DDR3 vs DDR4 in Gaming

[SteveGrabowski0]

Buying Skylake with DDR3 is ridiculously stupid. DDR4 support is one of the biggest reasons to go with a Skylake cpu. Old ideas seem to die really fucking slowly here, and RAM doesn't matter is one of those old ideas that isn't true. RAM speed matters when you're cpu bound (so a lot of times when you experience minimums in games), and DDR4 is significantly faster than DDR3 unless you're paying through the nose for the best of the best DDR3 kits.

[Sauron]

Digital Foundry would disagree, DDR4 2133MHZ vs 2666 mhz was almost always a 8-10FPS difference and that's without BCLK overclocking. Imagine 1600 vs 2666 

 

 

in situations with a dedicated gpu?

[SteveGrabowski0]

in situations with a dedicated gpu?

 

They tested with a Titan X

[Ansau]

in situations with a dedicated gpu?

 

Yes, the key is to test more cpu based and memory hungry games. Basically open world games, multiplayer or mmo that manage tons of data.

 

- Firestrike is a synthetic test and not a valid to prove the improvements of faster RAM, you need real stuff. Here we have a very small area with only 2 guys fighting each other. Only last test would see some improvements, but at the same time it seems to be more focused on cpu calculations than memory management.

- Alien doesn't stress that much the cpu. A FX-4300 barely exceeds 60% usage when I can see my 4690k @ 4.5GHz over 85% in Gw2 or GTA V.

- Valve Particle simulator is a very old benchmark, where ram speed was very low as ddr3 was not mainstream yet.

 

This is a much better test: http://www.purepc.pl/pamieci_ram/test_ddr3_vs_ddr4_jakie_pamieci_ram_wybrac_do_intel_skylake?page=0,11

[typographie]

Actually I would like to point out that plenty of boards that use DDR3 (non-L) on Skylake and don't have any issues.

 

The concern is that there could be damage from increased wear and tear over time, on the scale of months or possibly years. Obviously they're fine in the short-term. Intel does not recommend using those boards.

[TheCat257]

What about a hard core gamer for FPS and games with high to max settings with graphics mods and mods that add extra things in-game?  What kind of RAM would I need, how many GPUs, how much RAM?  I plan on getting three monitors.

[CodeBr3ker]

If you want to enjoy the benefits of skylake yes. Moreover get the R9 390 as it's usually cheaper and outperforms the 970 at 1440p and 4k. (if you can't find a R9 390 or it's more expensive get a EVGA or MSI 970 the turbo 970 isn't that good).

it´s not that simple. I can only choose between the asus gtx 970 or the msi one. I don´t choose all the parts, the side just gives me options. And no, I won´t build my own from top to finish

[CodeBr3ker]

DDR3 is fine, any reason why you don't want to build it yourself? It's much more fun

I won´t because it will be a little more expensive, since there is pretty hardcore deals on these machines 

because of January and all that, but I will upgrade it myself when I need to, and someday when I have the big pocket, I will probably build my own, but for now, no

[Sack]

I won´t because it will be a little more expensive, since there is pretty hardcore deals on these machines 

because of January and all that, but I will upgrade it myself when I need to, and someday when I have the big pocket, I will probably build my own, but for now, no

As long as it's cheaper, I wish I could build systems for a living tbh, that's how fun I think it is.

[CodeBr3ker]

I am not saying that DDR3 won't work, I am saying it can damage your CPU. Can you watch the video?

 

DDR3L is fine though, that will work fine without damaging your CPU.

The guys building my pc are pros, so I have faith in them. If for some reason the ddr3 damages my cpu, it´s on them, they guaranteed that. And also if it was something extremely serious, like it had a cpu fuck-up-rate at like 100%, I think Intel would have called their Skylake cpus back, or urged a solution for the customer-

[CodeBr3ker]

As long as it's cheaper, I wish I could build systems for a living tbh, that's how fun I think it is.

I believe it´s fun when you have the interest. But of course the custom build computer manufactures are overpriced. They are people that makes a living too, and pays an insanely overdone tax here in DK

[CodeBr3ker]

Then another question, if ddr3 combined with a skylake cpu is a bad idea, what about an intel i5 4690k instead? And what if I combine ddr4 memory and a skylake cpu, how much of an advantage could I draw from that, if we think all round, not just gaming?

