Relax the tRAS timings for better performance

[Bound_Vortex]

Here's something everyone should look at

I won't reiterate anything from the thread, but it's a great read.

My mem timings are now at 11-2-2-2 at 210 fsb, and my memory benchmarks higher in pcmark and sandra

 

[IsOs]

My MSI K7D Master only goes up to 9 for tRAS in the BIOS. Can someone clarify this for me?

 

[Frazas]

I can also confirm this. I get better performance not only in benchmarks but also in games using a timming of 2-2-2-11.

I have and Epox 8RDA (non-plus).

 

[maluckey]

The only reason you might want to relax the timing, is if you are saturating the bus, or are at the limits of the memory itself. Numbers stamped on memory, or in ths SPD mean little, when compared to quality of the hardware, memory, and setup overall..

 

[Jeff7181]

Originally posted by: maluckey
The only reason you might want to relax the timing, is if you are saturating the bus, or are at the limits of the memory itself.

Why?

 

[lchyi]

Good post. This is interesting.

 

[rogue1979]

Good info, it really works!

 

[Bound_Vortex]

Originally posted by: maluckey
The only reason you might want to relax the timing, is if you are saturating the bus, or are at the limits of the memory itself. Numbers stamped on memory, or in ths SPD mean little, when compared to quality of the hardware, memory, and setup overall..

Uh, I think you don't quite understand what tRAS is and how it works. Here from Mushkin

Some notes from that link for the lazy

"just by definition, the minimum tRAS must be the RAS-to-CAS delay plus the read latency (CAS delay). That is fine for FPM and EDO memory with their single word data transfers. With SDRAM, memory controllers started to output a chain of four consecutive quadwords on every access. With DDR, that number has increased to eight quadwords that effectively are two consecutive bursts of four"

"Any tRAS setting lower tRCD + CAS + 2 cycles will allow the memory controller to close the page ?in your face!? over and again and that will cause a performance hit because of a truncated transfer that needs to be repeated. Along with those hassles comes the self-explanatory risk for data corruption. That one is not a real problem as long as the system is kept running but in case it is shut down and the memory content is written back to the hard disk drive, the consequences can be catastrophic. For the drive, that is."

 

[Megatomic]

Hmmm, I have my tRAS set at 5 currently on my 8RDA+. I wonder how high I can set it? I will definitely check this out later tonight and see if I get better performance from changing this setting.

Thanks for the heads up Bound_Vortex.

 

[Bound_Vortex]

the 8RDA+ allows settings up to 15, but 11 seems like the sweet spot for fsb 200-210 (I noticed minor drops when tRAS was set to 10 or 12).

From the linked thread someone tested at 222 fsb, and performance peak was with tRAS at 11-13

Edit: Thought I would include this formula from that link

40ns(This is the number given as the minimum tRAS value) + 4clocks(Burst Length) - 1(Auto Precharge)=tRAS

e.q. 40ns at 200 fsb would give me 8 clock cycles

calc: (40x10^-9)/(1/(200x10^6)) = 8

8+4-1 = 11 tRAS

 

[maluckey]

I disagree, and maybe the structure of the KT266a makes a difference, but after 100 passes in UT2003, and twenty levels, I see degradation in-game with higher (numerically) Tras settings than the setting (5), that I am using. I run at 2-2-5-2 (CAS 2) and Tras 5, and 1t memory address. My HDD is a Maxtor 8mb cache 7200 RPM, and my current PCI bus is at 43.75 mhz. This may have something to do with it, but I'm not an engineer. I'm just someone who took the time to do the tests, and didn't see what they saw. Remember Bill Gates who stated that he could see no need for more memory? Remember when the 128mb Video cards were said to be slower than their 64mb counterparts? Test it outside the synthetic world. If you are happy, then stick to it.

I found many others who showed similar results, on AMDmotherboards.com. I also found many others, in a quick MSN search, who also saw similar results, though they did state that they sometimes encountered data corruption from the accelerated Tras settings. I run NTFS, so no worries here.

Another blatant example of synthetic error, is in SANDRA. If you set the CPU in Win-XP to utilize more resources for System Cache, you get outrageous scores for your HDD, but little real world anything.

 

[Jeff7181]

Originally posted by: maluckey
My HDD is a Maxtor 8mb cache 7200 RPM, and my current PCI bus is at 43.75 mhz

And it still works?

