The Truth About Hynix MFR-based Memory Kits - Overclockers.com's G.Skill TridentX 8GB DDR3-2933 Memory Kit Review

I must admit I have been getting increasingly annoyed with the noise regarding the Hynix MFR-based memory kits. As you know, the latest memory overclocking results - everything over DDR3-4000- were all achieved with a very specific memory IC: Hynix MFR. Now although I love seeing these insane memory frequencies, as an overclocker I do want people to understand that the highest clocks do not represent the highest level of performance. The highly clocked MFR-based kits are not slower because of the memory timings, but mainly because they are single sided. That means, only one side of the memory stick has ICs. So unlike with spending USD $1000 on a GTX Titan or a Core i7 3970X, spending money on the highest priced memory kit will not get you the best performance. Hynix MFR is for memory what the AMD FX-9590 is for processors. Have you seen any positive review, 90%+, or highly recommended award for that CPU?

Since Computex many reviews of Hynix MFR based memory appeared online. I must say most of them are very disappointing, as most kits get awards for these horribly priced underwhelming performance memory kits. The performance of MFR-based memory is very low. Have a look at this performance chart from TechpowerUP on the right.

As you can see, a DDR3-2933C11 MFR memory kit is slower than a DDR3-2400C11(!) kit. That is nothing short of sad.

That is the main reason why I would like to highlight the review over at Overclockers.com. They also reviewed an MFR-based memory kit, but did not give a review or any other "recommended,” "must have" or "this is amazing" award or paragraph. They said exactly how things are. To quote,

The 8 GB G.Skill TridentX DDR3-2600 kit used for comparison here is $139.99 on Newegg. That’s 37.8% of the price of this DDR3-2933 kit. Not nearly 40% less than the price of the kit being reviewed, it is under 40% of its price. Oh, and it performs better. So for anyone out there that wants memory currently on the market for performance, go for that kit. G.Skill will be happy; they’ll still be getting your money, just not as much of it.

That’s not the whole story though. For a subset of a subset of overclockers that really enjoy memory overclocking (of which yours truly is one), this kit is tailor made for you. If you find yourself in that subset that is willing to spend a good sum of money purely for the fun of overclocking your memory to DDR3-3000 and beyond – well beyond – this kit has your name written all over it. Believe me it is A. Lot. Of. Fun. Fun in spades. This is a memory geek’s dream at half the price of those DDR3-3000 kits. For the rest of you, those that value performance over frequency numbers, have a look at the frequencies this kit can achieve. Then say “Ooohh” and “Aaaahhh”. Then go buy a cheaper kit.

Because this kit does what it says it will do and it can overclock to the moon – with regard to frequency only – as designed, I will not give it a “meh.” However, because it fails to do the one thing that higher-priced, higher-spec’ed memory should do – out-perform its cheaper-by-a-long-shot little brother – it doesn’t get to be Overclockers Approved either. Sorry G.Skill – and all the other manufacturers putting out these crazy high frequency Hynix MFR kits like Corsair, Avexir, Adata, etc – you’ve got to bring the performance to go with the MHz. Without it, your kit is dead in the water aside from those that are either really into memory overclocking (see: subset of a subset) or those that are compensating for something else.

... and I take my hat off. For your information, I did link our partner G.SKILL to this review and they were pleased to see a media outlet being honest and correct about the memory kit.

I realize I might not be making a lot of friends with this news post as it seems that most of the media, journalists and marketing teams would rather have people be quiet about the performance of MFR-based memory. However, right is right and wrong is wrong. Anyone can put Hynix MFR on a memory kit and get to DDR3-3400 with a slight voltage boost, the right timings, and a good CPU. In fact, creating a highly clocked memory kit is not rocket science, it is only a matter of binning as much ICs as possible. That is exactly why they are so pricey - it takes a lot of man-hours to go through 30,000 ICs to find those that can to DDR3-3000 at 1.65V.

