The Corsair Support Forums  

Go Back   The Corsair Support Forums > Corsair Enthusiasts Section > Overclocking and Benchmarking

Notices


Important Links
Great Threads
Forum Rules
FAQs
Edit System Specs
How to run Memtest86+
How to read CPU-Z
Memory / PSU Finder
Useful Tools
Memtest86+ v4.20
CPU-Z v1.63
ATTO v2.47
HDBENCH v3.40 b6
Super Pi
TrueCrypt v7.1a
SD Formatter 3.1
Audio Drivers
Vengeance 1500 v1.1
Vengeance 2000 v2.0.7
Gaming Software
K60/K90 Update
-M90 Game Profiles
SSD Firmware
Sandforce 5.05a
Enthusiasts
DDR1 IC Inquiry
DDR2 IC Inquiry
DDR3 IC Inquiry
Where to Buy
Retailers / Distributors
Online Store
RMA Request
Reply
 
Thread Tools Search this Thread Rate Thread Display Modes
  #1  
Old 07-05-2004, 05:16 AM
eva2000's Avatar
eva2000 eva2000 is offline
Registered User
 
Join Date: Aug 2002
Location: Brisbane, Australia
Posts: 172
POST ID # = 130004
Thumbs down XMS3200XL on AI7

I received some XMS3200XL to play with and evaluate and thought I'd report the results I got here. It seems XMS3200XL doesn't like Abit AI7 865PE springdale board.

System Used

P4 2.8C SL6Z5 3343A746 March 18, 2004
Abit AI7 rev1.0 17 bios
256MB Gecube X800Pro @ stock
2x 512MB Corsair XMS3200XL v1.1 0422014-1 Dual Channel Kit
80GB WD 800JB
550W Antec TrueControl PSU

Cpu cooled with Corsair Hydrocool 200 external water cooling unit.

constants set in bios
vAGP = 1.65v
AGP/PCI = 66.66/33.33 fixed


Notes specific to Abit AI7:
-----------------------------------------------------------
-----------------------------------------------------------
1. GAT of A-A-A-D-D refers to settings in the Advanced Chipset menu:

Game Accelerator = AUTO
Refresh Cycle Time = AUTO
Read Delay (tRD) = AUTO
Read Delay Adjust (tRDA) = DISABLED
Command Per Clock = DISABLED

These settings tweak the actual board's chipset to boost memory bandwidth and system performance to a level that rivals or even beats it's big brother Canterwood (875P chipset) which has PAT enabled all the time. The last 2 settings offer a big boost in performance but with nearly all memory introduces ALOT of instability so best to leave at DISABLED.

GAT on Abit AI7 differs from GAT on IS7/IC7/IC7-6/IC7 Max3 series and doesn't work exactly the same.

-----------------------------------------------------------
2. For optimal performance on AI7 board, you need to have full PAT (Performance Acceleration Technology) enabled. Without full PAT, system performance and benchmark scores suffer tremendously.

To enable full PAT you need the following settings:

in 1:1 cpu/ram ratio

a. NB Strap 667fsb and GAT set to A-A-A-D-D or
b. NB Strap 800fsb and GAT set to either Turbo-A-A-D-D, Street Racer-A-A-D-D or F1-A-A-D-D (T-A-A-D-D, SR-A-A-D-D, or F1-A-A-D-D for short) - this applies to AI7 17 bios only.

in async 5:4 or 3:2 ratio

a. NB Strap 667fsb and GAT set to A-A-A-D-D is only way to enable full PAT. NB Strap 800fsb T/SR/F1-A-A-D-D in async mode only give Partial PAT and not full PAT.


-----------------------------------------------------------
3. For full PAT to be enabled, AI7 uses NB Strap + GAT settings in combination to enable full PAT, disable PAT or Partial PAT.


-----------------------------------------------------------
4. Only certain memory chips or type of memory modules are compatible with Abit's NB Strap + GAT settings. So only some memory will allow AI7 to enable full PAT. From my testing only Winbond BH-5/6 and OCZ EB based ram can take partial or full advantage of Abit's GAT settings. Hynix D43 based ram in PC4000-4400 ram can't use GAT settings to enable PAT.

Corsair's new XMS3200XL based on Samsung TCCD 4ns chips is too new to 100% conclude whether or not this ram will work with Abit's NB Strap + GAT settings. Hence, why I am testing 2x 512MB Corsair XMS3200XL v1.1 of this new ram for the first time on Abit AI7 board.

