|14-07-2005, 01:21 AM||#1|
Join Date: Aug 13 2004
G.Skill F1-3200BWU2-1GBGH 1GB Winbond BH5 UTT Review & Discussion Thread (>=1024x768)
Review: G.Skill F1-3200BWU2-1GBGH 1GB Winbond BH5 UTT Dual Channel RAM kit
Hey i4memory members, after extensive testing with this G.Skill 1GBGH kit, it's time I showed you guys what this RAM is capable of (original discussion here: http://i4memory.com/showthread.php?t=843). Having owned this RAM for 2 weeks, and majority of time being downtime, because of dead Venice 3000+ (DFI 704 BIOS VID control bug), I haven't had much chance to tweak this up alot until my new CPU had arrived.
I hope this thread can turn into the official discussion thread, so please, if you own this RAM, or can help me, then feel free to contribute!
This is the RAM being tested:
Package : 1024MB kit (2x512MB) dual channel pack
IC Spec : Winbond UTT BH5
CAS Latency : 2-2-2-5 (PC3200)
Test Voltage : 2.7 ~ 2.9 V
PCB Board : 6 Layer Brain Power 6U808 PCB
Speed : DDR 400 MHz (PC3200)
Type : 184-pin DDR SDRAM
Error Checking : Non-ECC
Registered/Unbuffered : Unbuffered
Quality Control : Comprehensively tested in pairs in a dual channel environment
Warranty : Lifetime
As you can see, unlike the CH based UTT modules like the VX series, these new UTT BH-die modules are rated for low latencies at low voltages, which, means, more room to OC - but more results later.
Test System Platform:
DFI NF4 ULTRA-D (thanks nintek) w/ 704-2 BIOS (optimized for UTT/BH5/CH5 - ALL USERS please, if using this BIOS, avoid Above VID Control >123% - may destroy your CPU - especially 90nm)
Venice 3200+ 0517BPDW (thanks T_M@ i4m)
Hyper 6 with dual fans
G.Skill GH winbond kit as mentioned (thanks mittoni)
eVGA 6800 Ultra (thanks Sky95@ OCAU)
IDE WD HDD
Coolmax Taurus 550w (thanks Agg@ OCAU)
Just to recap on initial results and observations on the RAM's behaviour - I was having problems with asynchronous overclocking - but preliminary testing with the 510 series BIOSs showed that following eva2000's tip about using Tref of 0648 helped, but I was still limited to something around 210mhz @ 3.2v, which is quite poor, when on 1:1 @ 3.2v I can do 230 on very very tight timings (haven't started benching or burning in, etc...)
The modules themselves run extremely cool - even with no active cooling, the chips are considerably cooler than anything TCCD-based, and safe to say slightly cooler than the OCZ Gold PC3200 VX that I had previously owned.
Fortunately the 704 BIOS has fixed all the issues, and has given me a couple days to really tweak up this RAM. With a wide range of voltages available from the DFI - get prepared for various results people!
Please see below for OC results, and also performance benchmarks - this will be as comprehensive as possible - just as a hint of what's to come - there will be 1:1 testing first off, at various voltages, then there will be similar tests using memory dividers, and performance results at each of those settings. Please be patient - I must reformat my system, since it is still the same problematic windows install from my old NF3 AGP system. Tomorrow will be full of testing for you folks!
Last edited by groovetek; 15-07-2005 at 01:03 PM.
|14-07-2005, 01:22 AM||#2|
Join Date: Aug 13 2004
1:1 Synchronous RAM Testing
1:1 testing/overclocking was performed first to investigate the true limits of the RAM. Whilst AMD64 systems do not lose any performance running dividers provided that overall memory and CPU frequencies are the same - there are undeniably more factors that come into play when attempting to achieve 100% stability. Later on in the performance benchmarks, dividers will in fact be used in order to compare the impact of memory frequencies without changing CPU clocks. Fortunately I have been using this RAM with a divider without issue for a number of days now, and found that overclockability is similar, if not identical to 1:1. Regardless, it's common practice to investigate the true limits of the RAM using 1:1 settings as I am about to show.
The following tests were performed on the testbed mentioned above. All tests were performed using the following configuration in the DFI 704-2 BIOS. These timings were found to be the best performing, yet with as little practical impact on overclockability as possible. It is simply worth noting that other than the common CAS-Trcd-Trp-Tras timings, the other parameters can greatly impact memory performance. So whilst many people claim their memory can run at timings of "2-2-2-5", it is rather incomplete. For each voltage I have shown a memtest screenshot to gauge memory performance. I have found that the memory bandwidth indicated corresponds quite accurrately with final system performance, in RAM-intensive benchmarks such as 3DMark2001SE and SuperPI.
All tests were performed at LDT 4.0x, with the following RAM timings:
*2.9V Testing (within manufacturer voltage specifications):
Voltage reported by BIOS - 2.96v
It's no secret that Winbond-based memory chips crave voltage for high performance. This has been well-documented since late 2002, with the introduction of the original BH5 chips. Unfortunately back then, no motherboard was capable of supplying more than approximately 2.9-2.95 volts without serious modifications. It was in fact DFI that started a new trend in motherboards capable of supplying up to and over 3.2volts to the RAM.
The G.Skill 1GBGH BH5 UTT kit is rated between 2.7-2.9V. This is rather low, considering that other manufacturers such as OCZ have chosen to warrant their high performance Gold VX series modules up to 3.5 volts. Furthermore, such modules are built around CH5 modules, which are built on a finer process than the 175nm on which the BH5 is based. In theory this means the BH5 ICs should be less susceptible to electron migration due to overvoltage! Is this an indication of how much voltage this GH kit would be able to handle? Keep reading to find out.
