Settling The DDR4 Vs DDR3 Debate: ASRock Fatal1ty Z170 Gaming K4/D3 ATX Motherboard Review

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

SteelCity1981

Distinguished
Sep 16, 2010
1,129
0
19,310
i just don't see a reason not to upgrade to ddr4 if you are building a new skylake system. ddr3 is on its way out thats the reality. amd will soon support ddr4 which will be the final nail in ddr3's coffin. people look at ddr4 latency and are put off by yet have to understand that ddr4 hasnt matured yet in its process and when it does the lateny will get lower then it is now, it's the same thing that happens to each new ram generation when they first come out soit's nothing new.
 

Crashman

Polypheme
Former Staff
The reality is you're talking about XMP, an overclocking technology, and the modules in this test were overclocked to DDR3-2400 and DDR4-2666. 2400 C11 isn't anything to complain about, so you're just shouting into the wind.

Two settings: Standard and overclocked. You're allowed to ignore the non-overclocked settings. Using the popularity of overclocked memory to state that non-overclocked settings must be removed, in order to justify weak argumentation incompatible with reality, is inadmissible...
 

DuncDog

Honorable
Jan 15, 2015
12
0
10,510
Simply put, you're not going to find that any game is RAM bound with a discrete GPU. Core i3 vs i5, doesn't matter. You'd be better off with the cheapest RAM and the i5 than the fastest RAM and the i3.


joex444,

You might want to give this a watch. I agree with you in a few areas, but also disagree in a few others.

It is a video from Eurogamer's Digital Foundry team looking at the comparisons between the 2500K and the 6500 i5s and their relationship by contrasting OC with increased memory speed. Some titles make a huge difference, and others don't. And as mentioned in the video, Sandy Bridge CPUs are limited to DDR3-2133, while the socket identical Ivy Bridge, allows for much faster RAM speeds lessening the gap.

https://youtu.be/frNjT5R5XI4
 

Crashman

Polypheme
Former Staff
First please allow me to apologize for responding directly to comments but not completely. How did "JEDEC Standard" become baseline?
1.) We performance test DDR4 boards at JEDEC standard settings because enabling XMP sometimes enables "enhanced" turbo multipliers while disabling certain power-savings features. In a motherboard review, it's important to have all the boards set up the same way. No cheating.
2.) The DDR4 board was reviewed first.
3.) The DDR4 board was retested with the new memory at the same JEDEC baseline, so you could look at the new numbers, and then look back to the old ones, and see if anything changed.

Since the DDR4 board went first, the only FAIR way to compare the DDR3 board was using the same method: First, test the JEDEC standard. Second, test the highest stable overclock.

We could have tested both boards dozens of other ways too, but the overclocked configuration was the target configuration and we'd already hit that target. Going back to test things below the target would have added excessive time to both the benchmarks and the analysis.

So we could have had five different configurations, but the lab time would have been excessive and two settings seen here (max O/C plus JDEC baseline) would have still been presented.
 

thequn

Distinguished
Nov 4, 2012
393
0
18,810
That's simply not true. You buy ANY of these modules and plug them in, the board detects them as DDR3-1600 C11. Manufacturers have virtually given up on the 1600 C9 standard, and the 1600 C9 memory you buy is programmed to boot at something other than 1600 C9.

I said standard. I didn't say custom timings (XMP or otherwise).
???
my ram is all c7.1600 that stuff is like from 2010.
 

Crashman

Polypheme
Former Staff
You know all the guys who come in here "Why is my DDR3-1600 C7 running at 1333 C9?" Yeh, you have to make changes in the firmware to enable the overclock (it's usually via XMP).

 

genz

Distinguished
I'm sure my 1.65v ram can run at 1.35v but I don't know how that will affect it's stability and performance and what settings/ timings I would need to use...

Ic doubt you can keep the same timings with at .3v under, and running it at 1.65v WILL kill your CPU. There's a reason the values are explicitly stated and that was because when they switched to 1.5v (I believe it was Ivy Bridge), there were MANY fried CPUs all over the forums from 1.65v RAM users. That 1.65v data line is directly connected to your processor. Don't do it.


