All Memory is Not Created Equal - HyperX 2133 CL12 versus Ripjaws 2133 CL11 versus Vengeance 2133 CL11

You got me excited

Where can I get them, model or SKU number?

I will probably go with 2x8GB as I don't feel like paying to upgrade Win7 to Pro. Although I am seeing higher utilization with my 850 Evo with RAPID enabled (14.6 GB today).

 

I hear ya. I got mine from Newegg and they were manufactured in February of this year. Too lazy to look up the lot at the moment. I'll put a link if you'd like, but it is a simple search on their site. I went to 32gb because I was regularly 14.5-15.6gb. Now, not an issue.

Sent from my SM-G900P using Tapatalk

 

Is this the one? What kind of performance gains can I expect, both synthetic and in gaming?

http://m.newegg.com/Product/index?itemnumber=N82E16820104562

 

http://m.newegg.com/Product/index?itemnumber=01N-01GF-00002

You want the set instead of one at a time. As for performance, depends on the program, your overclock, and where you're at right now. Meanwhile, ram upgrades generally do less than other upgrades, like the cpu. There are tons of pages dedicated to that question. ..

Sent from my SM-G900P using Tapatalk

 

We also have some comparisons inside this thread. Generally even if you have an synthetic increase of about 50% you'll feel nearly no difference in framerates (0-5% increase). If you want more frames, get a better graphics card. Your CPU is fast enough, vulcan.

 

Thanks guys, yeah I think I'm going to put that $100 or so towards moving from the R2 to a new chassis altogether.

With the R2, what has been holding me back from 980M is the throttling issue necessitating having to enter BIOS and switch back and forth from IGFX upon each and every boot and I'm not sure if this is still the case but having to move from Win 7 to 10 and also having to rely on the modification of newer drivers to work with that card. In the end I decided to save for a completely new build, ideally a Clevo / Sager NP 9870 G with a desktop 1080 or a used unit with a single desktop 980 for a bargain.

I have been keeping my eye open for 780 M but the lowest I've seen that for is around $400 each on ebay used, and who knows their condition.

Too bad 980M has turned out to be a nearly non-viable upgrade for the M18x R2. It would be nice if it could simply be installed and work with newer drivers as though the unit were a desktop PC going from GTX 680 SLI to a single GTX 980. I really don't understand why we don't have this ability and why 980M only works correctly doing the IGFX trick and why the drivers need to be modified.

Is it a Dell profit oriented issue?

Hey manufacturers, if youre listening, stop trying to squeeze blood from a stone, give us modular systems with the full ability to upgrade our mobile GPU's down the road just as though these were desktop PC's.

Paid $3500 for a system that cannot be upgraded. That's pretty F'ing lame.

 

So, can you SPD flash G.Skill sodimms with Thaiphoon Burner? I'm considering picking up a 32GB (4x8GB) 2133 G.Skill, but need to know I can flash them. They use samsung chips whereas my current 4x8GB 2133 HyperX use Hynix and I have running at 2400 GHz 11-13-14-28-2N that I flashed on the SPD with TB. I'm just wanting to know if I can manipulate the SPD and XMP profiles with G.Skill like I can Kingston before I purchase them (or if they are write protected). The slowest of them will be tightened at 2133 for my old P170HM while the faster running at 2400 (hopefully with tighter timings or with a faster speed) will be in my P770ZM.

 

Yes, they are flashable. No need to worry about write protection. My G.SKILL DDR3 and DDR4 SO-DIMM modules have been flashed multiple times with Thaiphoon Burner. In fact, the current modules in my P750ZM were modded with the values from you SPD dump for 2400. Before that, they had a different 2400 MHz profile custom tweaked for the P570WM (which had timings that did not work in the P750ZM).

 

When you run AIDA64's memory benchmark, your CPU speed doesn't affect it, right? I assume you close everything else running first?

 

I got more effect from the cache multiplier than core speed, but it is only a couple percent (1% usually or less) at most in difference. Obviously, cache effects your cache scores.

Closing everything, minimizing plugged in peripherals, etc., may help some, but not alot (except in cache scores), although it definitely helps by 1% in winsat (at least going by my observations). I'll do a run with everything closed and matching his CPU and cache speeds, then post.

