So how does PBO scale with CPU cooler effectiveness? (also AMD stock cooler review)

[Xlucine]

Everyone knows PBO likes a cooler CPU and will boost higher, but how much does it actually vary across low-end coolers? Is a £30 cooler a good investment for zen2, or does it not change much? I've got a couple of coolers lying around, so decided to find out.

Test setup:
Case: Silverstone TJ-08e
CPU: 3700X with PBO on, but no frequency offset (left on 'auto' in the bios)
Motherboard: Asus TUF B450M PLUS GAMING. The included fan control software was used to adjust fan speed, with the intake fan held constant around 700-750 RPM. This limited the testing somewhat, as the software is set up to put all fans to 100% when the CPU exceeds 95 C or so so I was not able to explore the full RPM range for all coolers.
Ambient: 22C +/- 1C throughout testing, according to a alcohol thermometer on my wall.
Load: CPU-Z stress test (frequency will vary with load, prime95 tends to be ~100 MHz higher but varies throughout the stress test)
Measurement software: HWiNFO64
Thermal paste used: Arctic MX-4
Coolers tested: Wraith stealth, Wraith prism, and a silverstone AR01 - a ~£30 120mm tower cooler with direct touch heatpipes comparable to a hyper 212 or similar. This particular AR01 has a noctua NF-P12 redux fan on it, as the original fan was too loud for my liking.

Results:

Temperature is the peak recorded over 10 minutes of the stress test, and frequency is averaged across the test period.

All in all, surprisingly little variation. Between the worst and best cases there's only ~150 MHz (~4%), giving very little real-world impact. This also shows the wraith stealth to be a pretty decent cooler, even on an overclocked 3700X. Package power did vary as well, and ranged from 86 W with the wraith stealth to 90 - 97 W with the other coolers. The single data point for the wraith stealth is due to the high temperature; my motherboard would not let me reduce the fan speed.

The two AMD coolers are surprisingly alike. They both have 85 mm rotors, they both have a minimum PWM speed of 550 RPM, and they both have 7 blades (probably the same geometry). The only difference is the top speed (1840 for the stealth and 2713 for the prism), the fins they blow on, and the RGB. At full speed the prism is quite loud, but this isn't needed for normal operation and at more sedate speeds they're both quiet. Compared to the 120 mm fan they both seem to spin very fast, but are still quiet due to their smallness.

Based on this, I recommend sticking with the stock cooler if you're got a wraith prism - they're decent coolers, and an aftermarket one really doesn't add much.


Building on this all, I had another question I wanted to answer - is zen2 limited by the heatspreader efficiency?

Cooler geometry is optimised for CPUs with a hotspot in the middle of the package, so with zen2 the heat has to percolate sideways across the heatspreader to the rest of the cooler. With the AR01 one of the direct touch heatpipes does touch over the chiplet, but if I can get more heat to get to the other heatpipes then all the better (in theory). Higher end coolers tend to have a copper block between the heatspeader and heatpipes, so I tested the effect of one of these.

I only have the AR01 to hand however, and a different non-direct-touch cooler would add additional unknowns to the testing, so I made some copper shims (in 2mm and 5mm thicknesses) to test this effect with the AR01. This would help heat get to the other heatpipes, but doesn't come for free - I'm adding a second layer of thermal paste and the non-zero thermal resistance of the copper to the equation. For the typical heat flux from the 3700X (100 W in 40 mm square, or 60 kW/m^2), and assuming 400 W/m.k for commercially pure copper, then this works out to be 0.16 C / mm - so if improving the efficiency of the heatspreader doesn't help the cooler, then the 2mm shim should be at least 0.3 C worse and the 5mm shim should be at least 0.8 C worse.

The shims were cut from some copper sheet of the required thickness, and some spacers were also made to ensure the mounting pressure was the same (luckily, the mounting system had room to fit the extra thickness)

Results:


A big ol' heap of nothing - they all performed the same. This shows more heat was getting to the other heatpipes, helping the cooler shed heat, however this was pretty much exactly countered by the increased thermal resistance - all results are within margin of error.


Finally, to answer the question I posed in the title - lumping all the results together in one graphs, PBO scales like this:

[DanceswithUnix]

Nice work, though I presume the last graph is really Precision Boost and applicable for those of us that don't overclock. Overdrive should just push you up to the hotter end?

