Question Does an Alder Lake CPU run hotter under load with DDR5 (especially fast DDR5) than with DDR4 ?

Apr 26, 2022
77
3
35
0
I had decided to return my AMD parts and get Intel 12900KS.

I was on the fence about DDR4 vs DDR5 but have done so much art swapping and selling the last year, I am going all Intel and had some early experience with Alder Lake and a 12700K almost 4 months back with both DDR4 and DDR5.

The memory of fast DDR5 6000 to 6200MHzMHz seemingly causing a 12700K to run 12-15C hotter doing a large data set variable OCCT test with same exact VCORE caused me to stick with DDR4.

Even when I ran DDR5 at SPD spec 4800MHz, it was better, but CPU temps were still at least 5to 7C higher.

Same exact LLC and vcore and clock speed and HT off.

The motherboards were an Asus Tuf D4 Z690 with DDR4

And for DDR5, they were Asus Z690-A Prime DDR5 and Asus Rog Maximus Hero. I had same exact experience of hotter CPU thermals running exact same test on each vs the Tuf DDR4.

Does DDR5 especially fast DDR5 really cause Alder Lake CPUs to run hotter all else being equal or just a coincidence?? Could it have been the Asus Prime Z690-A and Asus Rog Maximus Hero motherboards just cause Alder Lake to run hotter at same settings as opposed to Tuf?

I ran out of motherboards to try and do not really feel like more part swapping once I finalize this 12900KS build. I did not want any brand other than ASUS and Maximum Hero had other problems like the recall and Prime-A supposed to be a good board like the Tuf, but DDR5 with no WiFi and different color. It ran hotter on both.

Is there any truth to it. Does fast DDR5 work the IMC much harder to generate more heat in Alder Lake CPUs even though it is running in Gear 2??

Do the modules run hotter being the VRM for the DIMM now on them and with a large Noctua NH-D15S cooler could the extra heat generated form RAM being so close to CPU cause 12-15C higher temps especially with extra heat generated at fast speeds.
 
Last edited:
It would make sense that slower RAM access would hypothetically be a reduced load on the CPU, as if it were partially 'held back' (avoiding the B-Word!) whilst awaiting retrieval from/writing to RAM...

We began to see even back in 1998-99 that certain game engines (original Unreal) were leading to noticeable CPU temp increases, and particularly when given the best GPUs of the time (Voodoo1, TNT/TNT2, etc.). Slow down any part of the overall 'throughput equation', and the CPUs load is decreased somewhat.
 
I would say it depends on the workload. If your stress test includes frequent RAM access, then I would be inclined to agree that DDR5 will cause higher power consumption over DDR4 by virtue of it being faster.

However, say the amount of data being transferred is more or less the same, then the power over time is likely lower. Sure it might consume more power while doing work, but if the work is done sooner, then that means the CPU can idle sooner.
 
Apr 26, 2022
77
3
35
0
Would that have an affect running OCCT variable Large data set normal test which is a pretty simple and easy on CPU thermals stability that is good at error detection but generally does not put much heat on CPU except Alder Lake DDR5 lol from my experience where temps got into 80s where as mid 60s with DDR4.

And OCCT Variable large data set only used like 3-4GB of 32GB total RAM.

Compared to Small data set tests or Prime 95 Small FFT itself or part of a blend where CPU power and thermals get so much higher, OCCT Large Data Set Variable normal is nothing so it alarmed me when CPU temps on an Alder Lake running such a simple test caused CPU thermals to get into 80s where as other high end Ryzen 5900X and Alder Lake 12700Ks I had with DDR4 were only 50s and 60s with large data set tests in OCCT

Still would cause CPU it to run hotter since DDR5 is faster even though IC runs in Gear 2 and not Gear 1 with DDR5??
 
Last edited:

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
Alder Lake uses "compute fabric" which isn't too much different from AMD "infinity fabric". AMD's infinity fabric is notoriously sensitive to ram timings and I would expect similar results with Intel's rendition. The fabric is in essence the interconnect that lets the individual cores communicate. In stock configurations DDR5 tends to have better timings than DDR4 and has shown to be quite a bit faster in workloads that are memory movement intensive (I have seen up to 46% performance increase on lower timings). With Alder lakes hybrid big-little core setup the big cores are kind of like race horses. They work best when they can run hot and open up their individual clock frequency. The crazy high clocks they are pulling would benefit from DDR5, but how much I really cannot say. That benefit will result in higher clocks, and thus higher thermals, especially in loads that benefit from higher clocks. DDR 5 also moved voltage regulation onto the module, so there is another potential side effect in which there may be more heat being produce in the general CPU vicinity causing some heat increase.
 