[Castdeath97]

it´s not that simple. I can only choose between the asus gtx 970 or the msi one. I don´t choose all the parts, the side just gives me options. And no, I won´t build my own from top to finish

 

Get the msi one then you will be better off (as long as its the twin forzr).

[MageTank]

I am going to end up repeating this until i depart from this world, but so be it. VDIMM does not kill CPU's. That voltage comes from the board, to the ram, and nowhere else. VDIMM has absolutely no impact on the CPU in terms of longevity or damage. The IMC cant even be stabilized by adding more VDIMM. VCCIO (A+D) is the voltage that goes in and out of the IMC. VCCSA is the IMC and PCIE subdomain voltages. These two voltages are what can kill the IMC. 1.5 and 1.65v kits might use higher VCCIO/SA voltages in their XMP profiles, which is why Intel says not to use these kits. DDR3L follows very strict JEDEC standards, and are not going to go beyond those standards, which is why it is 100% supported and considered safe.

 

You can run 1.65v sticks on Skylake as long as you keep VCCIO/SA voltages within safe ranges. For 24/7 use, these numbers were determined to be 1.25v a piece. I personally try to run 1.15 and below, but as long as you stay under 1.25, you are fine. 

 

If anyone doubts my claims, feel free to respond to this comment. I can show you plenty of people running 1.85v-2.1v DDR4 memory overclocked to 4000mhz CL13. Don't even bother saying "but it might cause degradation" because that cant be proven. Skylake has not been out long enough to test that, and if Intel has tested it, they should have released proof of that to consumers. This is just another case of them listing "official support". After all, the very same claims were made back during Sandy Bridge, and we still broke the limitations back then. 

[Dabombinable]

The memory controller in the CPU is exposed to the same voltage as the memory runs at.

 

http://www.tomshardware.com/news/skylake-memory-support,30185.html#xtor=RSS-181

Uh,  no, no its not. Look at the bios on a Skylake motherboard, and you'll see what I mean. BTW that's still not a statement from Intel, I've already read it before, and the how is still missing.

[Sakkura]

Uh,  no, no its not. Look at the bios on a Skylake motherboard, and you'll see what I mean. BTW that's still not a statement from Intel, I've already read it before, and the how is still missing.

 

"After speaking with Intel about the issue, we finally know the truth about what RAM Intel's integrated memory controller (IMC) can support."

[Dabombinable]

"After speaking with Intel about the issue, we finally know the truth about what RAM Intel's integrated memory controller (IMC) can support."

Still not an actual statement from Intel. Still not how DDR3 can damage it. Still hearsay. If something like that was true, Intel would have actually come out with a press statement, and we wouldn't be seeing motherboard manufacturers making DDR3 Skylake motherboards.

[MageTank]

"After speaking with Intel about the issue, we finally know the truth about what RAM Intel's integrated memory controller (IMC) can support."

 

Still not an actual statement from Intel. Still not how DDR3 can damage it. Still hearsay. If something like that was true, Intel would have actually come out with a press statement, and we wouldn't be seeing motherboard manufacturers making DDR3 Skylake motherboards.

If you read closely, it says "using default voltages". Voltage. With an S on the end. Plural. Meaning more than one. I firmly believe they are talking about VCCIO and VCCSA here. Why else would they pluralize it? You could argue they meant 1.5 and 1.65, but 1.65 is not a default voltage. That is way outside of JEDEC standards. Not only that, you have to manually load a profile to get 1.65, so it cant even be used with the word "default". 

 

I believe Intel is warning people against loading their DDR3 XMP profiles on these boards, because the VCCIO/SA voltages might be out of spec with what the Skylake IMC can handle. Understand this. Those high end 1.65v kits are going to need more VCCIO/SA to keep that ram speed stable. No doubt about it. Loading it on Skylake might push you outside of safe limits, and could eventually cause damage. It is easier for Intel to tell people to stick to DDR3L, a 100% safe standard, than to try to explain what VCCIO and VCCSA is, and what are safe limitations, and trying to talk people into making their 1.65v kits stable. After all. Manually dialing in those voltages could mean your XMP profile speed is no longer stable, and that is a headache in and of itself.

