Friday, August 6th 2021

Corsair Postulates That DDR5 Memory Runs Hotter

Corsair DIY Marketing Director, George Makris recently confirmed in a recent video that DDR5 memory could "conceivably could run much hotter than DDR4" due to voltage regulation being moved to the memory modules from the motherboard. This was reiterated by Corsair Memory Product Manager, Matt Woithe, who notes that they are prepared to handle this increased heat in Corsair DDR5 modules using their Dual-path Heat Xchange (DHX) technology. The next generation of memory also mandates the inclusion of on-die EEC which while not confirmed by Corsair will also add to the power budget of the modules. Corsair is expecting to release their first DDR5 memory modules towards the end of this year which will coincide with the launch of Intel's 12th Generation Alder Lake processors. AMD fans will need to wait until 2022 with the launch of Zen 4 to take advantage of the new DDR5 memory modules.
Corsair Lab Tech Talk - DDR5 Memory

Source: Corsair Lab
Add your own comment

43 Comments on Corsair Postulates That DDR5 Memory Runs Hotter

#1
Valantar
How much, exactly, can the conversion losses from a ~5W VRM be? Don't most DC-DC buck converters run at 94+% efficiency? Oh no, my RAM is putting out an extra 0.3W of heat, how are we ever to cool this? Or are they just scaling these circuits incredibly poorly to the load at hand, tanking efficiency?
Posted on Reply
#2
Crackong
Hey if there's still RGB rainbow on your ram sticks, you are just lying on these so-called "concerns".
Posted on Reply
#3
TumbleGeorge
Maybe Corsair confirm their bad engineering and low quality parts that use for make DDR5 modules?
Posted on Reply
#4
DemonicRyzen666
ValantarHow much, exactly, can the conversion losses from a ~5W VRM be? Don't most DC-DC buck converters run at 94+% efficiency? Oh no, my RAM is putting out an extra 0.3W of heat, how are we ever to cool this? Or are they just scaling these circuits incredibly poorly to the load at hand, tanking efficiency?
If I'm not mistaken Intel did this with one line of their cpu's ? didn't they. Where the they had tiny Vrm's on it
Turned out the smaller they are, the more densely packed the heat is. Harder it was too cool them.
Posted on Reply
#5
TheLostSwede
DemonicRyzen666If I'm not mistaken Intel did this with one line of their cpu's ? didn't they. Where the they had tiny Vrm's on it
Turned out the smaller they are, the more densely packed the heat is. Harder it was too cool them.
Yes smaller parts are harder too cool, as there's less surface area to cool. This is why things like power amplifiers in Wi-Fi equipment are hard to cool.

This is a bit different though, as there's no need to use the tiniest components available.
Posted on Reply
#6
persondb
One thing to note is that DDR5 also runs at lower voltages, 1.1V I believe, so that's about 10% less than DDR4.
Also buck converters have very high efficiency and they don't need to put a small one.
The chips are being designed with the on-die ECC, so maybe it won't be a big increase in heat, who knows. As they go into new nodes though, they will likely get some power savings(which is probably not going to be very big, but even small things help).

I think Corsair is just saying that to cover their ass in case they release a kit which thermal throttles, "it's not our fault! It's how DDR5 is"-kind of thing.
Posted on Reply
#7
Valantar
DemonicRyzen666If I'm not mistaken Intel did this with one line of their cpu's ? didn't they. Where the they had tiny Vrm's on it
Turned out the smaller they are, the more densely packed the heat is. Harder it was too cool them.
True, but those were on the underside of a 65-95W CPU, separated from the heat generating silicon by a few mm (at most) and a thin substrate. This will be on the PCB alongside the RAM chips, likely in the middle or along the top of the PCB, so much further away from the silicon, and individual RAM chips barely generate heat at all - not to mention that the ~5W of the DIMM is distributed among the 8-16 chips present. That's quite a different scenario than a single chip outputting >10x the heat in near direct contact with the VRM. Of course CPUs are actively cooled unlike RAM, but again, 0.3W (assuming 5W power draw and 94% conversion efficiency) can be dissipated by a bare package with moderate airflow, even for very small VRM components. It might heat up the closest RAM chips by a tiny bit, but they already stick ridiculously oversized heatsinks on most RAM - maybe it's time for them to actually make those useful? As @TheLostSwede said, there's also no reason to use the smallest components availble - there's generally plenty of space on a DIMM. The parts just need to be relatively low profile - but laptops have low profile VRMs for 45-65W CPUs that turbo to 2-3x that much, so I doubt those are hard to find or cool for a 5W load.
Posted on Reply
#8
Bomby569
No problem on my Lian Li wind tunnel, but could be problematic especially for some integrated systems that are already an oven.
Posted on Reply
#9
dgianstefani
Eh. DDR4 even with it's supposedly tiny heat output of <5w can still run hot, trust me.

