What ram speed should I buy?1600,1866,2133,2400,2800,2933,3200 MHz

Memory is not just about speed, it is also about latency. Memory modules have their latency rated by a score (that references their timings). Generally the higher the frequency the memory operates at the higher the latency. Good 1600mhz memory is cl9 (Cas Latency) as standard, cl8 1600mhz will be quicker but also more expensive.

You need to balance these two factors when buying RAM. There are benchmarks showing that the performance in dual channel mode is optimal at 2x4gb 1600Mhz with as low a CAS as possible. 2x8gb is very good also, just not as cost<>performance effective.

G-skill ram is very good, especially if you want red modules. You shouldn't need to overclock your ram, most of the time it will make a small difference and make your memory controller work harder in the process.

CAD, Rendering, Adobe Suite, video editing all benefit from more RAM, faster RAM and lower CAS RAM.

As for whether faster RAM benefits Gaming there are a 1,00o answers cause there's a 1,000 games. Most RAM reviews are somewhat useless because they 1) don't look at multiple GFX card setups, 2) don't look at minimum fps and 3) draw conclusions from just 2 or 3 games. So does RAM speed / CAS matter in gaming ?

a) Yes
b) No
c) All of the above

The answer is c. First you start to see RAM Speed / CAS affecting more and more games in multiple GFX card situations, especially when looking at minimum fps. So if you have 1 card, you will be affected in less situations. Games like the STALKER series were affected by RAM speed as well as even PCI Bandwidth with 1 card. For 1 card users, the effect from RAM speed can vary widely.

Here we see that going from 1600 to 2400 with Metro LL (bottom of page) resulted in a whopping 0.1 fps increase in performance. But in F1, the difference is fps (top of page) is an incredible 11% (177/159).

How far to go up the speed ramp depends on not only how far up you go but where you live... here in US:

2x8GB of Gskill Ripjaws @ 1600 CAS 9 run about $125
2x8GB of Gskill Tridents @ 2400 CAS 10 run about $150

Is there anything else you could add you your PC that will produce an 11% improvement in fps (F1) for $25 ?
Does it make sense to spend $25 for virtually 0% improvement in fps (Metro LL) for $25 ?

So you see the dilemma. To my eyes, I'll spend the $25 tho in recent months its been as little as $7. Three weeks ago Mushkin 2400's w/ better timings than the Tridents were $131.

Going above 2400 just doesn't have a good ROI.... 2400 is reasonably priced cause yields are better tan they were way back when when DDR3 arrived. 2666 and up is still kinda low so prices are still high and the additional cost is much harder to justify.

As for what types, color doesn't matter though many peeps try to match their MoBo highlights or case color. My personal preference is for Mushkin Redlines .... they have the best timings, they use Hynix modules and they are reasonably priced ... they do have seasonal problems keeping up with demand.

The only think to watch out for is if you using a big air cooler, the tall toothy heat sinks tend to hit the coolers. My some bought LP (low profile) for his last build to miss the Phanteks PH-TC14PE and it tuned out he didn't need to, it would have fit just fine.

DDR-3 needs no extra cooling

I'd go the Trident X in 2400/10 for a 2x8GB set.

http://www.newegg.com/Product/Product.aspx?Item=N82E16820231589

The Mushkins Jack mentioned are good also, cost a bit more but negligible

http://www.newegg.com/Product/Product.aspx?Item=N82E16820226569

Very informative from JackNaylorPE! One final thought, do not use Seagate HDDs. Please add a WD or Hitachi or Samsung instead. You might save 20 bucks going for the Seagate but if it dies in 2 years you'll be really aggravated about the data loss.
Its also out of warranty so you will have a high capacity paperweight, like the 2000gb ST2000DM001 I have on my desk right now.

Seagate actually has a lower return rate than WD. These are reported failures between 6 and 12 months of operation. The first number is last reported period, the number in ( ) is the 6 month period before that.

http://www.hardware.fr/articles/927-6/disques-durs.html

- Seagate 0,69% (contre 0,86%)
- Western 0,93 (contre 1,13%)
- HGST 1,01% (contre 1,08%)
- Toshiba 1,29% (contre 1,02%)

Click to expand...