Honestly speaking, there are only two things I find impressive about MFR-based kits. One, getting the highest frequency record, something mostly (only) extreme overclockers are interested in. Two, having the highest clocked mass-production memory kit available in the market. None of that exclusive one-kit press release or incredibly limited sample stuff. Looking at Newegg, I only see G.SKILL having DDR3-2800+ memory kits in stock. At the Geizhals price comparison site, I find one Avexir and a couple G.SKILL sticks in stock. Others are unlisted or unavailable. Great work, both companies deserve an award for the work.


39

Belgium Massman says:

Probably the longest newspost on the frontpage. Sorry, had to let it aaaall out. Can't stand how review sites would give average marks for the AMD FX-9590 and be very positive for MFR-based kits. Just so it's clear to everyone, congrats to all the folks involved with hitting over DDR3-4200 (incl; memory vendors, mainboard vendors and overclockers). Congrats to the memory vendors getting those highly clocked kits in mass-production. Congrats to media that don't recommend these kits to their audience for anything but high frequency memory.

TaPaKaH says:

I understand your frustration but please be aware that 4GB high-rated modules is not the only place where MFR is used, as the newspost seems to claim.

If you're talking high-end memory (from daily user's perspective) then you can't ignore 8Gb modules. Obviously, those will always be dual-sided regardless of which ICs you use, hence the "single-sided argument" will not hold. At DDR3-2133+ there are only two choices for ICs - Samsung revision B and Hynix MFR. With Samsung not being capable of consistently doing above 1200 10-10-12, MFR is actually a better choice.

Now let's go to sub-100Euro 2x4Gb kits. You say (in the other thread) that 2400C9 kits will run circles around MFR, but have you actually had a retail set yourself? I've had and seen quite a few of those being so bad that I believe those wouldn't be able to beat decent MFR (that are "guaranteed" by Team on 2666C11 for same sub-100Euro) in a 32M low-clock competition.

Bottom line is that MFR don't make sense on high-rated 4Gb modules, but as you said - it's a subset of a subset and doesn't mean that ICs are bad in general.

Belgium Massman says:

You are right, Sam. I'm probably focusing too much on this high-end part of the scope. I understand your point of view and I agree for the most part. I would want to make a sidenote that the market for high-density configurations that absolutely require 8GB memory sticks for, I assume, 32GB or more, is pretty small as well. For most users, you're fine with 2x4GB or cheap(er) 4x4GB if you need more. The point of the news post is more to highlight the somewhat odd conclusions of a lot of the single-sided MFR reviews, not to dismiss Hynix MFR as a complete failure alltogether. MFR has its place in the market, especially considering Samsung and others are shifting production focus to mobile, leaving not too much options for high-end memory. It's becoming almost exclusively the only option for manufacturing high-end memory at the moment. Bottom line, the general lack of outrage about the high-dollar low-performance kits leaves me with a very bad taste in my mouth. The GTX Titan comes at a USD $1000 price-point, but for that you actually get the highest performance. The same goes for the Core i7 3970X - if you talk about out-of-the-box specs. Across the web there are many negative comments on the USD $800 FX-9590 as the performance you get for 5GHz turbo (the highest stock clock for CPU) is well below that of a mainstream CPU. I don't understand why there's so few negative comments about high-clock MFR.

K404 says:

Kinda off-topic....are all these extra symbols:

. That’s 37.8% of they’ll still That’s n beyond – well beyond – this geek’s dr say “Ooohh� and “Aaaahhh�. moon – only – a “meh.� higher-spec’ed memory should do – brother – it doesn’t get G.Skill – etc – you’
A result from copy-pasting from a different forum code to HWBots? Back on-topic. ok.... so 2x4GB is a writeoff at high speed and 2x8GB won't hit the frequencies and/or timings. We gotta pick carefully :D The next generation of memory will (theoretically) contain everything needed for SPi32M in ONE memory chip. Terrifying! We need to keep in mind the list of memory that is dual-sided at 4GB per stick :)

TaPaKaH says:

