Opinions on this DDR3 1600 for P5E3 Premium

[Noel 4,486]

Since I don't plan on more than a 410 or so Mhz FSB, what do you think of this stuff at $255? Won't be 1T, but timings are good and this would be a modest overclock.Model Brand OCZ Series Platinum Model OCZ3P16004GK Type 240-Pin DDR3 SDRAM Tech Spec Capacity 4GB (2 x 2GB) Speed DDR3 1600 (PC3 12800) Cas Latency 7 Timing 7-7-7-24 Voltage 1.9V Heat Spreader Yes Features 1.95V EVPPlatinum Z3 XTC Heatspreader Recommend Use High Performance or Gaming Memory ------This stuff is about $390:Model Brand Patriot Series Viper Model PVS34G1600LLK Type 240-Pin DDR3 SDRAM Tech Spec Capacity 4GB (2 x 2GB) Speed DDR3 1600 (PC3 12800) Cas Latency 7 Timing 7-7-7-20 Voltage 1.9V Heat Spreader Yes ---------This stuff is only $650. Since I'm not doing alot of o'c on this, probably not much point in this Corsair.Model Brand CORSAIR Model WIN3X40961600C7DHXIN Type 240-Pin DDR3 SDRAM Tech Spec Capacity 4GB (2 x 2GB) Speed DDR3 1600 (PC3 12800) Cas Latency 7 Timing Intel XMP SPD Certified Profile 7-7-7-20 Heat Spreader Yes Manufacturer Warranty Parts Lifetime limited Labor Lifetime limited Thanks in advance for your input,Noel

[McCrash 3]

Try and find some test reports on the memory your listed.

[sargeski 16]

Noel,I'm using the Patriot Viper with my P5E3 Premium. Works just fine with the motherboard no issues at all.

[Noel 4,486]

Thanks Sarg. Are you able (or have you tried) to run at 1640Mhz and retain CAS 7 timing?QX9650 w/ Retail HSF|8GB Muskin PC-6400|ASUS P5E|EVGA 8800GT @700|Seagate SATA 2 x 4|Seagate Cheetah 15K.x|XP Pro|Vista 64--soon to be installed

[Noel 4,486]

Sarg, can you run a Everest latency and read benchmark? Thanks alot in advance if you can . . .QX9650 w/ Retail HSF|8GB Muskin PC-6400|ASUS P5E|EVGA 8800GT @700|Seagate SATA 2 x 4|Seagate Cheetah 15K.x|XP Pro|Vista 64--soon to be installed

[Noel 4,486]

Sarg, do you mind running an Everest latency and read benchmark on your rig? Much appreciated!Noel

[sargeski 16]

Noel,Here you go.I really haven't tried to push the memory. I'm quite happy with it at stock settings. Running a TrD of 6.http://forums.avsim.net/user_files/189227.jpg

[Noel 4,486]

That's good nuf for me. I do wonder tho a few posters, I think the majority (of a small sample) said it wouldn't do CAS 7 if I recall. If it did CAS 7 I'd go for it at price.

[Guest D17S]

$70 worth-O-4Gs-O DDR2-800 @ 800 and 5-5-5-18http://forums.avsim.net/user_files/189273.jpgAnd you're gonna pay How much for "new and improved?" Speed costs, that's for sure, but isn't this gettin' a bit out of hand!?

[Noel 4,486]

Sam, I have a P5E3-Premium coming, and it needs DDR3. I know, I went for it. A 4th P5E I can't stomach!Sarg's OCZ is *only* $255. The Corsair Dominator DDR3 1800 2x2 was in 3 locations yesterday at $525, and is all sold out as of this morning. Yes, it has gotten way out of hand for me. I am determined to see 1640 @ CAS 7, and tRD of 6 or whatever works. If I have to cough up $550 I will! Makes perfect sense to someone who coughed up for a QX processor, no? I still have zero clear advise on where to go with vMCH. Just vague generalities ("you have to increase vMCH when you up the FSB etc etc). I will try consulting with multiple parties at Corsair and the other memory people, and do some more reading about that. One thing I also have gotten no clear answer on is what my BIOS does when you set, for example, NB voltage to AUTO. Does it scale up with demand? Or is it the same as default? I know with vCore it doesn't scale up. Anyway, probably have to call ASUS and see.