Apr 26, 2022
77
3
35
0
Alder Lake uses "compute fabric" which isn't too much different from AMD "infinity fabric". AMD's infinity fabric is notoriously sensitive to ram timings and I would expect similar results with Intel's rendition. The fabric is in essence the interconnect that lets the individual cores communicate. In stock configurations DDR5 tends to have better timings than DDR4 and has shown to be quite a bit faster in workloads that are memory movement intensive (I have seen up to 46% performance increase on lower timings). With Alder lakes hybrid big-little core setup the big cores are kind of like race horses. They work best when they can run hot and open up their individual clock frequency. The crazy high clocks they are pulling would benefit from DDR5, but how much I really cannot say. That benefit will result in higher clocks, and thus higher thermals, especially in loads that benefit from higher clocks. DDR 5 also moved voltage regulation onto the module, so there is another potential side effect in which there may be more heat being produce in the general CPU vicinity causing some heat increase.

Would OCCT Large Data Set Variable Standard test cause such a thing. When I ran the test, there was 32GB of RAM in system and it only used like 2-3GB of it. And OCCT Variable test is not a tough test. It is a basic test that is good at error detection and does not push thermals that hard. Yet it ran 10-15C hotter. And this was with a fixed static overclock speed and vcore and LLC exact same between the Asus Tuf DDR4 and Asus Hero/Asus Z690-A Prime mobos. I did not let the CPU boost to as high as it wanted on speed as it was a static overclock, so DDR5 would not have had the opportunity to force the CPU to run faster at least not in terms of all core clock speed.

I mean based on your theory would DDR5 especially fast DDR5 still cause CPU to run hotter even at static fixed frequency??
 

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
Would OCCT Large Data Set Variable Standard test cause such a thing. When I ran the test, there was 32GB of RAM in system and it only used like 2-3GB of it. And OCCT Variable test is not a tough test. It is a basic test that is good at error detection and does not push thermals that hard. Yet it ran 10-15C hotter. And this was with a fixed static overclock speed and vcore and LLC exact same between the Asus Tuf DDR4 and Asus Hero/Asus Z690-A Prime mobos. I did not let the CPU boost to as high as it wanted on speed as it was a static overclock, so DDR5 would not have had the opportunity to force the CPU to run faster at least not in terms of all core clock speed.

I mean based on your theory would DDR5 especially fast DDR5 still cause CPU to run hotter even at static fixed frequency??
Clock frequency in itself doesn't really reflect the whole thermal picture. If the CPU core is idling for a given percentage of the clock cycles due to being underfed from memory or the thread manager you will see a reduced thermal buildup. Dynamic underclocking/voltage reduction allows for reduced power consumption at idle and/or low loads but isn't really the whole picture. If it were you would see max heat from your processor even under idle loads under a static clock, which most processors were in the early days. Idle processor cycles use very little energy even though the clock cycle still occurs. So if your RAM is creating a bottleneck that is reducing the flops processed by a fair percentage you would definitely expect to see the processor running cooler. Benchmarks are usually the most noticeable because they push things well beyond normal operating conditions to test limits. Amounts of RAM used are really a point moot as most RAM is caching not what is actively being used by the CPU at any given time. This again is especially true when benchmarking.

That heat difference is coming from somewhere. Assuming Ceterius Paribus the only logical explanation is that the processor is running harder and idling less with the DDR5 and performing more flops for the same clock speed.
 
Apr 26, 2022
77
3
35
0
Clock frequency in itself doesn't really reflect the whole thermal picture. If the CPU core is idling for a given percentage of the clock cycles due to being underfed from memory or the thread manager you will see a reduced thermal buildup. Dynamic underclocking/voltage reduction allows for reduced power consumption at idle and/or low loads but isn't really the whole picture. If it were you would see max heat from your processor even under idle loads under a static clock, which most processors were in the early days. Idle processor cycles use very little energy even though the clock cycle still occurs. So if your RAM is creating a bottleneck that is reducing the flops processed by a fair percentage you would definitely expect to see the processor running cooler. Benchmarks are usually the most noticeable because they push things well beyond normal operating conditions to test limits. Amounts of RAM used are really a point moot as most RAM is caching not what is actively being used by the CPU at any given time. This again is especially true when benchmarking.