Agree that the standard fan is pretty good, I still haven't fitted my tower cooler. But seeing the lower fan speed for the same frequency is nudging me to maybe try it later.

[Zak33]

What's the maximum thickness of copper shim you think you could fit under the heatsink, using the standard fitting kit?

I ask because it's viable that a thicker copper shim would do a better job carrying the heat to the corners of the heatsink properly..... i think.
It's conducts better and has a higher volumetric heat capacity than aluminium and that combined MIGHT mean you'd get a better result with a thicker one.

not that you got a bad result tbh.. the Prism is a good cooler.

There's a new one too.. imminent...same model code.. will be hard to see which one people get.
https://hexus.net/tech/news/cooling/...ed-cpu-cooler/

[DanceswithUnix]

The heat spreader on top of the chip should really be doing the job of, well, spreading the heat across the heatsink base De-lidding shows that some people think the one existing shim is already too much.

Adding a shim will increase thermal soak, so spikes in heat generation will be fine, but it is still extra thermal resistance between the CPU and the air so overall not a good thing. Ideally you want the heatpipes in direct contact with the CPU heat spreader to get that heat away.

There's a new one too.. imminent...same model code.. will be hard to see which one people get.
https://hexus.net/tech/news/cooling/...ed-cpu-cooler/

If you read the forum comments on that article, AMD said that was an unauthorized fake, based on their design and using their company logo without permission. Interesting that they added heatpipes to the original design though.

[Xlucine]

Nice work, though I presume the last graph is really Precision Boost and applicable for those of us that don't overclock. Overdrive should just push you up to the hotter end?

Agree that the standard fan is pretty good, I still haven't fitted my tower cooler. But seeing the lower fan speed for the same frequency is nudging me to maybe try it later.

The testing was with PBO on, so not strictly PB, but really AMD boost acronyms are a mess so I'm not sure which one is which anymore

What's the maximum thickness of copper shim you think you could fit under the heatsink, using the standard fitting kit?

I ask because it's viable that a thicker copper shim would do a better job carrying the heat to the corners of the heatsink properly..... i think.
It's conducts better and has a higher volumetric heat capacity than aluminium and that combined MIGHT mean you'd get a better result with a thicker one.

not that you got a bad result tbh.. the Prism is a good cooler.

There's a new one too.. imminent...same model code.. will be hard to see which one people get.
https://hexus.net/tech/news/cooling/...ed-cpu-cooler/

5mm was the thickest I could fit, which is a pretty chunky copper plate. I've got some 10mm copper, but realised after buying it that it was too much for the mounting hardware. The 2mm and 5mm copper shims (really stretching the definition of shims at those thicknesses) performed identically, but I suspect that going thicker would start to hurt performance (thicker gives higher efficiency but also higher resistance, and efficiency can only go to 100% - 100 W passing through a 40x40x5 mm bit of copper can't be far off that). High heat loads ought to like thicker shims, but air cooled AM4 won't be much above 100 W (an octave at most)

The heat spreader on top of the chip should really be doing the job of, well, spreading the heat across the heatsink base De-lidding shows that some people think the one existing shim is already too much.

Adding a shim will increase thermal soak, so spikes in heat generation will be fine, but it is still extra thermal resistance between the CPU and the air so overall not a good thing. Ideally you want the heatpipes in direct contact with the CPU heat spreader to get that heat away.

It's more thermal resistance between the die and the heat pipe that happened to line up with the hot bit, but less thermal resistance between the hot bit and the other two heatpipes. Getting the heatpipes closer together and all contacting the bit over the CCD would be ideal, but surprisingly silverstone didn't design for zen2 back in 2014 when I bought the thing

[Domestic_Ginger]

Great review. I thought I could offer my findings from this evening.

I am a new owner of a 1600 and have had a go with PBO.

Stressed using prime95 max temp 60C (PWM not working on CPU fan so its on full whack; need to investigate) on a HDT-s1283.

PBO V VDD Power Settings Result all core Mhz
OFF Auto Auto off ~3.6
ON Auto 60A off bounces around 3.6
ON Auto 150A off ~3.6
ON 1.3 150A 120W, 120W, 120A ~3.0
ON 1.35 200A 120W, 120W, 120A ~2.0
ON 1.2 ALL AUTO ~3.7
ON 1.35 ALL AUTO ~3.7

They weren't kidding when they said you needed an x chip to run it!!!!!