Sent from my SM-G900P using Tapatalk

 

So, here they are:
48core 40 cache:

45 core 43 cache:

48 core 43 cache:


As you can see, it helps cache scores and latency the most. Performed all of these in diagnostic mode Win 7 Pro x64.

Edit: in fact, the variance in ram speed is likely variance in OS background behavior and not attributable to the multiplier on cache or core at all. My original 45 core 40 cache was on a regular boot with programs closed, not in diagnostic mode like the other three. Further difference between Mr. Fox's scores and mine would be differences in secondary timings or other factors that differ between G.Skill and Kingston HyperX ram (which I believe his G.Skill to have samsung chips and I know my Kingston have hynix)...

 

Eh, honestly, I only cared about the memory report. I wanted to see if my 11-11-11-31-376-T1 would get similar results to his 2400MHz 11-13-14-28-226-T2. I might run it later when I'm less lazy.

Interesting how the cache affects RAM latency.

 

So my boss/mentor has told me for years, faster, or rather higher clocks don't always equate to more speed. You, how ever are the first to have some sort of proof. BASICALLY just like how no 2 chips are the same, hence yields issues = hardware binning, no 2 chips clock or run the same.

SUPPOSEDLY out of all the various standards and APIs throughout the years, there is actually some basic redundancy checks and error checks going on but, not at the same level as full registered and error correcting server RAM but, enough so, that there is a point where you can pass stress tests and have a stable system at the expense of speed....

...ergo , ipso-facto the best way to go about performance gains is to actually test them, weather it be different ram or an overclock or timing adjustments etc etc.

THIS IS NOT FACT!!! This more of pseudo science so, take it with a grain of salt. HOWEVER as Mr. Foxes claims/evidence might show, there is some sort of merit to such claims. As long as this gentlemen has been in the business I'm not one to completely dismiss such claims especially after Mr. foxes' claims.

 

There are actually formulas to estimate speed and latency for the ability of ram, and you are correct, slower ram with tighter timings can run faster than faster ram with looser timings, but only to a degree. If the same latency exists at different speeds (say CL11 for both 2133 and 2400, simplifying the discussion of course), then the 2400 will win. If you have different latencies (say CL8 at 2133 and CL11 for 2400), then the 2133 is likely to come out on top. You can always test, as you suggested, but getting a rough idea before you start, such as reviews of ram and user's experience from forums, will help to narrow down the choices based on ram capacity, speed, and timings.

 

So, an update on not all RAM is created equal. I recently purchased some G.Skill 2133MHz 11-11-11-31 ram that was manufactured in August. I, once again, overclocked a 2133MHz set to 2400MHz, this time with 12-13-13-31-278 CR2. After comparing the results, I stayed with the Kingston set I previously flashed. The interesting thing was that AIDA64 had the G.Skill winning on speed. So did winsat mem. It was a slim win, but a win nonetheless (except on latency in AIDA64). But, even with this, it lost in MaxxMem2 and (where it really counts) SuperPi32M (by over 2 seconds)! Fully attribute the loss to the latency, but here are the pics!
Kingston

G.Skill


So, is seems I forgot to do screenshots of the SuperPi32M runs, BUT, the results were Kingston at 7M14.xxxS and G.Skill at 7M16.8xxS. Now, this is with a full CL setting slower AND much HIGHER tRFC (278 vs 228 on the Kingston). I wanted to point this out so that people would know, NOT ALL MEMORY IS CREATED EQUAL! My BIOS settings for 2400MHz on the G.Skill kit to follow (it was a 4x8GB Kit)...

 

Can Anyone recommend me which of these to get for my machine, I read the thread again and found Mr. Fox's reply of using the above stock 1600+ doesnt add up much with the Ranger machine. But I'm stuck with an 8GB RAM of 2 sticks, Want to upgrade to 16GB 2 sticks so want to get some quality RAM.

These are my options -
Corsair - CL9 1600MHz / CL11 1866MHz
HyperX - CL9 1600MHz / CL11 1866MHz
G.Skill - CL9 1600MHz (Cheapest of all, Maybe those kingston and corsair will also drop on 24) / CL11 1866MHz / the only RAM which has CL10 1866MHz kit is GSkill but only 32GB kit.

Thanks !!

 

why not go all out with 2133 mhz ram? as far as ive seen, the kingston sticks are most likely to hit 2400 mhz stable

Sent from my Huawei Mate 8 NXT-AL10