B die at 1.5/1.6v even with the overdesigned heatsinks (which really aren't that great) can hit 40-50c under stress, which is where they start to become unstable. With extremely high airflow they can be full load under 30c, but really shouldn't be necessary.

Hopefully manufacturers pull their head out of their ass and start making function over form heatsinks.
Posted on Reply
#10
SN2716057
I still have one of those memory coolers in the basement..hell, it's even an Corsair version.
Posted on Reply
#11
Octopuss
Haha so the RAM heatsinks will serve a purpose for the first time ever? :D
Posted on Reply
#12
Valantar
dgianstefaniEh. DDR4 even with it's supposedly tiny heat output of <5w can still run hot, trust me.

B die at 1.5/1.6v even with the overdesigned heatsinks (which really aren't that great) can hit 40-50c under stress, which is where they start to become unstable. With extremely high airflow they can be full load under 30c, but really shouldn't be necessary.

Hopefully manufacturers pull their head out of their ass and start making function over form heatsinks.
... That is at 1.5-1.6V. DDR4 starts at 1.2V, with 1.35V being the most common for XMP/DOCP. DDR5 starts at 1.1V, and with JEDEC ratings going to at least DDR5-6400 (and likely higher) at that voltage, while we'll no doubt see higher voltages even (or especially) in early high-clocked kits, what you're saying is ... kind of an edge case.

Put more simply: Your argument centers around an extreme edge case in the current generation. Yet you're (seemingly) applying it to non-extreme cases for the next generation. That's some deeply flawed logic. If DDR5 runs hotter than DDR4 when pushed to 33% higher voltage than stock, then ... that's not much of an issue. It might be an issue for overclockers or people running extreme high end RAM kits, but everyone else? Not a chance.

Also, what you're describing is a characteristic of a specific series of DDR4 DRAM dice from a specific vendor. We have literally zero way of knowing if those characteristics will transfer to future DRAM generations in any way, shape or form, beyond general facts like "more power and voltage = more heat".
Posted on Reply
#13
dgianstefani
Valantar... That is at 1.5-1.6V. DDR4 starts at 1.2V, with 1.35V being the most common for XMP/DOCP. DDR5 starts at 1.1V, and with JEDEC ratings going to at least DDR5-6400 (and likely higher) at that voltage, while we'll no doubt see higher voltages even (or especially) in early high-clocked kits, what you're saying is ... kind of an edge case.

Put more simply: Your argument centers around an extreme edge case in the current generation. Yet you're (seemingly) applying it to non-extreme cases for the next generation. That's some deeply flawed logic. If DDR5 runs hotter than DDR4 when pushed to 33% higher voltage than stock, then ... that's not much of an issue. It might be an issue for overclockers or people running extreme high end RAM kits, but everyone else? Not a chance.

Also, what you're describing is a characteristic of a specific series of DDR4 DRAM dice from a specific vendor. We have literally zero way of knowing if those characteristics will transfer to future DRAM generations in any way, shape or form, beyond general facts like "more power and voltage = more heat".
It's not flawed though is it when 1.5v kits are standard for 3800mhz plus high end ram.

There are even 1.6v kits in top end bins.
Posted on Reply
#14
Valantar
dgianstefaniIt's not flawed though is it when 1.5v kits are standard for 3800mhz plus high end ram.
It is, when you start to consider that 3800 is 18% higher than the highest JEDEC standard. For DDR5, that would equate to DDR5-7600 for current standards (DDR5-6400 is the highest standard for now), but JEDEC is working on DDR5 standards up to 8400, similar to how DDR4 went from 2133 to 2666 pretty quickly, but then took several years for JEDEC 3200 to become a thing. So if we start with the more realistic baseline of JEDEC DDR4-2666, you're instead speaking of a 43% uplift, or the equivalent of DDR5-9122 to reach those types of voltages and thermal issues. Or we could say "late standard +18%" like DDR4-3200, in which case we'd be looking at DDR5-8400+18% or DDR5-9900. Either way, normal DDR5 speeds in the 6000+ range should be entirely fine.

Of course, this is all pure speculation, and the specifics of how DDR5 will scale in terms of thermals, voltage and power draw are entirely unknown to people outside of the industry at this point. But this seems like a CYA announcement more than anything else, and I see no real reason to expect problems for the vast majority of users. Unless DRAM makers choose to cheap out on the parts or designs for their VRMs, in which case that's on them entirely.
Posted on Reply
#15
Chrispy_
If DDR5 runs hotter, can we please have some proper cooling fins on our RAM heatsinks instead of decorative plastic and RGBLED?
Also, RGBLEDs add heat - so if companies must include RGBLED in their product lineup, at least keep making the top-tier SKU without it too.
Posted on Reply
#16
dgianstefani
ValantarIt is, when you start to consider that 3800 is 18% higher than the highest JEDEC standard. For DDR5, that would equate to DDR5-7600 for current standards (DDR5-6400 is the highest standard for now), but JEDEC is working on DDR5 standards up to 8400, similar to how DDR4 went from 2133 to 2666 pretty quickly, but then took several years for JEDEC 3200 to become a thing. So if we start with the more realistic baseline of JEDEC DDR4-2666, you're instead speaking of a 43% uplift, or the equivalent of DDR5-9122 to reach those types of voltages and thermal issues. Or we could say "late standard +18%" like DDR4-3200, in which case we'd be looking at DDR5-8400+18% or DDR5-9900. Either way, normal DDR5 speeds in the 6000+ range should be entirely fine.