Up till now, I've only read what XMS3200XL ram can do on certain boards. No one has reviewed XMS3200XL on Abit AI7 which is my favourite P4 motherboard. So my results below may not be typical of all XMS3200XL and all Abit AI7 boards. I'll later test XMS3200XL on Asus P4C800-E to reconfirm the results I get on Abit AI7.

-----------------------------------------------------------

Memtest86+ v1.20 Preliminary Results
First, I always extenstively test new ram using Memtest86+ - latest version being v1.20 at http://memtest.org. I'll test the ram at a particularly setting for 24-72hrs but due to time constraints, I'll do a very short preliminary test - this doesn't mean the memory is 100% error free (well what I consider error free), but will serve to approximate 90-95% error free mark.

I've divided the results into 2 groups, one which uses full PAT enabled settings and one which disables PAT. Enabling PAT is more stressing on the chipset and memory controller so normally full PAT enabled max oc mhz speeds will be lower than that of when disabled PAT is set.

Short test criteria = Memtest86+ v1.20 for 2 full standard loops (test #1-7) + test #5 loop for <=50 passes

note AI7's vdimm undervolts above 2.8v on my board so

2.8v = 2.8v real
2.9v = 2.85-2.86v real
2.95v = 2.91-2.92v real
3.0v = 2.96-2.97v real
3.05v = 3.03v real
3.1v = 3.07v real
3.2v = 3.18v real

Memtest86+ reported bandwidth doesn't show the full real bandwidth differences between no PAT and full PAT below, to see real differences you need to bench in windows.
_____________________________________________________________
Memtest86+ v1.20
with PAT disabled
1:1 ratio


200mhz SPD (2.5-2-2-5) NBStrap 800fsb A-A-A-D-D - 2.8v - 2610MB/s = PASSES

210mhz 2-2-2-5 NBStrap 800fsb A-A-A-D-D - 2.8v - 2742MB/s = FAILED 2 errs 1st pass TEST#2 + 47 errs 1st pass TEST#3 + 2347 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 800fsb A-A-A-D-D - 3.0v - 2742MB/s = FAILED 2 errs 1st pass TEST#3 + 312 errs 1st pass TEST#5

210mhz 2-2-2-6 NBStrap 800fsb A-A-A-D-D - 2.8v - 2742MB/s = FAILED 26 errs 1st pass TEST#3 + 1561 errs 1st pass TEST#5
210mhz 2-2-2-6 NBStrap 800fsb A-A-A-D-D - 2.9v - 2742MB/s = FAILED 14 errs 1st pass TEST#3 + 956 errs 1st pass TEST#5
210mhz 2-2-2-6 NBStrap 800fsb A-A-A-D-D - 3.2v - 2742MB/s = FAILED 2 errs 2nd pass TEST#5

210mhz 2-2-2-7 NBStrap 800fsb A-A-A-D-D - 2.8v - 2742MB/s = FAILED 1 err 1st pass TEST#2 + 20 errs 1st pass TEST#3 + 1351 errs 1st pass TEST#5

210mhz 2-2-2-8 NBStrap 800fsb A-A-A-D-D - 2.8v - 2742MB/s = FAILED 2 errs 1st pass TEST#2 + 34 errs 1st pass TEST#3 + 2284 errs 1st pass TEST#5

210mhz 2.5-2-2-5 NBStrap 800fsb A-A-A-D-D - 2.8v - 2742MB/s = FAILED 4 errs 1st pass TEST#2 + 31 errs 1st pass TEST#3 + 2530 errs 1st pass TEST#5

210mhz 2-3-2-5 NBStrap 800fsb A-A-A-D-D - 2.8v - 2680MB/s = PASSED 5 passes of TEST#5 loop

220mhz 2-3-2-5 NBStrap 800fsb A-A-A-D-D - 2.8v - 2801MB/s = FAILED 4634 errs TEST#5 loop
220mhz 2-3-2-5 NBStrap 800fsb A-A-A-D-D - 2.9v - 2801MB/s = FAILED 782 errs TEST#5 loop
220mhz 2-3-2-5 NBStrap 800fsb A-A-A-D-D - 3.2v - 2801MB/s = PASSED 11 passes of TEST#5 loop

240mhz 2-3-2-5 NBStrap 800fsb A-A-A-D-D - 3.2v - 3059MB/s = FAILED 502 errs 1st pass TEST#1 + 30,000+ errs 1st pass TEST#5