 

I know I am not Intel. I know I am not a reputable source. However, many overclockers exist in this world. Google the impact VDIMM has on overall CPU longevity and health. I promise you, the other overclockers will tell you what I am saying to day. Go check out the people pushing 2.1V on HWbot. There are people that run DDR4 at 1.65v 24/7 over there too. DDR4 4266, a speed within JEDEC's standards for DDR4, runs at 1.4V out of the gate. Under load, that thing is going to hit 1.44v. If 1.5V is dangerous, 1.44v would be close to pushing it, no?

[Dabombinable]

If you read closely, it says "using default voltages". Voltage. With an S on the end. Plural. Meaning more than one. I firmly believe they are talking about VCCIO and VCCSA here. Why else would they pluralize it? You could argue they meant 1.5 and 1.65, but 1.65 is not a default voltage. That is way outside of JEDEC standards. Not only that, you have to manually load a profile to get 1.65, so it cant even be used with the word "default". 

 

I believe Intel is warning people against loading their DDR3 XMP profiles on these boards, because the VCCIO/SA voltages might be out of spec with what the Skylake IMC can handle. Understand this. Those high end 1.65v kits are going to need more VCCIO/SA to keep that ram speed stable. No doubt about it. Loading it on Skylake might push you outside of safe limits, and could eventually cause damage. It is easier for Intel to tell people to stick to DDR3L, a 100% safe standard, than to try to explain what VCCIO and VCCSA is, and what are safe limitations, and trying to talk people into making their 1.65v kits stable. After all. Manually dialing in those voltages could mean your XMP profile speed is no longer stable, and that is a headache in and of itself.

 

I know I am not Intel. I know I am not a reputable source. However, many overclockers exist in this world. Google the impact VDIMM has on overall CPU longevity and health. I promise you, the other overclockers will tell you what I am saying to day. Go check out the people pushing 2.1V on HWbot. There are people that run DDR4 at 1.65v 24/7 over there too. DDR4 4266, a speed within JEDEC's standards for DDR4, runs at 1.4V out of the gate. Under load, that thing is going to hit 1.44v. If 1.5V is dangerous, 1.44v would be close to pushing it, no?

I can understand increasing the voltage to the IMC at higher clock speeds. But with DDR3 the clock speeds are lower, and therefore you don't need to adjust it as the DDR3 won't put as much of a strain on the IMC as DDR4. VCCIO is the memory controller voltage, and VCCSA is the System Agent voltage, and VCCSA is the System Agent voltage, and helps with BCLK overclocking. And VDIMM is controlled by the motherboards memory power phases. Eg. 8+2 phase motherboards have 2 phases for the memory, 8 for the CPU . VDIMM has nothing to do with the CPU at all.

[MageTank]

I can understand increasing the voltage to the IMC at higher clock speeds. But with DDR3 the clock speeds are lower, and therefore you don't need to adjust it as the DDR3 won't put as much of a strain on the IMC as DDR4. VCCIO is the memory controller voltage, and VCCSA is the System Agent voltage, and VCCSA is the System Agent voltage, and helps with BCLK overclocking. And VDIMM is controlled by the motherboards memory power phases. Eg. 8+2 phase motherboards have 2 phases for the memory, 8 for the CPU . VDIMM has nothing to do with the CPU at all.

VCCIO is the voltage for the electrical path in and out of the IMC. VCCSA still handles IMC voltage and PCIE subdomain voltage. Also, the amount of voltage differs entirely, and can even depend on how the board trains the tertiary timings. Tighter tertiary timings require more VCCIO/SA voltages, while looser tertiary timings can reduce stress on the IMC, requiring lower voltages to be stable. Remember, Skylake has two different memory controllers on it. We do not know if the DDR3 memory controller is as strong as Haswell just yet, so I cannot speak from experience if it is actually capable of driving strong DDR3 kits at their rated speeds, but I do know that too much VCCIO/SA voltages are a bad thing. 

 

For Haswell, stock VCCIO was 0.850. Stock VCCSA was also 0.850 on Haswell. On Skylake, it is 0.950 VCCIO, and 1.0 VCCSA. This is with my DDR4 kit installed BTW, no idea if the DDR3 boards are different (Would need someone to test that to clarify for me). 

 

EDIT: This image might help people.