[watercooled]

Xlucine since you have a 3700X and a B450, if you have access to a power meter I wonder if you could get some proper idle readings? They seem surprisingly hard to find on chipsets other than X570 and I'd really like to know what real-world idle is like vs the previous generation, and many reviews I've seen that do measure it, have the power plan set to high performance which makes the readings a bit meaningless!

Unlike lots of reviews you also have a reasonably-sized PSU for the build which would also make readings more meaningful.

[Xlucine]

Xlucine since you have a 3700X and a B450, if you have access to a power meter I wonder if you could get some proper idle readings? They seem surprisingly hard to find on chipsets other than X570 and I'd really like to know what real-world idle is like vs the previous generation, and many reviews I've seen that do measure it, have the power plan set to high performance which makes the readings a bit meaningless!

Unlike lots of reviews you also have a reasonably-sized PSU for the build which would also make readings more meaningful.

Sure:
Desktop, only task manager open: 52 W
Sleep: 1.2 W
Fully off: 0.7 W
Full load: 420 W (ayyyy)

I also had a look at my HTPC (with a 4690k, 16 GB of DDR3 I had lying around, 1 HDD and 1 SSD, no dGPU, in an old lenovo Q87 SFF prebuilt w/ 240 W PSU);
Desktop: 26 W
Sleep: 1.4 W
Fully off: 0.7 W
Full load: 111 W

Idle is higher than I expected, but fits with reported power draws in HWiNFO64. The claimed package power for the 4690k was 3.5 W at idle (downclocking to 800 MHz), whereas the 3700X claims 16 W or so (and add 18 W for the GPU. Lowest CPU got was ~2.7 GHz). ~30 W more claimed draw by the components seems reasonable with the extra 26 W measured (the extra draw could have been counteracted by less chipset draw, potentially less losses in the VRMs, and no HDD in my ryzen system). I'm impressed with how little power the intel iGPU needs to display the desktop, my more modern GPU needs substantially more and looking at idle figures in hexus reviews shows this doesn't change much across dGPUs

I had task manager open in both cases to make sure there were no windows stuff in the background messing up the results, and power plan is left default ("better performance", i.e. middle of the slider between "best energy savings" and "best performance"). PBO is still on, but shouldn't affect things much (as the chip can still downclock when not loaded)

[watercooled]

Thanks, idle does seem a tad on the high side, but like you say it's hard to know where that power is going unless you're comparing like-for-like.

[cheesemp]

Thanks for doing this. I asked this question on HotUKDeals - is it better to go high end mb, 3600 and high end cooler, or mid range mb, 3700 and and free cooler (wraith prism?). From your results the second option might be the better performing option... Kind of surprised based on my very out of date experience with a 3570k where a £30 cooler gave you an almost 1ghz OC. Have I got it right? If so is there a minimum board price wise to aim for as a 3700 would be a stretch for me.

[CAT-THE-FIFTH]

Thanks for doing this. I asked this question on HotUKDeals - is it better to go high end mb, 3600 and high end cooler, or mid range mb, 3700 and and free cooler (wraith prism?). From your results the second option might be the better performing option... Kind of surprised based on my very out of date experience with a 3570k where a £30 cooler gave you an almost 1ghz OC. Have I got it right? If so is there a minimum board price wise to aim for as a 3700 would be a stretch for me.

Latter - the Ryzen 7 3700X consumes as much as my Ryzen 5 2600(IIRC). I tested both the Wraith Stealth and Wraith Spire on my Ryzen 5 2600 in a cooling limited mini-ITX case:
https://forums.hexus.net/pc-hardware...ml#post4023092

Even though temperatures were better and the boost higher with the better Wraith Spire,the differences were minor. The only thing which was noticeable was the lower noise with the Wraith Spire.

[Domestic_Ginger]

Thanks for doing this. I asked this question on HotUKDeals - is it better to go high end mb, 3600 and high end cooler, or mid range mb, 3700 and and free cooler (wraith prism?). From your results the second option might be the better performing option... Kind of surprised based on my very out of date experience with a 3570k where a £30 cooler gave you an almost 1ghz OC. Have I got it right? If so is there a minimum board price wise to aim for as a 3700 would be a stretch for me.

I think its all relative. If you are going to OC then get the high end mobo but tbh overclocking seems pretty fruitless as you'll get maybe 10% out of it. If you need the extra cores then get the 3700. A good B450 with either is probably fine. I recently got an MSI mortar and a 1600 AF; the idea being I can put in a Zen3 part in 3 to 4 years. I would be a little picky with the ram as you should try to get 3600mhz stuff if possible - it comes with a premium though and you should be able to find cheaper 3000 or 3200 kits and OC them.