250mhz 2.5-3-3-7 NBStrap 800fsb A-A-A-D-D - 2.9v - 3115MB/s = FAILED 16 errs 1st pass TEST#5 loop
250mhz 2.5-3-3-7 NBStrap 800fsb A-A-A-D-D - 2.95v - 3115MB/s = FAILED 32 errs 4th pass TEST#5 loop
250mhz 2.5-3-3-7 NBStrap 800fsb A-A-A-D-D - 3.1v - 3115MB/s = PASSED 103 passes TEST#5 loop
250mhz 2.5-3-3-7 NBStrap 800fsb A-A-A-D-D - 3.2v - 3115MB/s = PASSED 10 passes TEST#5 loop

260mhz 2.5-3-3-7 NBStrap 800fsb A-A-A-D-D - 3.2v - 3233MB/s = FAILED 128 errs 1st pass TEST#3 + PC Rebooted during Test#4

260mhz 3-3-3-8 NBStrap 800fsb A-A-A-D-D - 3.2v - 3233MB/s = FAILED 8 errs 1st pass TEST#3 + ALOT of errors TEST#4 + 524 errs 1st pass TEST#5 + Memtest85 screen FREEZE

260mhz 3-4-4-8 NBStrap 800fsb A-A-A-D-D - 3.2v - 2965MB/s = FAILED 32 errs 1st pass TEST#3 + TEST#4 FREEZE + 2300 errs 1st pass TEST#5

_____________________________________________________________
Memtest86+ v1.20
with PAT ENABLED
1:1 ratio



200mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 2.8v - 2740MB/s = PASSED 30 passes of TEST#5 loop
200mhz 2-2-2-5 NBStrap 800fsb T-A-A-D-D - 2.8v - 2886MB/s = FAILED 2 errs on 74th pass of TEST#5 loop
200mhz 2-2-2-5 NBStrap 800fsb T-A-A-D-D - 2.9v - 2886MB/s = PASSED 725 passes of TEST#5 loop
200mhz 2-2-2-5 NBStrap 800fsb SR-A-A-D-D - 2.9v - 2886MB/s = PASSED 13 passes of TEST#5 loop
200mhz 2-2-2-5 NBStrap 800fsb F1-A-A-D-D - 2.9v - 2886MB/s = PASSED 51 passes of TEST#5 loop
200mhz 2-2-2-5 NBStrap 800fsb F1-E-A-D-D - 2.9v - 2886MB/s = PASSED 195 passes of TEST#5 loop

210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 2.8v - 2878MB/s = FAILED 12 errs 1st pass TEST#3 + 1661 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 2.9v - 2878MB/s = FAILED 11 errs 1st pass TEST#3 + 1347 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 2.95v - 2878MB/s = FAILED 3 errs 1st pass TEST#3 + 668 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 3.0v - 2878MB/s = FAILED 1 err 1st pass TEST#3 + 326 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 3.1v - 2878MB/s = FAILED 110 errs 1st pass TEST#5
210mhz 2-2-2-5 NBStrap 667fsb A-A-A-D-D - 3.2v - 2878MB/s = FAILED 6 errs 1st pass TEST#5

220mhz 2.5-3-3-7 NBStrap 667fsb A-A-A-D-D - 2.8v - 2801MB/s = PASSED 39 passes TEST#5 loop
220mhz 2.5-3-3-7 NBStrap 800fsb T-A-A-D-D - 2.8v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads
220mhz 2.5-3-3-7 NBStrap 800fsb T-A-A-D-D - 3.2v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads

233mhz 2.5-3-3-7 NBStrap 667fsb A-A-A-D-D - 2.8v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads
233mhz 2.5-3-3-7 NBStrap 667fsb A-A-A-D-D - 2.9v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads
233mhz 2.5-3-3-7 NBStrap 667fsb A-A-A-D-D - 3.1v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads
233mhz 2.5-3-3-7 NBStrap 667fsb A-A-A-D-D - 3.2v = FAILED = NO BOOT INTO MEMTEST86+ screen hang just before Memtest loads

==============================================================
Preliminary Memtest86+ Summary:
As you can see the results don't look pretty at all :( Whether this is possible a bad pair of XMS3200XL or Abit AI7 not liking XMS3200XL, I'm not sure. If you own an Abit AI7 and XMS3200XL ram already, feel free to try the above settings I used in Memtest86+ v1.20 and see if you get the same results and report back here.