That heat difference is coming from somewhere. Assuming Ceterius Paribus the only logical explanation is that the processor is running harder and idling less with the DDR5 and performing more flops for the same clock speed.

So are you saying DDR4 RAM was actually bottlenecking the CPU?? Even though DDR4 RAM runs as well or even better than DDR5 in many cases on Alder Lake today?? Or not the kind of bottlenecking that I think??

Or even if it was bottlenecking CPU it is not a bad thing as DDR4 would bottleneck Zen 2 and Zen 3 and Rocket Lake and Comet Lake/Coffee Lake as well. Its just that there is no DDR5 nor will there ever be available for those platforms. Its just that slower RAM will bottleneck CPU more than faster and faster RAM?? As RAM is always a bottleneck to CPU unless it could run as fast which it never has. But DDR5 is faster so bottlenecks CPU less even though DDR4 is more than fast enough with ow latency and will be for a while.
 
Last edited:

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
So are you saying DDR4 RAM was actually bottlenecking the CPU?? Even though DDR4 RAM runs as well or even better than DDR5 in many cases on Alder Lake today?? Or not the kind of bottlenecking that I think??

Or even if it was bottlenecking CPU it is not a bad thing as DDR4 would bottleneck Zen 2 and Zen 3 and Rocket Lake and Comet Lake/Coffee Lake as well. Its just that there is no DDR5 nor will there ever be available for those platforms. Its just that slower RAM will bottleneck CPU more than faster and faster RAM?? As RAM is always a bottleneck to CPU unless it could run as fast which it never has. But DDR5 is faster so bottlenecks CPU less even though DDR4 is more than fast enough with ow latency and will be for a while.
Technically speaking there is always something causing a bottleneck. Whatever component it is there is never really a truly balanced system. It might be RAM, CPU, GPU, PCI interconnect but something will always limit. There is a reason they made DDR5. It is significantly faster with typically better timings than DDR4. Alder Lake is a pretty major design shift and is alot more like Zen and its infinity fabric (AMD). Intel calls it "Compute Fabric" but it is still a fundamental shift from the monolithic die structure that Intel had been using up until now in its core line. Each core connects via the fabric and to the memory via the fabric as well. Alder Lake has 1000GB/s Compute fabric and 204GB/s memory fabric. DDR4 only has 1.6 to 3.2 Gbps max data rate while DDR5 starts at 4.8 to 8.4Gbps. So because of the way the fabric works Alder lake can plug in legacy RAM, but yes you will take a performance hit. I cannot confirm any benchmark data on Alder lake, but I do know that there is a fair performance penalty on infinity fabric (on Zen) for poor RAM timings, and I would expect the same from the compute fabric. At stock timings DDR5 is benching somewhere around 14%-19% faster than DDR4 and has an obvious bandwidth advantage. So if there was any bottleneck existing there, it would be less with DDR5.
 
Apr 26, 2022
77
3
35
0
Technically speaking there is always something causing a bottleneck. Whatever component it is there is never really a truly balanced system. It might be RAM, CPU, GPU, PCI interconnect but something will always limit. There is a reason they made DDR5. It is significantly faster with typically better timings than DDR4. Alder Lake is a pretty major design shift and is alot more like Zen and its infinity fabric (AMD). Intel calls it "Compute Fabric" but it is still a fundamental shift from the monolithic die structure that Intel had been using up until now in its core line. Each core connects via the fabric and to the memory via the fabric as well. Alder Lake has 1000GB/s Compute fabric and 204GB/s memory fabric. DDR4 only has 1.6 to 3.2 Gbps max data rate while DDR5 starts at 4.8 to 8.4Gbps. So because of the way the fabric works Alder lake can plug in legacy RAM, but yes you will take a performance hit. I cannot confirm any benchmark data on Alder lake, but I do know that there is a fair performance penalty on infinity fabric (on Zen) for poor RAM timings, and I would expect the same from the compute fabric. At stock timings DDR5 is benching somewhere around 14%-19% faster than DDR4.