As for the article. This has been done to death and I really don't get it. Fast RAM just seems completely not worth the price, and slow ram always gives you 99.9% of the performance for far less than 99.9% of the price. I don't get the interest. I don't even get the push for DDR4 other than a reason to hike up prices.
 


Sandy was already supposed to use 1.5V RAM before Ivy came out. Sandy just wasn't as bad at handling higher voltages as Ivy and newer. Running higher voltage RAM is also not inherently so dangerous for the CPU. The problem is that the there is a significant discrepancy between the voltage of the data line (1.65V) and the voltage of the memory controller (generally about 1.0V). The discrepancy should not exceed 0.5V. The CPU's memory controller and related functions need a voltage boost in turn when running higher voltage memory to stop the discrepancy from going over 0.5V. So long as temps are kept under control (higher voltage means higher current which is exponentially worse for the CPU as temps increase linearly), this greatly reduces the number of fried CPUs from running higher voltage memory. Intel (or the motherboard/BIOS companies) makes it difficult because they feel they need to name the settings that need adjusted a different name every generation.

Overall, you're right in that the marketing that goes along with fast RAM is ridiculous. Companies try to give huge frequencies with fairly tight timings and try to sell them to gamers who don't need fast RAM at exorbitant prices and from the customer's perspective, that doesn't make sense. Regardless, DDR4 isn't really more expensive than DDR3 anymore (and will likely drop below DDR3 later this year as DDR3 production is reduced) and higher memory performance does help certain things. For the home user, DDR4 is mostly superfluous, but for many professional workloads (such as rendering), memory performance can be hugely important and the reduced voltage reduces electricity bills for such uses considerably (when a server board has say 12 modules or so per CPU and much more with upcoming CPU generations, it makes a difference). Arguably, DDR3L has pretty similar capabilities to DDR4 nowadays (1.35V DDR3-2133 CAS 11 exists and with a voltage drop to 1.2V or 1.25V, it might still run DDR3-2133 with timings similar to DDR4-2133), but DDR3L might be plateaued in how much performance it can deliver without more voltage whereas DDR4 is almost certain to continue improving.
 
My testing so far has suggested that memory is not the bottleneck. DDR4 significantly outperforms DDR3 in Sandra, but in other tests, there may be no difference, and sometimes the DDR3 system is faster.
 

Crashman

Polypheme
Former Staff

it's been explained to death that CAS 9 at 1800MHz data rate is the same amount of latency as CAS 15 with 3000MHz data rate.
 
"Consider that CAS 11 is the most-common latency standard for DDR3-1600"
Simply not true, the most common latency for DDR3 1600 is CAS 9, I don't think I've ever even seen a 1600MHz stick with a CAS latency of 11 (other than cheap value RAM), CAS 10 is more common than CAS 11 even, but really it should be CAS 9. In fact CAS 9 can and has been used on DDR3 modules up to 2400MHz, further making your use of CAS 11 on 1600MHz bunk...

The only reason I can think of for them wanting to use CAS 11 DDR3 for a benchmark like this is because someone doesn't want them to prove DDR4 isn't an upgrade. I.E. a sponsor or the bigger corps. There is no other logical reason other than that they got bought out... or they just know they'd be in deep shit from their sponsors if they came out with those benchmarks.
 

Crashman

Polypheme
Former Staff
Exactly. Like I said above, if you want a significant memory capacity upgrade to go with your CPU upgrade, it makes sense to buy the two larger-capacity DDR4 modules that match your target capacity. If you're happy with the amount of memory you already have but unhappy with your CPU, it makes sense to get the DDR3 version of the board and keep your DDR3 modules.

I just wouldn't run them at 1.65V with stock memory controller voltage. As shown in this article, most 1.65V DDR3 runs perfectly fine at 1.50V, which is already included in the JEDEC configuration set for the test modules. If you want more, this is also a good starting point for manual tuning, again as shown in the article.

Pure insolence. The DDR4 memory timings started out with JEDEC defaults and scaled up to full O/C. The DDR3 memory timings started out with JEDEC defaults and scaled up to full O/C. Note that both have full O/C: If anything, your banter only proves that your opinion cannot be trusted.