From the above results:
1. XMS3200XL performs wonderfully at 200mhz 2-2-2-5 being able to handle all 4 combinations of NB Strap and GAT to enable full PAT. However, even a mild overclock of 210mhz 2-2-2-5 failed miserably at all vdimm voltages between 2.8-3.2v. 250mhz 2.5-3-3-7 needed alot more vdimm that other reviews reported, only being stable at 3.1v - 3.2v.

2. Other reviews online showed XMS3200XL could handle between 214-220mhz 2-2-2-5 and 240-250mhz 2.5-3-3-7 at 2.85v and under on different boards such as Asus P4C800-E. And that more vdimm voltage doesn't help XMS3200XL which I have confirmed above. I have P4C800-E here as well so will test that as well.

3. XMS3200XL Plug N' Frag auto set 2-2-2-5 timings, doesn't work on Abit AI7 maybe due to Abit's bios incorrectly reading XMS3200XL's SPD as 2.5-2-2-5

4. In NB Strap 800fsb mode, loosening tRAS from 5 to 6 so 2-2-2-6 was used, resulted in less errors in memtest than at 2-2-2-5, but nonetheless didn't help @ 210mhz 2-2-2-6 and still FAILED.

To contrast, this exact same Abit AI7 board allows the following ram to achieve the following:


2x 512MB Corsair XMS3200C2v1.1 BH-6 @ 240mhz 2-2-2-6 at 3.2v
2x 512MB Corsair XMS3500C2v1.1 BH-5 @ 240mhz 2-2-2-6 at 3.2v
2x 256MB Corsair XMS3500C2v1.1 BH-5 @ 240-252mhz 2-2-2-6 at 3.2v
2x 512MB OCZ PC3500 Platinum Limited Edition BH-5 @ 240mhz 2-2-2-6 and 250mhz 2-2-2-7 at 3.2v
2x 256MB OCZ PC3500 Enhanced Bandwidth @ 240mhz 3-2-2-6 and 250mhz 3-3-2-7/8 at 3.2v
2x 512MB OCZ PC3500 Gold @ 200mhz 2-3-2-5 at 2.65v and 239mhz 2.5-3-3-7 at 2.65v
2x 512MB OCZ PC3700 Gold rev1 @ 250mhz 2-3-3-7 at 3.1-3.2v vdimm
2x 512MB OCZ PC3700 Gold rev2 @ 200mhz 2-3-2-5 at 2.65v and 250mhz 2.5-3-3-7 at 2.65v
2x 512MB Mushkin PC3500 Black LVL2 BH-5 @ 238-240mhz 2-2-2-6 at 3.2v
2x 256MB Kingston PC3200 KHX3200K2/512 BH-5 @ 240mhz 2-2-2-6 at 3.2v


i.e. 2x 512MB XMS3200C2v1.1 BH-6 @240mhz 2-2-2-6 http://www.houseofhelp.com/v2/showthread.php?t=26403

Again, this could be a bad pair of XMS3200XL i got or it could be Abit AI7 not liking XMS3200XL ram. I'll test on Asus P4C800-E as well to see if the results differ. If you own an Abit AI7 and XMS3200XL ram already, feel free to try the above settings I used in Memtest86+ v1.20 and see if you get the same results and report back here.

To prove nothing wrong on cpu side, here's 2.8C@275fsb 5:4 with 2x 256MB KHX3200K2/BH-5 @ 220mhz 2-2-2-6 prime stable.

PLEASE, read further below in this thread for further progress and updates or other settings I may try to figure out what settings work for this XMS3200XL ram on AI7. It could be due to the tight SPD timings that XMS3200XL uses/read by Abit bios that needs GAT settings to be further tweaked.

I will test on P4C800-E later as well.
Reply With Quote
  #2  
Old 07-05-2004, 05:17 AM
eva2000's Avatar
eva2000 eva2000 is offline
Registered User
 
Join Date: Aug 2002
Location: Brisbane, Australia
Posts: 172
POST ID # = 130005
Default

While it seems I still haven't figured out why XMS3200XL can't even do a 210mhz 2-2-2-5 oc at 2.8-3.2v vdimm at either NB Straps! sigh

But this XMS3200XL is fast at DDR400 (note prelim memtest so this would not be 100% stable but 90-95% stable in my books to be 100% stable i'd need to memtest for 24-72hrs heh)

---------------------------------------------------------------------------
This is Sandra 2004 SP2 STD unbuffered bandwidth

To run unbuffered uncheck 9 check boxes in memory module's options panel shown here