Of course, this is all pure speculation, and the specifics of how DDR5 will scale in terms of thermals, voltage and power draw are entirely unknown to people outside of the industry at this point. But this seems like a CYA announcement more than anything else, and I see no real reason to expect problems for the vast majority of users. Unless DRAM makers choose to cheap out on the parts or designs for their VRMs, in which case that's on them entirely.
As if JEDEC standard is relevant past providing OEMs with cheap excuses to install slow ram :roll:
Posted on Reply
#17
Valantar
dgianstefaniAs if JEDEC standard is relevant past providing OEMs with cheap excuses to install slow ram :roll:
Given that it's the only existing baseline for RAM specs, what else are you suggesting we base our speculations on? Wishful thinking? FUD?
Posted on Reply
#20
DeathtoGnomes
TheLostSwedeWhy? You don't have enough fan headers in your board?
people use fan headers? :p
Posted on Reply
#21
Chrispy_
dgianstefaniIt's not flawed though is it when 1.5v kits are standard for 3800mhz plus high end ram.

There are even 1.6v kits in top end bins.
I'm calling BS. Source?

Patriot's 4266 kits and below are all 1.35V
viper.patriotmemory.com/products/viper-steel-ddr4-performance-memory-ram-viper-gaming-by-patriot-memory

Browsing through Newegg the majority of the 4000+ kits from Corsair/Crucial are also 1.35V or 1.40V. The reason for this is that the maximum safe voltage the CPU's IMC can drive at is 1.444V (1.35V spec + 7%). Of cource you can overvolt your CPU if you want, but it's a risk that many people won't take, thus capping the market appeal of higher-voltage kits and why 1.5V is oft-cited as the maximum anyone should run.
DeathtoGnomespeople use fan headers? :p
ofc, you can set fan speed based on the temps of the nearest motherboard sensor to the RAM slots without any garbage software running in the background.
Posted on Reply
#22
Vader
Corsair makes a wide range of ram modules. People are discussing high-end kits, but i think this might be more of an issue with budget sticks that don't come with heatsinks. Plus with that "corsair cooling" this reads more like an ad. I mean, ddr5 will likely run hotter than ddr4, we don't know to what extent will this be an issue yet
Posted on Reply
#23
mechtech
The next generation of memory also mandates the inclusion of on-die EEC which…

Im assuming that’s ECC?

and I don’t think the market segregation team at Intel is gonna like that. ;)
Posted on Reply
#24
dgianstefani
Chrispy_I'm calling BS. Source?

Patriot's 4266 kits and below are all 1.35V
viper.patriotmemory.com/products/viper-steel-ddr4-performance-memory-ram-viper-gaming-by-patriot-memory

Browsing through Newegg the majority of the 4000+ kits from Corsair/Crucial are also 1.35V or 1.40V. The reason for this is that the maximum safe voltage the CPU's IMC can drive at is 1.444V (1.35V spec + 7%). Of cource you can overvolt your CPU if you want, but it's a risk that many people won't take, thus capping the market appeal of higher-voltage kits and why 1.5V is oft-cited as the maximum anyone should run.


ofc, you can set fan speed based on the temps of the nearest motherboard sensor to the RAM slots without any garbage software running in the background.
lots of 3800/4000+ kits are Micron E die which have loose timings and get away with lower voltages.

Tight timing high end B die 3800mhz+ is typically 1.5v-1.6v. With higher voltage XMP for tighter sub timings.

There's a few fast kits with decent timings at 1.45v, but I haven't personally tried them so maybe they're not completely stable, or rather optimistic.

Do your own research, B die finder is a good tool to identify B die.

On your own link, the 4000+ have profile 1 at 1.45, and profile 2 at 1.35.
Posted on Reply
#25
Chrispy_
mechtechThe next generation of memory also mandates the inclusion of on-die EEC which…

Im assuming that’s ECC?

and I don’t think the market segregation team at Intel is gonna like that. ;)
\o/
Win-win; we get ECC and we get to flip the bird to Intel's shittiest department.
Posted on Reply
Add your own comment
Copyright © 2004-2021 www.techpowerup.com. All rights reserved.
All trademarks used are properties of their respective owners.