Lower return because their warranties are only 12 months and they don't accept them for replacement. After 18months your Seagate will have a good chance of dying.

I am quite familiar with the backblaze study but it has nothing to do with desktop drive environment. As for the backblaze study, I have to agree with the comments here ... has absolutely nothing to do with the desktop environment.

http://www.anandtech.com/show/8764/seagate-enterprise-nas-hdd-wd-red-pro-gets-a-competitor

Although I do appriciate the backblaze study, they make it CLEAR their failure rates for all the drives ranked are for a SPECIFIC application, one that none of the drives are designed for (these are consumer drives running in a cold-storage scenerio among other drives in RAID.) Almost none of those drives have firmware that supports staggared spinup, vibration monitoring, harmonic balancing, differential queuing, and so on.

To be fair, Backblaze is using regular desktop drives in environments they were not designed for: high vibration from 40+ drives in a pod. The results of that study are surely valid if you use the drives in such an environment.. but they don't necessarily apply to other environments.

The backblaze data also mentioned Seagate AND WD 5400 RPM drives that routinely failed nearly instantly. There's a lot of evidence that supports their environment is not suited towards certain drives. I don't think the average consumer should put much, if any, weight on those results.

Click to expand...

I have Seagate SSHDs all over the place, the oldest is going in 4 years...had an NAS with 7200.10's ... NAS died from a storm electrical surge, the drives came out and are still in use as backup drives (they are almost 9 years old.)

That is a fair point sir, especially about the 6-12month failures.

Is it not akin to stress testing a CPU though? We operate it in an environment it will not likely be in to make sure it will be stable. If the Hitachi drives survive in the extreme circumstances is that not better?

Reliability aside, because of Seagate's warranty policy their OEM drives only carry a 12month warranty. Which is awful for a hdd. I would buy the other brands on warranty length alone.

Ability. If you buy 1600 cas9's, that's what you got, maybe some OC to get them to 2133 cas11. Or maybe not. If you buy the trident 2400/10, you could run 1600, 1866, 2133 all day long at 1.5v, not needing to OC cpu until running the full 2400, so you have options.

Just as we are now recognizing that drives are set up for certain environments ....

Red = NAS
Purple = Surveillance
Black = Desktop

Drives are also designed for different uses.... why backblaze uses desktop drives in "cold storage scenario " is beyond me. Many consumer drives have a feature called "head parking" which Seagate has used very aggressively ever since the 7200.10. This places the drive head in a "parked" position when not being used so that is least likely to cause damage to the drive if it is subject to vibration (dropped lappie, bumped desk etc,). This is a bonus for consumers, but a detriment in a server farm ... its been reported that the drives are rated for like half a million cycles. In a server farm you can wind up doing 50,000 cycles a month or more and every time the drive tries to protect itself, it's yanked back into service. Desktops don't see that type of usage.... you open a file in your word processor, the drives idle till it gets saved..... browsing even is not that bigga deal especially if ya have a decent amount of RAM.

From the same article

The reason the Hitachi drives do so well in the Backblaze study is because Hitachi never even implemented the SMART 0xC1 command.

Click to expand...

I have never bought one of these CCTV OEM drives and I don't know why anyone would for a desktop. I haven't bought a HD in 4 years (I have bought a ton of SSHDs).... but last I recall the barracuda's were 2 or 3 years. Back in the day 3 and 5 was the norm. .... but price has become the primary factor w/ system builders. You always see the advertisements naming the CPU and the GFX maybe even memory but no one every puts the HD on the placard other than "1TB HD". So the vendors have been going cheap to maintain market share and one way to reduce costs is to reduce the cost of the warranty. Seagate and other vendors are producing drives to meet this demand but just cause they make a drive for this market, it's not like they are no longer serving the market we are interested in.

The WD Blue 1 TB and the Seagate Barracuda 1 TB have the exact same price, have the exact same warranty and have the exact same % of 1 egg reviews on newegg. There really is no justification for saying one brand is worse than any other brand. What ya can say is that every vendor produces great drive models and every vendor produces duds. One brand might have a good run for a few sales periods and then 1 bad model line can bring the entire average crashing down. When storagreview.com was maintaining the storage reliability database , Seagate had both the very best and the very best drive model. Just like PSUs....