Isn't DDR5 faster speed, but slower latency?? Like CAS latency is like 32-40 where as on DDR4 it is 14-18?? And the fact DDR4 can run in Gear1 and if Alder Lake has an Infinity Fabric like thing where running it in 1:1 ratio. DDR5 has to run in 1:1 ratio. Does that hurt performance of Alder Lake? And is it important that Intel is able to get DDR5 to run in 1:1 to realize its better performance due to faster MHz speed?
 

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
Isn't DDR5 faster speed, but slower latency?? Like CAS latency is like 32-40 where as on DDR4 it is 14-18?? And the fact DDR4 can run in Gear1 and if Alder Lake has an Infinity Fabric like thing where running it in 1:1 ratio. DDR5 has to run in 1:1 ratio. Does that hurt performance of Alder Lake? And is it important that Intel is able to get DDR5 to run in 1:1 to realize its better performance due to faster MHz speed?
Depends on the module: https://www.tomshardware.com/news/gskill-unveils-extreme-low-latency-ddr5-kits

There are faster and slower timings depending on the RAM you purchase. Also CAS latency isn't really the whole picture when it comes to real performance. latency = clock cycle time x number of clock cycles. So while you may be looking at stock CAS timing of 22 on DDR4-3200 it times out to roughly a 13.75 ns latency, while DDR5-4800 at 40 CAS times to roughly 16.67 ns. In this case the DDR4 has a better latency figure, but bandwidth has to be taken into account. Depending on the module DDR5 could be significantly faster as newer modules are going as low as 28 with a 5600MT/s rate which will be as responsive as any DDR4 with DDR5 bandwidth advantages. First generation memory on new standards always have this argument thrown at it, but real latency hasn't changed that much generationally for quite a while. However bandwidth definitely has increased. Some of the best latency figures actually existed in the DDR3 days.
 
Last edited:
Apr 26, 2022
77
3
35
0
Depends on the module: https://www.tomshardware.com/news/gskill-unveils-extreme-low-latency-ddr5-kits

There are faster and slower timings depending on the RAM you purchase. Also CAS latency isn't really the whole picture when it comes to real performance. latency = clock cycle time x number of clock cycles. So while you may be looking at stock CAS timing of 22 on DDR4-3200 it times out to roughly a 13.75 ns latency, while DDR5-4800 at 40 times to roughly 16.67 ns. In this case the DDR4 has a better latency figure, but bandwidth has to be taken into account. But depending on the module DDR5 could be significantly faster as newer modules are going as low as 28 with a 5600MT/s rate which will be as responsive as any DDR4 with DDR bandwidth advantages. First generation memory on new standards always have this argument thrown at it, but real latency hasn't changed that much generationally for quite a while, but bandwidth definitely has increased.

It seems DDR5 has lots more potential, but motherboard support is buggy and does not support SPD specs past 4800MHz and rest is XMP and often buggy and not stable even with good motherboards?? It seems it takes time for new memroy XMP timings and official JDEC specs to get better and support and stability at higher speeds to get better.

What future chipset do you think it happens with where DDR5 is superior to DDR4?? Z790 or X670?? Or something afterwards? Or newer revisions of Z690 motherboards? And how long do you think it will take.

And will DDR5 ever be able to run 1:1 mode with Zen 4 and Alder Lake/Raptor Lake. It seems 1:1 mode is important for overall performance.
 

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
The Gear 1 (1:1) and Gear 2 (2:1) setup is fairly well documented with DDR4 but I am not entirely sure how Intel handles it with DDR5 and Alder Lake. The overall issue is that RAM transistor/capacitor latency time really hasn't changed much in quite a while. So while marketed frequency has increased, parallelism has been the magic that has allowed this not some magic trick that changes the speed of electricity. This is why you see the CAS latency timing increase every time higher bandwidth RAM is released. It is also why really good RAM has lower timings, because the transistor/capacitor interaction time is really short in those modules. High end RAM is expensive because they have to bin the chips like crazy. DDR5 is a new controller with design improvements on that parallelism allowing for alot more bandwidth. So CAS timing or not comparing DDR4 to DDR5 is not an apples to apples comparison.
 
