DDR5: how effective for gaming?

[wheresmycar]

Any cited reports or are we all waiting for post-NDA reviews (TPU, 7 days is a lifetime, give us a peek)?

BTW im not in a rush to upgrade (possibly towards the end of 2022) just a little axe-sigh-tid. To be frank, i like knowing all the little nitty gritty stuff and even getting questions answered which i didnt ask. Even rumours and the wealth of information surrounding them is a treasure for me little eyes and 2 brain cells.

 

[Deleted member 185158]

Waiting is the fun part!

 

[ir_cow]

DDR5 has been out for nearly a year now. No NDA stuff to worry about.

 

[sepheronx]

HW unbox did a video and found that with prices gone down on DDR5, there is actual value in it. Not a lot but does add to performance.

 

[dgianstefani]

HW unbox did a video and found that with prices gone down on DDR5, there is actual value in it. Not a lot but does add to performance.

That HW unboxed video has been the target of much derision from many other techtubers.

DDR5 is really good in some games, if you use the top tier kits 6400/6600 c36 etc. The mid and low end are pretty crap compared to good DDR4.

Tuned Samsung B Die dual rank kits are much lower latency than any DDR5 kit, which really helps in high frame rate esports games or for minimum lows for instance, but not many people use 240/360Hz monitors so...

 

[P4-630]

DDR5 is really good in some games, if you use the top tier kits 6400/6600 c36 etc.

Seems I'm Ok-ish with my Corsair Dominator Platinums 6000MHz 36-36-36-76.

 

[dgianstefani]

Seems I'm Ok-ish with my Corsair Dominator Platinums 6000MHz 36-36-36-76.

That's still a high end DDR5 kit.

 

[P4-630]

That's still a high end DDR5 kit.

At least SK Hynix.

 

[steen]

That HW unboxed video has been the target of much derision from many other techtubers.

DDR5 is really good in some games, if you use the top tier kits 6400/6600 c36 etc. The mid and low end are pretty crap compared to good DDR4.

Tuned Samsung B Die dual rank kits are much lower latency than any DDR5 kit, which really helps in high frame rate esports games or for minimum lows for instance, but not many people use 240/360Hz monitors so...

Latency matters & bandwidth difference isn't relevant for most titles. DDR5 7k-8k with decent CL & newer CPU mem controllers will change things. Hard to argue against 12700k/DDR4 vs 12600k/DDR5 in value/$ terms though. By the time current DDR5 makes sense, it's obsolete anyway.

 

[RandallFlagg]

Latency matters & bandwidth difference isn't relevant for most titles. DDR5 7k-8k with decent CL & newer CPU mem controllers will change things. Hard to argue against 12700k/DDR4 vs 12600k/DDR5 in value/$ terms though. By the time current DDR5 makes sense, it's obsolete anyway.

It's not just the DDR5 itself that will be obsolete though. HWUnboxed was arguing that you 'should' use DDR5 in a *budget build* (12100) and get cheap 2x8GB DDR5-4800 now at about the same price as 2x8GB DDR4, and later on you would have an 'upgrade path' as DDR5 gets better and cheaper.

His selection of DDR4 to test against was dubious to start with (3200 C14 and some horrifically bad 4000 C16) - you can go to Newegg and get DDR4-3600 C14 2x8GB right now for about 70-90 bucks, $15 less than 2x8GB DDR5. But even with his lesser DDR4-3200, DDR4 was faster than DDR5-4800 by 5-10% in games.

The big problem with that theory though, is that 'cheap' 6-layer motherboards mostly won't run DDR5 beyond DDR5-6000.

So what happens is, on a budget build you wind up getting 5-10% less performance than DDR4, and in a year or two when say DDR5-6600 becomes cheap, you are not going to be able to just go buy some 6600 and drop it in. Your motherboard won't be good enough.

To make matters worse, from many other tests out there, if you drop in your best DDR5-6000 guess what? It's gonna perform about like that DDR4-3600 C14.

So then you wind up tossing your motherboard and your now worthless DDR5-4800 2x8gb.

I would actually agree with HWUnboxed *if* they were talking about building a high end rig, with a high end motherboard, and forking up for DDR5-6000 C30 right now. The high end 8-layer motherboards are already able to hit 7600 on some of the brand new Hynix A-Die. With a mobo that can handle it, you can beat DDR4 with high speed DDR5.

But they were talking budget, and that's just very misleading.

 

[ir_cow]

I think lots of people here don't understand what being GPU bound means.... Just look at a recent DDR5 review on TPU. Strictly speaking in terms of out of the box XMP performance (no extra tweaking), DDR5 pull ahead in many games on medium settings. Max out the settings and become GPU bound, and it matter very little what you use.

 

[sepheronx]

I got 32gb ddr5 5600 with cl 36.

I'm haply with it.

 

[steen]

@RandallFlagg, agreed. For brevity I left out MB (& Gear1/2) & GPU from the equation.

I think lots of people here don't understand what being GPU bound means....

I doubt that.

Just look at a recent DDR5 review on TPU. Strictly speaking in terms of out of the box XMP performance (no extra tweaking), DDR5 pull ahead in many games on medium settings. Max out the settings and become GPU bound, and it matter very little what you use.

At XMP the perf spread is small regardless of DRAM with just as many DDR4 wins. Perf/$ means diverting extra $ into a higher tier CPU/GPU instead yields > gains.

RPL/Z690/Z790 & RPL/X670/B650 will make for interesting DDR5 reviews, esp with new gen halo GPUs.

p.s. Any chance of including Intel MLC results? The Github GUI is quite stable.

 

[ir_cow]

p.s. Any chance of including Intel MLC results? The Github GUI is quite stable.

Sure. Seems that this program is only for Intel. Tried it on a AMD 5800X just now. 500ns for the memory doesn't seem right..

 

[Deleted member 185158]

Sure. Seems that this program is only for Intel. Tried it on a AMD 5800X just now. 500ns for the memory doesn't seem right..

Correct. Intel only. A bit more accurate than AIDA64. And also, if you ran that, IMLC will show quite a bit more information at different latencies using different amounts of memory for testing IE 128mb 256m and so forth. The actual latency is looser than what AIDA64 claims it to be.

I found 4800mhz default to be super horrible for me. 100ns latency is ouch. And that's IMLC numbers AIDa64 would report low 90s. Either way blah.

Just hope 13th gen does better on the high frequency end and the newer boards of mid and low range better handle 6000mhz and up.

 

[ir_cow]

I found 4800mhz default to be super horrible for me. 100ns latency is ouch. And that's IMLC numbers AIDa64 would report low 90s. Either way blah.

Interesting I heard the same about AIDA64 vs IMLC. I will have to do some comparison testing of my own during some downtime.

 

[Deleted member 185158]

Interesting I heard the same about AIDA64 vs IMLC. I will have to do some comparison testing of my own during some downtime.

Yes good. We need more IMLC in our Intel lives.

Here, I'll hook you up with MLC and the GUI. The GUI is not included in the DL, so this will make your life easier when you get there.

Looking forward to seeing your results!!!