[djgar]

Quote:

Originally Posted by 100pipito

CPU Intel Core i7-6700K 4.0GHz Quad-Core Processor $329.99 @ SuperBiiz
CPU Cooler Cooler Master Hyper 212 EVO 82.9 CFM Sleeve Bearing CPU Cooler $29.99 @ NCIX US
Motherboard Asus Z170-A ATX LGA1151 Motherboard $147.98 @ Newegg
Memory G.Skill Ripjaws 4 Series 16GB (2 x 8GB) DDR4-3000 Memory $99.99 @ Newegg
Storage Samsung 850 EVO-Series 500GB 2.5" Solid State Drive $169.49 @ OutletPC
Storage Western Digital Caviar Blue 1TB 3.5" 7200RPM Internal Hard Drive $49.55 @ OutletPC
Video Card Gigabyte GeForce GTX 1070 8GB G1 Gaming Video Card $399.99 @ Newegg
Case Fractal Design Define R5 w/Window (Titanium) ATX Mid Tower Case $94.99 @ SuperBiiz
Power Supply Corsair RMx 650W 80+ Gold Certified Fully-Modular ATX Power Supply $99.99 @ Newegg
Case Fan Akasa AK-FN073 109.5 CFM 140mm Fan -
External Storage Seagate Expansion 1TB External Hard Drive -
Prices include shipping, taxes, rebates, and discounts
Total $1421.96

I have bought the g skills ram of 16 gb (2x8 sticks) and putted them into dual channel slot and then I tried to enable xmp profile and had no luck with it as it always kind of puts 2133 MHz in dram configuration.

Any suggestions to try something?

Thanks in advance.

Take this information and put it in your sig using RigBuilder so we always have it available - check Jpmboy's signature.

 

[Bojamijams]

Does anyone have any tips for SK Hynix ICs? I have a set of HyperX Savage that runs at 2400 12-13-13-35 (as per XMP) @ 1.35V, but I'd like to get it up to DDR4-3000 but am not having any luck.

 

[JackCY]

With most RAM what you buy is what you get, they are binned and sold as fast as they will run most of the time. Gotta find some rare kit and batch that was sold at lower price as lower speed but will run faster in fact.

 

[Bojamijams]

Quote:

Originally Posted by JackCY

With most RAM what you buy is what you get, they are binned and sold as fast as they will run most of the time. Gotta find some rare kit and batch that was sold at lower price as lower speed but will run faster in fact.

If I wanted to upgrade to faster ram, what should I look for in terms of overclockability? I don't really want to pay for DDR4-3600+ but would love to buy a 3000 kit that can overclock to that. I hear Samsung ICs are better, is that true? Is there a low profile kit (for heatsink clearance) that you can recommend? How do I find the IC used on it? (sorry for the noob questions)

 

[Jpmboy]

Quote:

Originally Posted by Bojamijams

If I wanted to upgrade to faster ram, what should I look for in terms of overclockability? I don't really want to pay for DDR4-3600+ but would love to buy a 3000 kit that can overclock to that. I hear Samsung ICs are better, is that true? Is there a low profile kit (for heatsink clearance) that you can recommend? How do I find the IC used on it? (sorry for the noob questions)

how high in voltage have you gone with your current 2400 kit in trying to reach 3000? (you are asking a lot from that kit).
Have you been able to run 2666 13-14-14 at 1.4V

 

[Bojamijams]

Quote:

Originally Posted by Jpmboy

how high in voltage have you gone with your current 2400 kit in trying to reach 3000? (you are asking a lot from that kit).
Have you been able to run 2666 13-14-14 at 1.4V

I've tried 1.4 Vdimm and 1.2 VCCIO and VCCSA.

What do you recommend for the 4th value with the 13-14-14... 39? Auto?

And holy crap.. You got your ram at C13 3400?? that's ridiculous! I want that but I don't think TridentZ are low profile enough

 

[Jpmboy]

Quote:

Originally Posted by Bojamijams

I've tried 1.4 Vdimm and 1.2 VCCIO and VCCSA.

What do you recommend for the 4th value with the 13-14-14... 39? Auto?