Apr 26, 2022
77
3
35
0
Technically speaking there is always something causing a bottleneck. Whatever component it is there is never really a truly balanced system. It might be RAM, CPU, GPU, PCI interconnect but something will always limit. There is a reason they made DDR5. It is significantly faster with typically better timings than DDR4. Alder Lake is a pretty major design shift and is alot more like Zen and its infinity fabric (AMD). Intel calls it "Compute Fabric" but it is still a fundamental shift from the monolithic die structure that Intel had been using up until now in its core line. Each core connects via the fabric and to the memory via the fabric as well. Alder Lake has 1000GB/s Compute fabric and 204GB/s memory fabric. DDR4 only has 1.6 to 3.2 Gbps max data rate while DDR5 starts at 4.8 to 8.4Gbps. So because of the way the fabric works Alder lake can plug in legacy RAM, but yes you will take a performance hit. I cannot confirm any benchmark data on Alder lake, but I do know that there is a fair performance penalty on infinity fabric (on Zen) for poor RAM timings, and I would expect the same from the compute fabric. At stock timings DDR5 is benching somewhere around 14%-19% faster than DDR4 and has an obvious bandwidth advantage. So if there was any bottleneck existing there, it would be less with DDR5.

Does dual rank Samsung B Die 3600MHz or faster CL14 compensate for the advantages DDR5 has at least today's variants??

Like Build Zoid did a thing on how dual rank is great and significantly improves DDR4 performance. But going to quad rank would be bad. And he mentioned that DDR5 is very good at single rank already and going dual rank with it is bad.

Also mentions that with DDR5, any capacity below 16GB DIMMs is bad and go for DDR4 in such a case if you do 8GB DIMMs

View: https://www.youtube.com/watch?v=CtGXAZznKSc


And I wonder if the tests that show DDR5 6000MHZ ahead of CL14 3600MHz DDR4. Are they actually using Dual rank DDR4. It seems all DDR5 is single rank anyways and for a good reason. But DDR4 should be dual rank for performance, but cheaper to make single rank sticks.

I am sure in future DDR5 is going to slaughter DDR4 when faster and more refined versions come out, but it seems now it is very close or better to DDR4 like fast DDR3 was to DDR4 at initial release. Plus it seems DDR5 is buggy now as it is so new. and only 1 generation supports it and we have 6 months or more to wait before Zen 4 and Raptor Lake mobos come to the table to support it. Though DDR4 at initial release on mainstream platforms had less issues as DDR4 was more mature as Z170 was not first platform to support it but first mainstream non HEDT platform to do so. But will be great with time on future platforms as the spec has great promise and potential like never seen. But now is probably not the time for it even if cost is not an issue.
 
Last edited:

jasonf2

Honorable
Oct 11, 2015
633
158
11,390
66
Does dual rank Samsung B Die 3600MHz or faster CL14 compensate for the advantages DDR5 has at least today's variants??

Like Build Zoid did a thing on how dual rank is great and significantly improves DDR4 performance. But going to quad rank would be bad. And he mentioned that DDR5 is very good at single rank already and going dual rank with it is bad.

Also mentions that with DDR5, any capacity below 16GB DIMMs is bad and go for DDR4 in such a case if you do 8GB DIMMs

View: https://www.youtube.com/watch?v=CtGXAZznKSc


And I wonder if the tests that show DDR5 6000MHZ ahead of CL14 3600MHz DDR4. Are they actually using Dual rank DDR4. It seems all DDR5 is single rank anyways and for a good reason. But DDR4 should be dual rank for performance, but cheaper to make single rank sticks.

I am sure in future DDR5 is going to slaughter DDR4 when faster and more refined versions come out, but it seems now it is very close or better to DDR4 like fast DDR3 was to DDR4 at initial release. Plus it seems DDR5 is buggy now as it is so new. and only 1 generation supports it and we have 6 months or more to wait before Zen 4 and Raptor Lake mobos come to the table to support it. Though DDR4 at initial release on mainstream platforms had less issues as DDR4 was more mature as Z170 was not first platform to support it but first mainstream non HEDT platform to do so. But will be great with time on future platforms as the spec has great promise and potential like never seen. But now is probably not the time for it even if cost is not an issue.
This really just seems to be the norm when it comes to new bus standards. Early integrators really are just tacking things together at first out of existing parts to get to market quick. It takes a while for the pieces to catch up and surpass the old standards abilities.
 

ASK THE COMMUNITY