Thaiphoon burner and dram calculator can be used to tell what a set is like before you buy it; there is a data base in TB and dram calculator will tell you what to expect if you OC it. CtV got some nice B-die from what I saw in his excellent review but it has a price premium. Micron Rev-E is good and should clock quite high without loads of voltage. Hynix CJR (?) is the third choice from what I understand.

[CAT-THE-FIFTH]

I think its all relative. If you are going to OC then get the high end mobo but tbh overclocking seems pretty fruitless as you'll get maybe 10% out of it. If you need the extra cores then get the 3700. A good B450 with either is probably fine. I recently got an MSI mortar and a 1600 AF; the idea being I can put in a Zen3 part in 3 to 4 years. I would be a little picky with the ram as you should try to get 3600mhz stuff if possible - it comes with a premium though and you should be able to find cheaper 3000 or 3200 kits and OC them.

Thaiphoon burner and dram calculator can be used to tell what a set is like before you buy it; there is a data base in TB and dram calculator will tell you what to expect if you OC it. CtV got some nice B-die from what I saw in his excellent review but it has a price premium. Micron Rev-E is good and should clock quite high without loads of voltage. Hynix CJR (?) is the third choice from what I understand.

When I bought my RAM it was the cheapest 3200MHZ kit I could get,so I got lucky as it was either Samsung or Hynix RAM:

https://www.overclockers.co.uk/team-...my-0a2-tg.html

But I would be aiming for now is the Micron E-die RAM,as it's cheaper and should overclock reasonably well.

[cheesemp]

Thanks for the responses. I tend to buy to keep longer term so figure the leap to 8 core was a better longer term bet but it is s stretch financially (but good to know it makes sense). I guess I'll see what I can afford when the time comes. I'm not looking to purchase for a month anyway.

[DanceswithUnix]

Thanks for doing this. I asked this question on HotUKDeals - is it better to go high end mb, 3600 and high end cooler, or mid range mb, 3700 and and free cooler (wraith prism?). From your results the second option might be the better performing option... Kind of surprised based on my very out of date experience with a 3570k where a £30 cooler gave you an almost 1ghz OC. Have I got it right? If so is there a minimum board price wise to aim for as a 3700 would be a stretch for me.

The free cooler on the 3700X is pretty good, and it is only having to cool a 65W TDP chip. If you are *very* noise sensitive then you would probably end up buying an aftermarket tower cooler, but honestly the tower cooler I mentioned at the start of this thread is still sat in the box unopened for months because I just haven't felt the need to install it.

When I wanted a cheap board to play VR games downstairs I went for an ASUS TUF B450M-PLUS (same as used in these tests), basically because is was on sale so only paid £75 for it: https://www.amazon.co.uk/gp/product/B07F6XBJVG/

You can get cheaper boards, but I am a sucker for 4 ram slots for that machine end of life boost and it has plenty of USB3 which the old Rift headset needs. There aren't a gazillion phases on the VRM, there is only one "heatsink" on the main CPU phases with nothing on the other phases and that heatsink is more of a heatsoak than a sink thanks to its stylised lack of proper surface area and airflow (that's still better than no heattsink at all). But really, none of that particularly matters and even in a limited airflow shoebox case the machine is fine.

I draw the line at most A320 boards, they seem to be engineered a bit too cheap for my liking so I wouldn't feel comfortable putting more than a 35W Athlon in one. The B450 boards have to be able to handle some overclocking without going pop, so if like me you think Ryzen overclocking isn't worth bothering with then they all have a bit in reserve to keep them unstressed. I've yet to see a really bad B450 board.

[DanceswithUnix]

Got bored of saying I haven't fitted my aftermarket Arctic Freezer 34 eSports cooler, so I fitted it.

Machine is slightly quieter, which makes me think a lot of the noise is probably from the case fans (120mm intake and 140mm exhaust). My software build times have dropped from 80.9 seconds to 79.8s, but most of that is probably because the old fan was packing a lot of dust and I suspect if I had cleaned the old fan out and replaced it together with fresh paste it wouldn't be far off.

If anything, this exercise has given me more respect for the stock AMD fan. I shall clean it up and pack it carefully away, I'm sure it will get used again at some point.