Does Corsair make great PSUs ?
Does Corsair make crappy PSUs ?

Yes (HX) and yes (CX)

As I said, I haven't bought an HD in 4 years..... using strictly SSHDs. I have (2) 256 GB SSds an (2) 2 TB SSHDs in my personal builds....my son has one of each. All the office desktops and lappies I have purchased in last 3 years have SSHDs ... no failures to date.

Even when ya have a SSD, they rock for gaming. If you are playing Far Cry 3 all the most used files get moved to the SSD portion .... then when ya done w/ FC3 and move to FC4, the FC3 files get moved off and the FC4 take their place. Fior anyone who can't fit an SSD and a HD in their budget the SSHD is the obvious choice. Boot time on the SSHD (16.5) is less than a second slower than on the SSD (15.6).

I loved the WD Blacks but the SSHDs are 50% faster than the Blacks and almost twice as fast as all the other colors. Now WD has an SSHD but last I heard it was only being sold to OEMs

Thank you for the detailed response! That was very informative.

My personal experience still leans me away from Seagate though. I have Samsungs 6 years old and flawless but had 2 2000Gb Seagates die in the last 12 months. One was 18months old, the other was 3 months old.

SSHDs do seem like the way forward

Ford, Chevy... Intel, AMD.... WD, Seagate. Brand is useless. The item in question is everything. If it sucks, it sucks, if it's good, its good. That's all there really is to it.

For 2400 A slight CPU OC is often required, though it depends on your individual 4790K some are stronger than others.

I have not had an instance of XMP requiring a CPU overclock......it does require a bit more CPU and DRAM voltage from 44 on up (this will vary for every CPU of course .... the test CPU wasn't a particularly good sample)

Here's the results of my testing on 4770k on the Muskin 2400s

Legend:

42/A/A/Auto = 42 CPU Multiplier / Auto Max cache / Auto Min Cache / Auto DRAM Setting (1600)
46/43/A//XMP = 46 CPU Multiplier / 43 Max cache / Auto Min Cache / XMP DRAM Setting (2400)
DRAM Setting: Auto (1600 / DRAM Voltage = 1.65)

================================================

All BIOS Settings Stock except for XMP: All stable at Vcore = Auto / DRAM Setting = Auto (Observed = 1.671)

These were the other stable settings in the test. Note that when I tested and failed at 1.65 DRAM voltage I went back and retested at 1.700 ... someday I will see if there's any stable settings in between

42/A/A/Auto: Vcore = 1.200 / DRAM Setting = Auto (Observed = 1.671)
42/42/42/Auto: Vcore = 1.200 / DRAM Setting = Auto (Observed = 1.671)
42/42/42/XMP: Vcore = 1.200 / DRAM Setting = Auto (Observed = 1.671)

44/A/A/Auto: Vcore = 1.250 / DRAM Setting = Auto (Observed = 1.687)
44/44/44/Auto: Vcore = 1.250 / DRAM Setting = Auto (Observed = 1.687)
44/44/XMP: Vcore = 1.260 / DRAM Setting = 1.700 (Observed = 1.712)

46/A/A/Auto: Vcore = 1.375 / DRAM Setting = Auto (Observed = 1.687)
46/43/43/Auto: Vcore = 1.380 / DRAM Setting = Auto (Observed = 1.687)
46/43/43/XMP: Vcore = 1.385 / DRAM Setting = 1.700 (Observed = 1.712)

CPU temps are 70, 70, 67, 58 at last setting

Haswell cpus have a strong affinity towards ram at 1.5v or lower, its even gone so far as to be a incompatibility in pcpartpicker to use a Haswell cpu and 1.65v ram. About 1/2 the ram at 2133, and most of the 1866 runs at 1.5v. The other 1/2 2133 and higher runs at 1.65v. Because of the demands of OC by the cpu, it 'raises the bar' so to speak on ram voltage, and will allow the ram to reach its full potential at rated voltage. This only affects Haswells and later, as the ram speed/voltages are on chip, whereas the speed/voltages for Ivy Bridge and earlier are on the motherboard.