I have chosen to include the following 2.9V results, to give an idea of how far this RAM can be taken whilst staying within the manufacturer's specifications, however conservative they may be. Whilst this RAM is targetted to enthusiasts, it is nonetheless important to consider those who are either uncomfortable with exceeding the specified voltage (and hence technically voiding warranty), or those who own motherboards that are capable of only supplying up to approximately 2.9 volts.
Using the timings shown above, the following overclock was achieved. Stability was tested using a brief memtest86+ v1.60 3 loop test #5 pass to guage suitability to boot windows, followed by a SuperPI 8M calculation. Whilst such tests may not seem to be incomplete - they are the best that can be done without spending an excessive amount of time. Being able to complete a SuperPI 8M run is a good measure of the memory's perfect stability point minus 2-3mhz MHz at most, at a given voltage.
>Maximum Memory OC
At this voltage, 225mhz returned an error during the SuperPI8M run, and lowering to 222 has alleviated the problem. Remember voltage is only at 2.9V, thus this RAM has exceeded the rated frequency by 22mhz at its rated voltage. It should be noted that for this test and all following test in this review, there was active air cooling on the modules, thus potentially increasing the maximum overclock at any given voltage, especially when exceeding 3 volts.
Voltage reported by BIOS - 3.28v
3.2V was chosen as the next voltage to test the GH kit, simply because a range of DFI boards offer up to this voltage level (eg, DFI NF2 LP, NF3, NF4-DAGF series motherboards). Even the DFI NF4 Ultra-D motherboard being used here today requires the movement of a jumper to enable voltages beyond 3.2 volts.
>Maximum Memory OC
As you can see, the memory scales very well with voltage - as expected with BH5 modules. A +0.3V bump equated to a 20mhz improvement, which also equates to a healthy performance boost. Just as a note - please take these 8M times as a grain of salt - I was doing other things at the same time for the sake of entertaining myself while it was testing, but more so that the system can be stressed out more . Please see the benchmarking results section below for some real performance comparisons.
Voltage reported by BIOS - 3.46v
3.4V is a common voltage for BH5 users - it offers near-maximum overclockability whilst being on the borderline for safe 24/7 use. Heat produced is elevated at this point - but most people have direct air cooling on the modules anyway so this is not a concern.
>Maximum Memory OC
Again, as you can see - the memory continues to scale with voltage. Do keep in mind that the A64 timings being used are extremely tight, and can easily outperform relaxed memory at higher frequencies. Whilst 252 is only a 10mhz improvement, it has crossed the PC4000 barrier which is already a respectable achievement.
*MAXIMUM OC @ MAXIMUM VOLTAGE:
Voltage reported by BIOS - 3.72v
Here is where it gets really fun. The following tests were performed at 3.6v + 0.3v option, which somehow results in 3.72V reading in BIOS. This kind of voltage is getting rather extreme, and is not advised. Besides, the gains are not very big, as you can see below.
>Maximum Memory OC
Yes, the memory continues to scale, however, it's difficult to say that the voltage required is justifed for the 9mhz increase over the previous voltage setting. I found that 264 gave me a few errors after a couple passes of memtest86+ v1.60, which indicates to me that i'm usually a few mhz off proper stability. I've been playing around with this setting for quite some time now, without any stability issues... but I am of course feeling abit worried about the reported 3.72volts passing through my RAM.
So there you have it - a run down of how this G.Skill 1GBGH BH5 UTT kit overclocks and scales with voltage. I may update this in the near future with a maximum overclock using relaxed settings - but as I mentioned before, the performance tradeoff simply for a few MHz is not worth it.
As for now, please read on for performance benchmarks.
Last edited by groovetek; 15-07-2005 at 03:15 PM.
|14-07-2005, 02:56 PM||#7|
Join Date: May 21 2005
I just picked up a pair from Mittoni about an hour ago on my way home as well.
In your opinion, do you think burn-in would be neccessary or helpful for G.Skill GH series?
If so, would it be the same as OCZ's recommendation like running the ram at 2-2-2-5 @ 3.2V using Memtest's Comprehensive Test for 48 Hours.
Was it actually 3.2V or 3.3V?
Thanks very much.
|14-07-2005, 05:14 PM||#8|
Join Date: Aug 13 2004
OK guys, i installed win xp x64, and everything is fine - EXCEPT... I've spent the whole afternoon trying to get my USB wireless adapter working - but no go - there simply are no drivers for it yet... but should be very soon...
Issue described here:
However, this won't hold me back from some overclocking/benchmarking now... so I'll be back every now (to this computer, with an internet connection) and then with some feedback on how I'm going.
|1gb, bh5, discussion, fbwu21gbgh, gskill, review, thread, utt, winbond|
|Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)|
|Similar Threads for: G.Skill F1-3200BWU2-1GBGH 1GB Winbond BH5 UTT Review & Discussion Thread|
|Thread||Thread Starter||Forum||Replies||Last Post|
|**i4memory SuperPI Mod v1.4 Rankings**||T_M||General Benchmarks||1006||07-08-2006 01:54 PM|
|[May 2005] Corsair revision -> ICs list||eva2000||General Hardware||12||31-07-2006 10:08 AM|
|G.Skill F1-3200BWU2-1GBGH 1GB Winbond BH5 UTT Discussion Thread||groovetek||Gskill DDR1 on AMD||58||06-07-2005 09:24 PM|
All times are GMT +11. The time now is 08:49 AM.