2x 512MB XMS3200XLv1.1
@ 200mhz 2-2-2-5

NB Strap 800fsb A-A-A-D-D [PAT disabled] = 2790/2799
NB Strap 800fsb T-A-A-D-D [PAT enabled] = 3147/3143
NB Strap 800fsb SR-A-A-D-D [PAT enabled] = 3137/3170
NB Strap 800fsb F1-A-A-D-D [PAT enabled] = 3164/3244
NB Strap 667fsb A-A-A-D-D [PAT enabled] = 3034/3034

---------------------------------------------------------------------------
This is Aida32 v3.93 Read and Write bandwidth

Read/Write

NB Strap 800fsb A-A-A-D-D [PAT disabled] = 4297/1710
NB Strap 800fsb T-A-A-D-D [PAT enabled] = 5067/2344
NB Strap 800fsb SR-A-A-D-D [PAT enabled] = 5087/2336
NB Strap 800fsb F1-A-A-D-D [PAT enabled] = 5341/2339
NB Strap 667fsb A-A-A-D-D [PAT enabled] = 5089/1872


---------------------------------------------------------------------------
This is CTIAW bandwidth

NB Strap 800fsb A-A-A-D-D [PAT disabled] = 5168.9 MB/s
NB Strap 800fsb T-A-A-D-D [PAT enabled] = 5676.4 MB/s
NB Strap 800fsb SR-A-A-D-D [PAT enabled] = 5682.7 MB/s
NB Strap 800fsb F1-A-A-D-D [PAT enabled] = 5722.6 MB/s
NB Strap 667fsb A-A-A-D-D [PAT enabled] = 5639.0 MB/s

As you can see the difference in performance due to the GAT and NB Strap settings. For NB Strap 800fsb and T/SR/F1-A-A-D-D I would memtest thorough for 24hrs and prime95x2 instances before deeming it stable if I was you.



.
Reply With Quote
  #3  
Old 07-12-2004, 05:39 PM
Clevor Clevor is offline
Registered User
 
Join Date: Jul 2002
Posts: 159
POST ID # = 131397
Default

Well Eva, you know the story of the AI7: picky, picky, picky. I still have my board up. Hell, my OCZ PC3700 Golds Rev. 1.0 won't run on it in any shape or form (Sammy TCCB). Maybe 1:1 did, I forget. Sammy stuff don't like it either, AKA your results with the 3200XL.

I did some testing of real late Sammy TCCCs on the P4C800-E and while they did 258, 1:1, 3-3-3-5 at 3.1 volts, they would not do any other timings nor run 5:4 whatever.

I notice my Buffalo 2x512 with -5BC Micron pretty much matches your EB3500 on the AI7 as I could do 248, 3-2-3-5 at 2.8 volts, more didn't help.

I have not read all the reviews but I bet even 3200XL sticks capable of 250, 1:1, 2.5-3-3-7 will not run mem ratios at all.

BTW, what ICs are in the OCZ PC3700 Golds Rev. 2.0? I believe they are speed binned Hynix, maybe same or similar to my ADATA PC450 with -4.5 ns Hynix. I know OCZ was real secretive with this when they came out. Isn't it just -4.3 ns Hynix? Stuff you can find in some Valueram?

Last edited by Clevor; 07-12-2004 at 05:52 PM.
Reply With Quote
  #4  
Old 07-12-2004, 05:45 PM
Clevor Clevor is offline
Registered User
 
Join Date: Jul 2002
Posts: 159
POST ID # = 131399
Default

Last night I tried three-four 3.0Cs on my late AI7 and I see something disturbing: the system cuts out completely on max overclocks around 255 fsb or 3.82 gig and 1.68-1.75 VCORE. The system just shuts down in the middle of Aquanox or PCMark2002. CPU temp is 35-37 C with my HC200 and I have all the temp/voltage monitoring in the AI7 BIOS disabled.

Eva, I know you don't have any obstinate 3.0Cs but have you ever experienced anything similar? I had the same experience with a Rev. 1.2 IS7-E and the DFI LanParty 875B. It's almost like Intel rigged the CPU to short all voltage on max overclocks to prevent the CPU from frying. I don't know if this is possible.

I'm getting fed up with this new AI7 and will be putting the old C3 board in tonight. If the Micron -5BC ram performs acceptable with the undervolting VDIMM I may get rid of the board.
Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes Rate This Thread
Rate This Thread:



Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -5. The time now is 01:58 AM.


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2013, vBulletin Solutions, Inc.