Massman said: I don't understand why there's so few negative comments about high-clock MFR.
I guess the main reason is that not enough 'decent' people can actually get a hold of a kit. Corsair didn't send out any 3000C12 press samples, G.Skill seem to be very careful with 2933C12/3000C12 as well (I asked for some mid-end stuff and a 2933C12 a month ago, mid-end stuff arrived in two days and had consecutive serial numbers meaning that G.Skill "cared", 2933C12 - no reply). There are some Adata 2800C12 samples floating around but I guess all reviews will mark those kits down for redonkeyless pricing before they even get to performance evaluations.
Massman said: The GTX Titan comes at a USD $1000 price-point, but for that you actually get the highest performance. The same goes for the Core i7 3970X - if you talk about out-of-the-box specs. Across the web there are many negative comments on the USD $800 FX-9590 as the performance you get for 5GHz turbo (the highest stock clock for CPU) is well below that of a mainstream CPU. I don't understand why there's so few negative comments about high-clock MFR.
The most expensive 4x8Gb 3000C12 set also features the best possible combination of density and performance, so? :)
High-rated 2x4Gb MFR sets can be compared to low-end VGAs with huge amounts of slow VRAM, i.e. not a high-end branch that has gone pointless beyond some point in its evolution.

Romania Alex@ro says:

Sam OCX said: I've had and seen quite a few of those being so bad that I believe those wouldn't be able to beat decent MFR (that are "guaranteed" by Team on 2666C11 for same sub-100Euro) in a 32M low-clock competition.




Excuse me?Single-sided MFR are miles behind double-sided Samsung(2400 cas9).Also there is potential for overclock,don't tell me that with a 2400 9-11-11 rated kit you can't reach 2500 9-11-11 or 2600+ 10-12-12 with 1.8V+ cause i have a very hard time believing that ;)

TaPaKaH says:

Alex@ro said: Excuse me?Single-sided MFR are miles behind double-sided Samsung(2400 cas9).Also there is potential for overclock,don't tell me that with a 2400 9-11-11 rated kit you can't reach 2500 9-11-11 or 2600+ 10-12-12 with 1.8V+ cause i have a very hard time believing that ;)
I've had three mid-2013-made GSKill 2400C9/2600C10 sets, neither of which could do 1333 CL9 or 1350+ CL10 with any voltage. websmile has had a recent 2600C10 set that, by his words, wouldn't run 1333 CL10. Such kits might have a hard time doing sub 6min05 at 5GHz.

K404 says:

Sam OCX said: I've had three mid-2013-made GSKill 2400C9/2600C10 sets, neither of which could do 1333 CL9 or 1350+ CL10 with any voltage. websmile has had a recent 2600C10 set that, by his words, wouldn't run 1333 CL10. Such kits might have a hard time doing sub 6min05 at 5GHz.


Ah, now you understand more of what I was saying last week ;)

Romania Alex@ro says:

Sam OCX said: I've had three mid-2013-made GSKill 2400C9/2600C10 sets, neither of which could do 1333 CL9 or 1350+ CL10 with any voltage. websmile has had a recent 2600C10 set that, by his words, wouldn't run 1333 CL10. Such kits might have a hard time doing sub 6min05 at 5GHz.


Even at 2666 C10 sub 6.05 is doable,a optimized result with 2860 9-12-12 scores 5.58.2-5,so you can add even 5 seconds and you still get 6.03.~5.

In my experience every 50 mhz counts as 0.5 sec to 32M,so make 2700 9-12-12 even let's say 2 sec slower that 2860,add another 0.5-0.6(which C-N proved on xs that is the diff from C10 to C9) ,make it round even 3 seconds and you get 6.01.5xx.Still let's say you run looser tertiary and secondaries,even 2666 c10,hell even 2600 c10 and it is still faster than fastest Hynix MFR run(dumo's 6.03.750 at almost 3100 11-14-13 which is faaar more than team 2666c11 MFR kit's will ever do).

All in all,single-sided memory sucks for 32M so bad it is not even funny....