And holy crap.. You got your ram at C13 3400?? that's ridiculous! I want that but I don't think TridentZ are low profile enough

the 4th value is tRAS, and tRAS = CAS(CL)+tRCD+tRTP. So for your setup the correct timing is most likely 39.

3400c13 is on my x99 with 64GB of ram. The z170 boards run 3600 to 4133 with decent ram kits. Check the table in the OP for some settings.

 

[Bojamijams]

Quote:

Originally Posted by Jpmboy

the 4th value is tRAS, and tRAS = CAS(CL)+tRCD+tRTP. So for your setup the correct timing is most likely 39.

3400c13 is on my x99 with 64GB of ram. The z170 boards run 3600 to 4133 with decent ram kits.

Given that formula it should be 41 then right? (your suggestion being 13-14-14 = 13+14+14)

Also why do you think I'm asking for too much from my kit? When I tried 3000 I didn't mean I did it with cas12

I tried 15-15-15-35 @ 3000 but now, knowing the full formula, I see I should've been trying 15-15-15-45

 

[JackCY]

Some RAM will OC to higher clock speed but you have to adjust the latency which hurts performance, so you're trading one thing for another. I don't know of any magic DDR4 kits that would be sold "under binned" such as buy 2133 and OC to 3200 while using not much higher latency at all. Even with DDR3 it was rare, sort of buy many and see or watch online forums, reddit etc. but it's often getting a specific kit from a specific batch that was binned lower so it sells because the fastest binned RAMs sell poorly due to their ridiculous pricing.
2133-2400MHz was normal for DDR3, with DDR4 the average kits to buy are around 3000-3200MHz but the pricing I do not know right now, maybe they are still milking DDR4.

 

[Jpmboy]

Quote:

Originally Posted by Bojamijams

Given that formula it should be 41 then right? (your suggestion being 13-14-14 = 13+14+14)

Also why do you think I'm asking for too much from my kit? When I tried 3000 I didn't mean I did it with cas12

I tried 15-15-15-35 @ 3000 but now, knowing the full formula, I see I should've been trying 15-15-15-45

you are confusing tRP with tRTP. Dl a copy of the latest Asrock Timing Configurator for your MB and post a snip back here.
And getting a 2400 kit o orun 3000 is not impossible, it just depends on the capabilities of the ram ICs.

this was with a GSkill economy kit (I think they were 2133) for my wifes Tax prep PC... been running at that speed for ~ a year now.. and she never shuts the damn thing off.

 

[tknight]

Quote:

Originally Posted by Jpmboy

the 4th value is tRAS, and tRAS = CAS(CL)+tRCD+tRTP. So for your setup the correct timing is most likely 39.

3400c13 is on my x99 with 64GB of ram. The z170 boards run 3600 to 4133 with decent ram kits. Check the table in the OP for some settings.

Where did you get that formula for tRAS ?

Because even when I check my XMP Timings for my 3600mhz C16 kit, the tRAS set by G.Skill, does not add up to the sum of those figures.

 

[Bojamijams]

Quote:

Originally Posted by Jpmboy

you are confusing tRP with tRTP. Dl a copy of the latest Asrock Timing Configurator for your MB and post a snip back here.
And getting a 2400 kit o orun 3000 is not impossible, it just depends on the capabilities of the ram ICs.

Ahh yes you're right. I see now.

Here is what I'm currently running from that ASrock timings configurator

 

[Bojamijams]

Quote:

Originally Posted by tknight

Where did you get that formula for tRAS ?

Because even when I check my XMP Timings for my 3600mhz C16 kit, the tRAS set by G.Skill, does not add up to the sum of those figures.

Mine doesn't add up either. Math is 33 but my XMP sets it to 35. I'm thinking the formula is the MINIMUM and since it probably has very little effect on the actual speed, XMP sets it just a bit outside to facilitate compatibility

 

[tknight]

Quote:

Originally Posted by Bojamijams

Mine doesn't add up either. Math is 33 but my XMP sets it to 35. I'm thinking the formula is the MINIMUM and since it probably has very little effect on the actual speed, XMP sets it just a bit outside to facilitate compatibility

Personally I don't think that formula is applicable whatsoever, because as you can see below from my XMP timings, using that formula adds up to 44 and my tRAS is set to 36 by default from G.Skill and that is a considerable difference

From all the memory overclocking I have done with DDR4, lowering tRAS always gives greater performance every time and still maintains stability.