 

[maluckey]

3 hour a day, six days per week. I sometimes run as high as 45mhz PCI bus speed (180 mhz FSB, 1/4 divider), but the numerous fans want to hover the case around the room. I'm not alone in the high PCI bus list On this very site. Theres a guy running 47.5 mhz PCI bus...stable. Last I heard, on a KT266a, he was shooting for 200 FSB x 1/4 divider. That blows my modest OC to hell.

 

[Jeff7181]

Seems a little far fetches since most hard drives get flakey at 38 Mhz and above... but hey... congrats on the unusually high overclock.

 

[maluckey]

Thanks, but that isn't the point of my post. Just trying to dispell urban myth and legend, by doing it myself, instead of repeating what others say. For six months I listened to people tell me about doom and gloom for high FSB, then I tried it, and no explosions.

I can post results of any test you want, (if your're curious how an old KT266a holds up at these speeds). You'll be very surprised at the performance. It's for the most part on-par with the KT333's that are mildly overclocked.

As far as the legendary 40mhz PCI bus wall, it has more to do with design of the board, and less to do with an arbitrary number. Things like signal path length, ripple, and the ESR of the capacitors used in the circuit construction, greatly afect the overall outcome of the overclock. Lastly, the temp of the entire system is of crucial importance. A capacitor (or any component for that matter) can only last so long. They are rated for their effective life at no more than the rated temp. Higher, and you risk degrading the component significantly faster. This doesn't mean you will, only that you can.

 

[Amorphus]

at 180fsb you shoudl be running a 2/5 divider, it'll get you closer to 33...

 

[maluckey]

Trust me, if it were at all possible, I would run the PCI bus slightly lower, but I run an Abit KR7A-133, which is KT266a, and as such, has no 1/5 divider. Thei highest available is 1/4. with 1/3 being the only other option. This affects the entire system, in a positive way, (as long as it remains stable), as every component is accelerated.

So far, the myth about the KT266a Epox boards having a 1/5 divider, seems to be just that, a myth.

Getting back to the original subject. There are few, if any, real world reports that I've found on otherharcore sites, anecdotal or otherwise, that show decreased productivity, and/or FPS in games, while running numerically lower Tras settings, (in systems that can handle it). I can imagine memory that is so poor as to not be able to handle it, and systems that cannot take advantage of it, but in a properly optimized system, in real world apps, nobody is showing it to be slower..........yet.

As a side note, Kingston now warantees it's PC2700 HyperX memory to run at CAS 2, Trcd , Trp 2, Tras 5 at 1t address! Seems odd doesn't it? maybe Mushkin, like OCZ has something to gain from the posting?

 

[Bound_Vortex]

Originally posted by: maluckey
I disagree, and maybe the structure of the KT266a makes a difference, but after 100 passes in UT2003, and twenty levels, I see degradation in-game with higher (numerically) Tras settings than the setting (5), that I am using. I run at 2-2-5-2 (CAS 2) and Tras 5, and 1t memory address. My HDD is a Maxtor 8mb cache 7200 RPM, and my current PCI bus is at 43.75 mhz. This may have something to do with it, but I'm not an engineer. I'm just someone who took the time to do the tests, and didn't see what they saw. Remember Bill Gates who stated that he could see no need for more memory? Remember when the 128mb Video cards were said to be slower than their 64mb counterparts? Test it outside the synthetic world. If you are happy, then stick to it.

I found many others who showed similar results, on AMDmotherboards.com. I also found many others, in a quick MSN search, who also saw similar results, though they did state that they sometimes encountered data corruption from the accelerated Tras settings. I run NTFS, so no worries here.

Another blatant example of synthetic error, is in SANDRA. If you set the CPU in Win-XP to utilize more resources for System Cache, you get outrageous scores for your HDD, but little real world anything.

What do you mean by degradation in-game? Might this be that after 100 passes and twenty levels that your system is paging memory from the game to your hard disk and that is the cause for the slow down? I would be very interseted to see a link which shows faster effecient performance with lower tRAS settings, because with the 15 or so machines I have tested (various mobos, processors, and memory brands/types), I have not been able to come up with one instance.

Oh and running NTFS does not make you immune to data corruption as you seem to think.

I'm not sure what you mean by this "synthetic world". I merely presented tRAS, what it means, and how it works. This is by no means synthetic.

Originally posted by: maluckey
in real world apps, nobody is showing it to be slower..........yet

Did you even go to any of the links I provided?

Originally posted by: maluckey
As a side note, Kingston now warantees it's PC2700 HyperX memory to run at CAS 2, Trcd , Trp 2, Tras 5 at 1t address! Seems odd doesn't it? maybe Mushkin, like OCZ has something to gain from the posting?

