originally Posted by RaZ
The ZX is probably the Infineon.
Nobody ever figured out what the HX is, except we know the chips are expensive. COuld be Hynix. The same chips are probably on the Value NG set, although I get the value kit gets any leftover chip once in a while.
What we need is a photo database of the chip's edges.
Yes, Samsung UCCC?Quote:
Originally Posted by Martin Cracauer
F1-3200USU2-2GBHS = Samsung UCCC
F1-4000USU2-2GBHZ = Samsung UCCC
The 9th character of the part number would be denoting the chip manufacturer.
But we are pretty sure that the ZX is G.Skill's Infineon set.Quote:
Originally Posted by RaZ
Although the price might indicate that the HX is the Infineon (G.Skill technician said the HX chip is more expensive).
We need that photo database :fact: :fact: :fact:
just let you know ZX is not using Hynix
PH means value ram :)
G.Skill Technician...
yes i am also the one that can vouch that a 3-3-2-5 timings and 260MHz 1:1 DRAM FSB.... the PHU-2GBZX 2x1gb ram...in dual channel.... will prime stable and also memtest fully stable...
the system will only boot with a CL3 timing @ 1:1 260MHz FSB...
but the minute you run a game like nfsmw, or 3dmark05, it will crash and hang on you....actually it will hang on you.. =)
but like i said...memtest is good/prime is good....just games no good......
windows is also good =) i can surf the net/chat/do documents in 260Mhz dram fsb...at 1:1.....
now if i run divider it'll be fine...9/10 divider gives me 236MHz FSB....with 2.5-3-2-5....(now i can reduce CL to 2.5...2.0CL won't boot...but 2.5 lets me boot..with 236MHz Dram FSB)
but no games at 1:1 260Mhz FSB.....
even though your first page did mention it CAN do DDR520 @ 1:1... that is probably not going to be game stable at all.....
i got no problems with the RAM...i paid a decent price for it on thanksgiving day at newegg..they had a nice special price..couldn't resist...
in case you wonder, here's the complete full system specs:
DFI LanParty uT ULtra-D Motherboard (BIOS 6/23/2005)
AMD X2 3800+ Dual Core Processor (OC'd to 2.60Ghz) 260x10FSB @ 1.550Vcore
G.Skillz 2GBZX 2x1GB 2.5-3-2-5 DDR PC3200 RAM @ 9/10 divider = 236MHz DRAM FSB.
eVGA 7800GT 256-P2-N516 PCI-xpress
Ultra X-Finity II 500watt PSU
2x300GB Seagate 7200.8 8MB Buffer PATA HD (raid 1)
2x300GB Seagate 7800.8 16MB Buffer PATA HD (raid 0)
Windows XP 32-bit SP1
Thermaltake Big Water SE Water Cooling Kit
Aquamark score: GFX 16,374. CPU 11,811. TOTAL Aquamark score: 96,707.
That is the exact same problems I have. Its really very frustrating :(.Quote:
Originally Posted by greendestiny
After reading the last two post I started thinking about the two sets of zx that I have. I knew I had run 05 before with them but couldn't find any ss. I threw a set in and ran them at 263x11 with a 7800gtx not oc'ed. Here is what I got.
http://img226.imageshack.us/img226/9...zx052xj.th.jpg
Will try the other set also. The two post above can you show a tweaker ss?