United States hokiealumnus says:

Thanks Massman, I'm flattered to have been mentioned on the bot's homepage. :) FWIW, ADATA now has their 2800 single-sided MFR kit in stock at Newegg in addition to G.Skill. I'm working to wrap up this review shortly. Hopefully ADATA takes the conclusion as well as G.Skill did, because results are exactly the same - high clocks, but no performance increase.

Belgium Massman says:

Looking good. Quite pricey, though. I'd love to see some of this good quality MFR end up at the same price-point of DDR3-2400C11. If you can get a kit that goes up from 2400C11 to 3400C1x with a minor voltage bump, that'd be more interesting from a price/performance point of view.

TaPaKaH says:

Massman said: I'd love to see some of this good quality MFR end up at the same price-point of DDR3-2400C11. If you can get a kit that goes up from 2400C11 to 3400C1x with a minor voltage bump, that'd be more interesting from a price/performance point of view.
2133C11/2400C11 is a loose spec, doesn't guarantee good IC quality. 2400C10, however, does and some makers like Corsair and Patriot actually utilise MFR on such specs.

South Africa Vivi says:

Massman said: I don't understand why there's so few negative comments about high-clock MFR.

cause no one is that crazy to buy it, and people who get sent kits wont speak bad of it :P

United States dumo says:

In the era of haswell oc, "one ram chip fits all" way of thinkin' is no longer work

MFR, HCH9 and PSC/BBSE all served very specific purpose when use for benching

United States sin0822 says:

as long as the reviewer does his job correcyl and benches the memroy consistently the single sided HFR will lose out to cheaper/slower kits, if this isn't shown then that is a HUGE red flag. However the OC potential is HUGE, i mean you are taking a 2933 kit and doing 3.4ghz ,2.93ghz to 3.4ghz, um that is impressive and provides giant epeens, it is worth something. Not everything, but something. Also you need to focus on your target audience TPU's target audience is not the same as overclockers.com can focus on benching and such, while TPu will focus on real world applications, and while they are the same, high speed is always cool. Having one of the fastest kits on the market regardless of performance is still a big deal. I like TPU b/c they link my reviews to my site :D

United States hokiealumnus says:

I think it's cool, but not worth the price. It goes back to why I started overclocking in the first place - to get more out of the hardware I had. To me, overclocking is not only about getting records, though there is that component. It's also about getting the most out of your hardware for a given budget. These kits have a 'wow' factor for frequency, but they do not even increase performance a little bit.

They're on the market, and that's great, but to actually expect people to spend $370 on an 8GB kit of memory, you need to improve performance. Otherwise, it breaks most people's budgets on what doesn't help them - indeed, it hurts them. Let me know next time you see someone running DDR3-3400 at a CPU frequency worth mentioning.

It's very, very difficult to recommend 99% of overclockers (which are already a subset of the market as a whole) to buy a kit like this rather than to get a cheaper kit that out-performs one of these and spend the extra $230 on a better GPU.

EDIT - I also think you missed Massman's point about TPU. They recorded worse performance, just like we did....then on their final page, titled "Value and Conclusion" (emphasis mine), they gave it a 9.0 and a "Highly Recommended" award.

Norway knopflerbruce says:

I wonder how a setup with 4x2GB set of Hyper/PSC/BBSE would perform compared to these newer 2x4GB ones. Anyone compared these configs?

United States sin0822 says:

yea i didn't read either review honestly haha, i don't have time and I have been sick. But I see where the issue is. Also remember that 2666c10 16GB kit, top of the line at the time, cost what, $900? Then when others came out with 2666 kits the price dropped by half, same wil happen here, but yea ATM its not good Price/performance but is good price/epeen which is pretty big for some OCers. it depends what overclocker you are referencing too, the ones that come on here and do subzero or those who are on the forums like hardocp, andatech, or ocn and who would rather have a faster memory speed than the performance. There is a MAJOR divide in the OC community, very hard to provide for both sections with reviews on certain products like this lack performance/price but which do give epeen performance/price. Id probably have given it a 7 or 8/10, and I have no memory loyalty. If you want to compete at 4ghz you need this type of kit if you don't have industry hookups, and there are normal guys out there doing 4ghz OCes pretty easily with kits like this, an that is something that should earn points b/c this memory can do something special. you gotta realize there are guys out there who go for max frequency instead of max performance, people who enjoy the rush of max clocks rather than beating someone in superpi by tweaking the shit out of an OS. I am guessing if your audience was HWBot then this kit would get a higher rating than if it was for general population, however the other way around might also work. Always two ways to skin a cat, believe me I have skinned cats before(for science in labs, its quite gross)