 

[Dabombinable]

VCCIO is the voltage for the electrical path in and out of the IMC. VCCSA still handles IMC voltage and PCIE subdomain voltage. Also, the amount of voltage differs entirely, and can even depend on how the board trains the tertiary timings. Tighter tertiary timings require more VCCIO/SA voltages, while looser tertiary timings can reduce stress on the IMC, requiring lower voltages to be stable. Remember, Skylake has two different memory controllers on it. We do not know if the DDR3 memory controller is as strong as Haswell just yet, so I cannot speak from experience if it is actually capable of driving strong DDR3 kits at their rated speeds, but I do know that too much VCCIO/SA voltages are a bad thing. 

 

For Haswell, stock VCCIO was 0.850. Stock VCCSA was also 0.850 on Haswell. On Skylake, it is 0.950 VCCIO, and 1.0 VCCSA. This is with my DDR4 kit installed BTW, no idea if the DDR3 boards are different (Would need someone to test that to clarify for me). 

 

EDIT: This image might help people.

 

OK. But the VDIMM is still separate from VCCIO and VCCSA. Now as for the strength of the memory controller, DDR3 and DDR3L run at the same frequencies so that still isn't doesn't matter as both have the same effect. And again I can do a direct comparison to DDR2+DDR3 motherboards+AM3 Phenom II. The different voltages (VDIMM) of the RAM did not affect the IMC in the slightest, only the frequency or speed. As you increase the frequency or speed of the RAM you need to increase the VCCIO and VCCSA if its not on auto as a greater strain is put on the IMC by the increasing frequency or speed of the RAM.

[MageTank]

OK. But the VDIMM is still separate from VCCIO and VCCSA. Now as for the strength of the memory controller, DDR3 and DDR3L run at the same frequencies so that still isn't doesn't matter as both have the same effect. And again I can do a direct comparison to DDR2+DDR3 motherboards+AM3 Phenom II. The different voltages (VDIMM) of the RAM did not affect the IMC in the slightest, only the frequency or speed.

You don't have to remind me that they are separate, i've been preaching that since the news broke that Vdimm was killing Skylake IMC's. Also, DDR3L and DDR3 is not always the same frequencies. The highest i have ever seen DDR3L is 2133mhz, and that was in SO-DIMM form factor. DDR3 itself has kits available at even 3200mhz at 1.65v. A speed DDR3L will NEVER hit. 

 

My mentioning of the IMC for DDR3 being slightly weaker was not even in regards to frequency. It's voltage tolerances could be lower. It might not be physically capable of handling the same VCCIO/SA voltages that were considered safe on Haswell. If i had a DDR3 Skylake board, i would be able to test this, but I do not. 

 

I am still quite certain VDIMM is not the culprit for Skylake's IMC degrading or dying. I am willing to bet its VCCIO or VCCSA doing it. However, I cannot speak for certain until i get a board to test it. 

[Dabombinable]

You don't have to remind me that they are separate, i've been preaching that since the news broke that Vdimm was killing Skylake IMC's. Also, DDR3L and DDR3 is not always the same frequencies. The highest i have ever seen DDR3L is 2133mhz, and that was in SO-DIMM form factor. DDR3 itself has kits available at even 3200mhz at 1.65v. A speed DDR3L will NEVER hit. 

 

My mentioning of the IMC for DDR3 being slightly weaker was not even in regards to frequency. It's voltage tolerances could be lower. It might not be physically capable of handling the same VCCIO/SA voltages that were considered safe on Haswell. If i had a DDR3 Skylake board, i would be able to test this, but I do not. 

 

I am still quite certain VDIMM is not the culprit for Skylake's IMC degrading or dying. I am willing to bet its VCCIO or VCCSA doing it. However, I cannot speak for certain until i get a board to test it. 

Thing is though, as long as the DDR3 is within JDEC specifications then it will be fine. Anything past the JDEC standards is as usual at your own risk. Also note that the difference in VCCIO and VCCSA voltages in Haswell and Skylake is because of the different types of RAM. DDR4 will always strain an IMC more than DDR3-unless its a higher speed DDR3 kit being compared to a lower speed DDR4 kit.

[FruitBasketSilex]

holly crap, that's overpriced AF[/quote

Actually, its 950$ USD.

Not that overpriced.