__________________________

PC partpicker identifies any DRAM for Intel CPUs above about 1.575 as a flag, because Intel recommends 1600 at 1.5, there is no danger of running DRAM at 1.65, in fact Intel even has certification list for high end DRAM and the bulk of all of it is 1.65, with some 1.6 thrown in also.

http://www.intel.com/content/www/us/en/gaming/xmp-for-core-processors.html

Additionally for info purposes on Haswell the MC (memory controller) is indeed in the CPU, but contrary to the above - it also was in Ivy Bridge, and backe to Sandy bridge and back to the socket 1366 and back to the socket 1156, in other words Haswell operates with DRAM as the last few previous generations of Intel CPUs have

I stand corrected. It's the VRM's I was thinking about.

I was indicating that the CPU 'may' need an OC to run 2400 sticks at 2400, it appears all the testing you did was with an OC on the CPU, the 4770K base is 3.5 raising the multiplier to even 3.7, or 4, or 4.2 is OCing the CPU.

In the instance here the 4790K has a base of 4GHz, yet some still need an OC to run 2400 at 2400, normally for those that do it's nominal

I have a question for anyone who can answer me, I bought a Z97 Gigabyte mobo and an I7-4790k cpu , now I also bought 2400 mhz DDR3 Corsair Vengeance Ram , 4 x 8 GB sticks, this is the board I bought :GA-Z97X-UD3H. under the specs for RAM it says this,DDR3 3100*(*O.C.)/ 3000*/ 2933*/ 2800*/ 2666*/ 2600*/ 2500*/ 2400*/ 2200*/ 2133*/ 2000*/ 1866*/ 1800*/ 1600/ 1333 , What I take this to mean is the two 'standard memory speeds of RAM should be 1600 or 1300 mhz and all the other speeds are 'overclocked memory'. My question is , I am building a gaming rig for the first time and when I put this 2400 mhz memory into the RAM slot and go to boot up the BIOS for the first time will it even recognize this memory , being that it is not 1600 mhz but 2400 mhz. In other words would I have to buy 1600 mhz memory first just to post with this board and then later on once I build the thing would I be able to use the 2400 mhz sticks? I'd also like to add that Gigabyte provides a pdf QVL list of qualified vendors of RAM for this board , what exactly does this list mean for RAM that is not on it but is still DDR3 RAM ? Does it mean I can only use what is on the QVL list?

Pc's have a set default. For a gpu this is VGA. For a cpu it's the factory set speed. For ddr3 ram it's usually 1600 or 1333. And it'll stay that way until you change it. With gpus, you add specie drivers for that card, cpus get turbo or OC and ram has xmp or higher frequency.

QVL is the qualified vendors list. It's a list of the exact models of ram that the motherboard has been tested with and found to work. With companies like g-skill, they test their ram on many boards, so is almost always a safe bet, but this is their testing, not the motherboards test, so different models won't show on the QVL.

your reply does not answer this simple question, what speed ram should he buy....simple....what speed ram will not work....basically this is a novice question. Look at it this way...you have your mobo on the antistatic bag plugged in to the power supply , ram inserted ( pay attention because this is the crux of the query.....lol) , a monitor plugged into the mobo, nothing else.....(maybe the stock cpu cooler just to be safe....) you short those power switch pins ( or press the on board power test button if you have one) and then....WHAT HAPPENS , Error code 55 maybe? there are so many DDR3 memory modules out there with different speeds already on them what is that all about? are they basically made as 1600 mhz modules and Corsair or Kingston or Crucial over clock them for you on their factory rigs and then sell them to beginning pc builders to install in a new board before that sucker is even in the case only to get an 'error code 55' ( that means there is no memory on the board....by the way....)...so what should this person buy for said 'puter to make it work? Basically it seems that the industry seems to sell this over clocked memory to rip people off or confuse them or both. Then we have water cooling ...which is another matter entirely. A new builder can literally crush the cpu into the socket with those bulky , tarded 'pumps' that go right on top of the most necessary part of the entire computer , two of them actually , the cpu and the mobo socket it interfaces with.