Belgium Massman says:

sin0822 said: I am guessing if your audience was HWBot then this kit would get a higher rating than if it was for general population, however the other way around might also work.

Always two ways to skin a cat, believe me I have skinned cats before(for science in labs, its quite gross)


The appreciation for a kit might be higher, but that doesn't mean it's "more" recommended. Every vendor can make a memory kit with single sided MFR and claim high overclocking potential. It's the nature of these ICs. In my opinion, there are three (and not two as I mentioned in the news article) things that set kits/vendors apart:

- having the highest clocked mass-production SKU (= "binning quality")
- having the overall overclock record ("engineering/bin quality")
- offer best price/perf or price/oc

The bar for highest retail-available mass-production SKU is currently set at DDR3-3000. That's a thumbs up to the vendors. The products itself are definitely not recommended or must have since they are too pricey and come with low performance. Anything below DDR3-3000 is not impressive. The best price/perf or price/oc comes from comparing a SS-MFR with a similarly priced other IC kit. As Sam already mentioned, it's possible that at the lower end of the spectrum, the overclockability of MFR is so good it will surpass other kits in performance with higher clocks. If SS-MFR offers the best price/performance after OC, that should be recommended obviously.

The highest overall overclock record is somewhat of a sensitive topic since we all know how easy it is to re-flash an SPD with a different vendor name or even product SKU name. If the validation says (just an example) Vengeance Extreme, we can't know if it's actually Vengeance Pro or TridentX. So just like with the highest mass-production bin, any award or recommendation should actually go to the engineering team rather than the product itself.

Anyway, the point of the news post is to emphesize the silly "recommended" or "must have" for high-clock single-sided MFR when the performance is so low.

United States hokiealumnus says:

My other MFR kit review: [CENTER]
[/CENTER]

Single-sided Hynix MFR ICs have been making the rounds through various manufacturers recently. I didn't give G.Skill a pass for them when they came through last time. They just don't have the performance to go with the MHz due to the timing compromises needed for these kinds of frequencies. ADATA is here to give it a try this time. Let's hope they reverse the performance sacrifices made for the MHz. Hey, if nothing else, MFR kits are always fun to see how far they can be pushed!
... Return to article to continue reading.

Belgium Massman says:

Cool! I like the conclusion (again) - well-balanced. I like how you also highlight the kit is on par with the other ones. Nice!

United States hokiealumnus says:

Thanks! (Thanks for the FB share as well.) :)

United States sin0822 says:

nice, i am gonna get some single sided MFR too soon

Belgium Massman says:

Adata press release: http://www.techpowerup.com/188040/adata-launches-newest-xpg-v2-3100-overclocking-memory.html This ...

This new product's dual channel is designed to bring ultimate performance to gamers
This unprecedented speed will gratify those who always seek an edge in boosting system performance.
Not only can speeds of 3100MHz be reached, but bandwidth can attain rates of up to 24,800MB/second, once again breaking through top performance levels for overclocking memory.
... is utter crap.

Germany der8auer says:

Marketing....

Belgium Massman says:

Actually, "marketing" would be telling potential customers you can reach the highest clock frequencies and not say mention anything about the performance. This is just lying ;).