Some people just HATE a fair comparison, and those who troll are quick to label as trolling anything that approaches fairness. You're using pandering and propaganda to refer to something else as pandering and propaganda.
 


I was speaking of common vs common.
CAS 11 1600 is not common and hardly ever used so why include it in a benchmark here to help users decide on what to buy?
 


Your two posts are contradictory, and you are making a lot of assumptions. First, you are assuming that we are getting paid off by companies to alter our findings in a way that benefits them. That is completely wrong. We are journalists, and we take reporting on products very seriously. To alter our findings for monetary gain would be unethical, and as any site publishing false information would get a bad reputation and ultimate become known for publishing lies instead of the truth, doing so would eventually lead to our sites destruction.

The reason Thomas used DDR3 with CAS 11 and 1600 MHz has been posted several times. It is because that is the JEDEC DDR3 standard. The reasons why this is a fair comparison has also been listed in several posts. All of the tested RAM settings are common either as the JEDEC standard, which is heavily used, or by performance RAM.
 

Crashman

Polypheme
Former Staff

For the what, sixteenth time? The DDR4 was tested at JEDEC defaults to set the BASELINE. I know you know what those are. The only fair way to compare that BASELINE was to use JEDEC DDR3 defaults as the other BASELINE.

Now, for only the second time, the reason to include the JEDEC DDR4 Baseline is so that you could compare the DDR4 numbers generated here to those of the DDR4 motherboard's full review.

If you only wanted O/C numbers, you could easily just ignore the baseline numbers rather than following the troll who started this "Simply not true" myth.
 

Crashman

Polypheme
Former Staff

It supports any JEDEC standard for DDR3/DDR3L plus XMP profiles within the hardware overclocking limit. Having said that, I haven't seen proof that the processor can handle the high-voltage XMP profiles over the long term, and DDR3L would be the best way to go.
 

f-14

Distinguished
sandra is a bogus testing platform for performance based on new encryption standards for 90% of the difference which is actually about a 1-2% performance change where the cross cut of the benchmark charts focuses solely on the new encyption difference. furmark and sandra attribute is due to new AES aka encryption and other such the i2700k still shines bright, you give the i7-2700k the new sky lake die shrink, you will have a 4ghz base clock that puts it on par minus the new encryption instruction.
smoke and mirrors from intel, and amd goes right along playing it when they could just skip to ddr5 where true performance gains beyond 1-3% don't get exaggerated in narrowly focused cross cut chart sections.

this articles ddr3 eco ram is a joke, like pitting a kid in a wheel chair against a 8th round JUNIOR olympics sprinter in the 100m
 

Crashman

Polypheme
Former Staff
Would you care to explain your thoughts? Since the article didn't focus on Sandra performance but used it as an example of the extreme? And since the two test settings for each memory standard was standard (JEDEC) and O/C? What you wrote appears to be an argumentative agreement.

 

ceh4702

Distinguished
Jan 1, 2011
305
0
18,790
Often the first batch of motherboards that come out for a chipset are not the highest quality. Often a second version of the motherboard has better specifications and a better BIOS overall. Personally, I hope motherboard manufacturers drop some of the features like SATA Express. It isnt very useful unless their is hardware available that uses it.
 


Why would it be cheating to actually compare the capabilities of the mb?

Is it cheating in F1 racing because not everyone has the same engine?
 

Crashman

Polypheme
Former Staff
it's cheating for two boards to be benchmarked at different settings. This is particularly true with overclocking and for overclockers. These "Enhanced" modes, which disable the lower four-core Turbo ratio and instead force all four cores to the maximum single-core ratio, are still overclocks. And overclockers do that manually.

What I'm saying is that two boards might be equally capable (say, 4.60 GHz at 1.28V), produce the same performance at that setting, but benchmark differently because one is running the CPU by default at 4.40 GHz and the other at 4.20 GHz. Since both boards produce the same performance at max and the same settings at true CPU defaults, an automated O/C produces misleading data.

 
Status
Not open for further replies.