[Guest D17S]

This is my ol' DDR2-800 about maxed out. This is 4, 1gig sticks too. My P5K-e does not allow me to directly see tRD. I can only adjust it in increments of -1, -2, -3. I found the -3 won't run, even at a Northbridge voltage of 1.7v. It would be Very handy to know what my actual tRD was. Latency is additive. If I knew what tRD number I was using, I could calculate it's contribution to the latency pie. I could then know what "the rest" was. That might be helpful in isolating other player's latency contributions. Consider: If Anand's description of tRD of 12 being typical, the tRD's crossing latency is worth 2.5ns (at a 400Mhz FSB) times 12, or 30ns. If I could get rid of the crossing delay, I could reduce my latency by 30ns. Gentlemen, that's a BuncH. 50% for me. I expect Nehalem's QuickPath is a LoT about that. This and the previous 800MHz run (at 4-4-4-12. Yes, I goofed the text) is at a setting of -2, with NB volts at 1.55v. http://forums.avsim.net/user_files/189274.jpgSo, how much difference does tRD make? Look at this: http://forums.avsim.net/user_files/189275.jpgAll I did was reset turbobuster (or whatever they call it) to Auto. What a difference. My arithmetic tells me I lost 3 FSB cycles. (2.5ns x 3 = my latency loss of ~ 7.5ns). If Anand is correct, my tRD increased by three. How about that. That tRD explanation might have just proven out. It's pretty close, actually. Does that mean that at a 400Mhz FSB (aka, 2.5ns per cycle) each notch of tRD will reduce latency by 2.5ns? That supertuned northbridge in the X48 may be worthwhile. 6 notches of tRD is worth 15ns of latency? Hummm. Try it. Now look at the tRD's effect on the rest. Read ramped, Write didn't budge and Copy only came along. What's up with that? I'm still not entirely sure what these numbers represent. 8GB/s is Lot of data . . . and those are BiG "B"s That's 64Gb/s? What's up with that? I'm almost thinking those might be ram module internal numbers. The FSB is only clocking along at pokey 400Mhz, or quad pumped for a potential transfer rate of 1.6Gb/s ('ittle "b"). Something's not tracking. Mr. Wizard. Need more data!

[Noel 4,486]

I believe what you are getting you can read under PERFORMANCE in MemSet. I know with the 4 x 1Gb OCZ 800 ram I had running at 1050Mhz, CAS 4, and had Everest latency of 51.8ns at the best ever. If I recall correctly, I was able to set Transaction Booster (must be similar to directly setting tRD on an X48, tho I don't know for sure. I DO know that going +1 translates to 1 less in MemSet Performance.

[Reset XPDR 1,271]

What I don't get with this whole memory optimisation thing is that a drop of main memory access latency from say 60ns to 54ns is only a 10% improvement in memory access speed. Yet main memory access latency is only a subset of instructions that a CPU performs, especially since the CPU cache handles a lot of these memory access requests, hence this 10% performance boost at the system level is likely much, much lower perhaps in the 1-2% band level. Dedicated memory performance benchmarks show this, so how does this exactly translate into phenomenal performance increases in real world applications, like a major performance boost in overall FS9/FSX texture loading performance. The cynic in me says "I don't think so".Gary

[Guest D17S]

Yea, Performance Level on Memset is tRD. The darn thing will adjust on the fly. Just locked myself up at a tRD of 5. That makes sense. My "-3" setting won't even boot. This P35 just doesn't have the chops. So a tRD of 7 means that 2.5ns x 7 or 17.5ns of my ram's latency is tRD related. On the other hand, how much more can even a super-duper NB offer. I don't think I'd spend even $3 to pick up 3 more tRDs. Nehalem's just around the corner and that may take that down a bunch. But do we really need it? Consider: 40ns are still happening somewhere else. What's going on there? Why would speeding up the ram's internal processes (timings & clock) allow faster transfers (lower latency)? Not that it couldn't, but if it does, why? If an operator is using the same tRD rate (for instance exactly every 7th cycle), how can data transfer rates increase? We can "quad pump" a cycle to load up 4 bits on a single cycle (Anand didn't address that), but that's the end of it. I have a feeling increases in performance is going to be about latency. The rest may be just a look inside the ram module's machine room.For this circumstance we are observing to occur, tRD - Must - mean that these transfer events are only Available every (for instance) 7th cycle. If we were actually using every 7th cycle, the latency would be 7 x 2.5ns or 17.5ns. At a 60ns latency, only every (60 / 2.5ns) 24th cycle is being used. It appears that ram's transfer capability is way behind the FSB's ability to carry that data. Maybe that's why a 266Mhz FSB works just fine with whatever ram you want to throw at it! High FSB speeds really are totally meaningless. I'm getting dizzy. Your turn.

[Guest D17S]

"Yet main memory access latency is only a subset of instructions"(So far) I'm arguing that Latency is not just another factor. The latency number represents the actual flow of data transfered between the ram and the CPU. The ram's internal timings and clock simply make the ram's internal data more available to transfer. For instance, with a tRD of 7 on a 400Mhz FSB, the FSB provides a data transfer opportunity every 7 cycles (7 x 2.5ns) or every 17.5ns. If they were all used, my latency would be 17.5ns . . . but it's not. Why? The ram cannot produce data that fast. So, how fast - Can - my ram produce data? My latency is 60ns. The FSB is not the bottleneck. It's provides a transfer opportunity every 17.5ns. So what's holding up production? The RAM(?!).So why does increasing the CPU's ram-Feed - Not - translate on a 1:1 basis into tangible, non-subjective increased performance. It surely doesn't and that's DefinaTelY going to be another one of our $64 questions.