Mushkin just happened to be the site of a quality explanation of tRAS that I found. Personally I use Corsair XMS 3200 in my system, and I have noticed an increase in not only performance at equal clock but also the ability to clock higher.

And as a side note, this would not be the first time engineers overlooked a basic principle when designing new faster components. It is not beyond the realm of possibility (I would go as far as to say it's probable) that when designers achieved optimum timings of PC-100 (which shown above would be your 5-2-2-2), they would keep those same timings as standards for higher memory clocks (assuming it would be the same). And who would question it...because on face value, you ARE slowing down a process in the function of the memory, so how would that be marketable or appealing?!?!. Sure the memory may run at 5-2-2-2, but running at a higher tRAS will be better not just for your data but your hard drive(s) as well. And as shown in many benchmarks it improves performance. This is not some synthetic anomaly, it's engineering.

 

[BDSM]

running it at 11 does yield better performance, at least on nforce 2 boards. This isn't just in sandra but in every benchmark I tried.

The End

 

[Jeff7181]

I'm off to run PCMark and find out if it makes a difference with mine =)

*EDIT* Ran it and got 6817, 6117, 1109

 

[SpideyCU]

Heck, I haven't downloaded PCMark since I reinstalled Windows on this rig a week ago, but I will now because I want to test this out!

 

[Bound_Vortex]

Jeff,
Looking at what your mem timings are at (7-3-3-2.5), did you just change to 11-3-3-2.5?

In my case, I was at 6-3-3-2.5 at 210 and going with tRAS of 11, allowed me to run at 11-2-2-2...just something you might want to try out

 

[SpideyCU]

Hmm, have to edit my message here, heh. Prime95 didn't settle in for stability when running at 11-2-2-2.0. I settled back to 8-3-3-2.0 and all is fine (180 MHz for the record). I'll be testing soon around 200 MHz to see how the timing work up there.

 

[KhoiFather]

Here are my results, messing with the settings.

MSI K7N2-Delta-L
2x 256MB Kingston Hyper X PC-3000 running dual channel
AXP 1700+ Tbred-A

PCMark2002
CPU is at default FSB!!!

cpu score 4331
mem. score 3484
HDD score 820
Memory Timing: 2-3-3-5 at 200MHz

4397
3500
916
Memory Timing: 2-3-3-11 at 200MHz

4406
3560
891
Memory Timing: 2-2-2-11 at 178MHz

4405
3594
908
Memory Timing: 2-2-2-10 at 178MHz

4399
3500
916
Memory Timing: 2-2-2-4 at 178MHz

BUT CHECK THIS OUT

4406
3632
903
Memory Timing:2-2-2-10 at 133MHz

 

[Bound_Vortex]

I'm putting together a comprehensive list for both pcmark and sandra

ETP of 45 min

Edit:

Using an Abit NF7-S rev2.0.
1700+ at 210.48x9 (as displayed in CPUid)
Memory is in sync and AGP is locked to 66 MHz

You can know more about my setup in my sig under toy

Each test was run 3 times with the average of all 3 taken for each score. After each trial of PCmark and Sandra (memory benchmark), machine was rebooted. Highest scores in Bold

Mem timings (PCMark cpu/mem/HD) (Sandra Int/Float)
12-2-2-2 (5893/5426/1329) (3236/2981)
11-2-2-2 (5898/5449/1325) (3236/2983)
10-2-2-2 (5898/5419/1310) (3235/2983)
9-2-2-2 (5894/5390/1318) (3221/2970)
8-2-2-2 (5892/5364/1316) (3208/2959)
7-2-2-2 no boot
6-2-2-2 no boot

for comparison, I ran the default settings for corsair only changing the tRAS
6-3-3-2.5 (5880/5261/1318) (3181/2911)
11-3-3-2.5 (5885/5342/1323) (3191/2921)

Note: about the ranges. As you can see, the cpu or HD didn't vary much regardless of settings, but the variance in memory results varied no more than 12 points total (from highest to lowest) in PCmark and no more than 6 points in either of the Sandra tests. And I should not that the 3 pcmark scores for 11-2-2-2 were 5447,5448,5452. Any of those would have been higher than any other score I received. As shown for Sandra tRAS of 10, 11, and 12 were very close together and the 9 results from those 3 settings were fairly identical.

As you can probably guess, i am running my memory at 11-2-2-2.