My kits XMP default timings.

 

[Bojamijams]

Quote:

Originally Posted by tknight

Personally I don't think that formula is applicable whatsoever, because as you can see below from my XMP timings, using that formula adds up to 44 and my tRAS is set to 36 by default from G.Skill and that is a considerable difference

From all the memory overclocking I have done with DDR4, lowering tRAS always gives greater performance every time and still maintains stability.

My kits XMP default timings.

Hmm you are correct. Well back to the drawing board I guess :/

 

[Jpmboy]

Quote:

Originally Posted by tknight

Where did you get that formula for tRAS ?

Because even when I check my XMP Timings for my 3600mhz C16 kit, the tRAS set by G.Skill, does not add up to the sum of those figures.

Depends on ther tRTP value assigned by the IMC. Posted 'bout this many times, the ras windows needs to be open for the entire time it takes to complete all three operations, and when this is set below the required window (it's all explained in the JDEC spec) the chipset subs in a value to correct the timings error... and you cannot interrogate this value. For vendors, there is no loss (or risk) to setting this value below the required window since the chipset will correct it. It is possible for our manual value to be an offset to a floor hence you see values below the sum. It's just how ram works.
Raja has posted about this many times over the past year(s).

Here are two:

No need for tRAS at 30. It's below the minimum time so the chipset will have to resort to some arbitrary timing. tRCD is the time it take to latch the row and transfer the data into the sense amps. CAS is the time it takes to find the column address have have the data ready for burst. Adding those two together brings you to 30 clocks. Each burst is 4 clock cycles in length. That brings you to 34. However, tRTP is set to 10. Which means that 40 clocks must elapse before tRAS elapses and the precharge command can be sent to transfer the data in the sense amps back into the dram cells. The minimum proper tRAS value for your setup is therefore 40 clocks.
All of the timings follow the same laws as DDR3 for minimum value, apart from tRRD_L which has a minimum spacing of 6 clocks

tRAS is the minimum time the row should be active. The row needs to be active for the entire duration it takes to perform tRCD, CAS and tRTP. Any lower and the chipset has to apply the minimum value arbitrarily - there may be an additional penalty for the collision as well.
So while it may look nice in screenshots to set tRAS to some low value (below the min threshold) in reality it is not helping and may be worse than setting the correct minimum value instead on relying on the IMC to correct the timing issue.

Believe me, many moons ago I examined JDEC timing schedules (charts) and it is indeed so. Looks good to have a low number, but may not actually be what the memory start-up tests end up with.

 

[[email protected]]

Quote:

Originally Posted by tknight

Where did you get that formula for tRAS ?

This diagram shows the RAS line held low until tRCD, CAS, and at least 2 clocks of data have been output. If the RAS line goes high before that, you're asking the memory controller to unlatch the row before you've got the requested data from it.

http://www.electronics.dit.ie/staff/tscarff/memory/DRAM_READ_CYCLE.jpg

What hangs in the balance at the board level - not the chipset level - is how the number of clocks you've entered for tRAS is interpreted. Most of the time, it should be a 1:1 relationship. Any timing set below the minimum "electrical" value is simply substituted so that the operation can be performed. The penalty of this is unknown, but there's usually a drop in performance at a certain point. Indirect realtionships to other timings can play a part in perceived performance gains (in this case it would likely be from tRC). Make no mistake that the law is real, it's just the interface layer that is sometimes fuzzy.

-Raja

 

[Jpmboy]

^^ yeah, what he said.

not sure if this is still valid:



Thanks for pointing out the interface layer issue.

 

[[email protected]]

Yeah, it's valid if you know what to look for. It shows a two column burst, not a single column read followed by a page close...

 

[Jpmboy]

Quote:

Originally Posted by [email protected]

Yeah, it's valid if you know what to look for. It shows a two column burst, not a single column read followed by a page close...

lol - no idea what it was.