Netherlands rsnubje says:

Well, in the end it's all about timings. that kit makes me wonder, it could just be the fastest kit on the market, at 3100 it's not the best timings, but neither the worse. It could perhaps equal a 2400MHz samsung kit? Just wanted to test something too, here have a look. Hynix CFR 2x4GB:
Hynix MFR 2x8GB: (sorry didnt have anything else laying around.)
Samsung(G.Skill tridentX 2400C11):

United States hokiealumnus says:

ADATA emailed and, like G.Skill before them, iterated appreciation for our candor, so that's two for two manufacturers that actually listen to their feedback. Whether it will affect things in the grand scheme of things is doubtful, but both of their reactions were admirable. On the plus side, the price for this kit just dropped drastically, so I added an Author's note to the end of the article and updated the pricing.

Author's Note - ADATA has adjusted the price of these kits and the article is edited to reflect that change. They went from their previous $364.99 (gold) & $359.99 (gray) to both kits retailing for $299.99. In their category, that makes this kit from ADATA especially tempting. Still not where they should be in terms of performance, but if you're looking for a frequency clocking kit, this is a good price for it.

Belgium Massman says:

Cool!

United States hokiealumnus says:

Double-sided memory puts performance right back where it should be: http://www.overclockers.com/adata-xpg-16gb-ddr3-2600-cl11-memory-review . Give me better performance than high clocks any day. Now, if only every memory company would see the wisdom of my ways and quit it with this frequency game. ;)

United States xxbassplayerxx says:

Great to know... I'm glad it isn't that MFR inherently sucks :p So basically, 8GB sticks using MFR are safe but 4GB sticks are not?

South Africa Vivi says:

start up that steam engine that made the PSC!

United States hokiealumnus says:

xxbassplayerxx said: Great to know... I'm glad it isn't that MFR inherently sucks :p

So basically, 8GB sticks using MFR are safe but 4GB sticks are not?


Basically, yes; 4GB sticks seem to all be single-sided & 8GB are double-sided. They don't overclock as far, but they perform as you'd expect for their speed & primary timings.

Vivi said: start up that steam engine that made the PSC!


+10000000. I would absolutely love if PSC-style memory made a comeback. Unfortunately at this point they're probably spending all their R&D on DDR4 (or that new 3D stuff floating around recently).

Looks like Micron bought Elpida & PSC, so the chances of that kind of comeback are probably extremely slim.

Germany Hyperhorn says:

I just had to bring this topic up again as I spent many many hours on doing comparisons with a Haswell system, grabbing for technical background information etc. In the end it comes to a single question: If you equal all RAM parameters from clock frequency over main timings up to things like tertiary subtimings, have a correct Dual Channel allocation and have reinsured, that the total amount of installed memory does not get exceeded in a single case - why will a setup with two dual-ranked sticks outperform the exact same setup with two single-ranked sticks? Because if you can elliminate differences in things like tRFC it comes down to: Dual-ranked is superior to single-ranked. And technically I've not found a single reason why it should be that way, as the ICs share the same datapath and require an additional chip select line so there's no simultaneous access. Note: I'm talking about ranks as the IC placement itself (single/dual sided) doesn't affect the addressing, but the amount of ranks does.

K404 says:

This applies to all RAM, I hope this info doesn't get "lost" in a thread about PR and reviews :)

Germany Hyperhorn says:

Wow, how could I forget rank interleaving. Thx Pieter! Most stuff I've studied were either too narrow to IC or general DRAM level. Quite amusing how I was missing the forest for the trees ... :)

I suggest everyone who wants to compare the performance to do this in a multi-threading environment or add some kind of consistent memory-consuming background load (e. g. HCI Memtest). You will see bigger differences that way. And of course it's not limited to a certain type of ICs, so it's not necessary to have single-ranked sticks mounted with Hynix MFR.

United States hokiealumnus says:

http://www.overclockers.com/gskill-16gb-tridentx-ddr3-1600-cl7-memory-kit-review Love this kit! Sure, I wouldn't use it for benching really (there are better 8GB kit options on the market) but as a general kit with great overclocking ability, it's great. :)

United States Splave.ROM says:

XMP for 2933 I imagine will have looose seconds and thirds, Id like to see this done again but with the memory tightened down as far as possible at each base frequency.

Please log in or register to comment.