AMD Precision Boost 2 and Wraith Prism Deep Dive @ [H]

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,601
AMD Precision Boost 2 and Wraith Prism Deep Dive

When AMD presented it new Precision Boost 2 technology, I have to think AMD undersold the feature in terms of what it was actually able to accomplish. The original Precision Boost on the 1st gen Ryzen processors was anemic at best. After seeing what Precision Boost 2 can accomplish, do we even need to try to overclock our new 2700X? Blasphemy!
 
Awesome review Kyle, not only AMD have learned from its mistake with PB1 but have gone back to the drawing board to build PB2. Looking forward on retiring my i3570k system for a 2600X.
 
yes....

thank you very much, i have been waiting for this article since your 2700x review when you mentioned this was coming... I think i too will pick up a 2600x now
 
I think if AMD had more OC headroom, PB2 wouldn't be preferred to good-ol elbow grease and manual OCs.

But the fact that this chip is pretty much at peak performance out of the box is awesome. good effort by AMD.
 
How were your temperatures using the Wraith Prism? That would be my only concern with using it over an AIO system I think. Also like you said in the summary page, it's awesome that the chip is almost at peak performance out of the box, and manual OCing might not warrant many gains. I am looking forward to building around an AMD processor soon, first one since my s939 3800+.
 
Will testing be done with OCing the base clock? I heard that you can extend PB2 clocks by OCing the Base clock with the clock generator. Maybe a viable performance increase beyond PB2 will be using a higher base clock with PB2, getting the top performing cores over 4.4-4.5 ghz.

Interesting take on PB2, it shows favorably across many different type of workloads! Rather amazing. I wished power measurements were taken between the different settings in the review.
 
What? No more manual overclocking? It's even easier now, the CPU does it for you and AMD provided a decent heatsink?

skynet confirmed


Oh well, at least I still have the option of installing an even beefier dual tower heatsink or 240/280mm AIO
 
noko said:
Will testing be done with OCing the base clock? I heard that you can extend PB2 clocks by OCing the Base clock with the clock generator. Maybe a viable performance increase beyond PB2 will be using a higher base clock with PB2, getting the top performing cores over 4.4-4.5 ghz.

Interesting take on PB2, it shows favorably across many different type of workloads! Rather amazing. I wished power measurements were taken between the different settings in the review.
Click to expand...
That is for another article. Rome was not built in a day.
 
I simply went to pull the cooler off and yanked it right out of the socket. And in doing that, I bent several pins
Click to expand...
My biggest fear when removing the cooler to sell off the cpu.
 
Thanks for the excellent article, of course you managed to post it while I was busy ordering the parts for my build so I didn't see it until after but I can't really blame you for that. Luckily it just confirms my decision to start with the Prism cooler and decide where I want to go from there.

Those retention clips remind me of old Pentium 2 era Celeron heatsinks that still give me nightmares so I was glad to hear you mention that they're simple to use, I would actually prefer a more traditional backplate with screws myself but maybe I'm in the minority.

I agree with the suggestion to take a look at BCLK overclocking, if that works well it might give me the incentive to actually upgrade the cooling. As it is I decided to get a better kit of memory that will hopefully allow for high speeds and tight timings just so I have something to play around with since I'm not planning to overclock.
 
Skylinestar said:
My biggest fear when removing the cooler to sell off the cpu.
Click to expand...
push down and twist, usually does the trick every time even if the thermal paste is cold.

Ebernanut said:
Thanks for the excellent article, of course you managed to post it while I was busy ordering the parts for my build so I didn't see it until after but I can't really blame you for that. Luckily it just confirms my decision to start with the Prism cooler and decide where I want to go from there.

Those retention clips remind me of old Pentium 2 era Celeron heatsinks that still give me nightmares so I was glad to hear you mention that they're simple to use, I would actually prefer a more traditional backplate with screws myself but maybe I'm in the minority.

I agree with the suggestion to take a look at BCLK overclocking, if that works well it might give me the incentive to actually upgrade the cooling. As it is I decided to get a better kit of memory that will hopefully allow for high speeds and tight timings just so I have something to play around with since I'm not planning to overclock.
Click to expand...


i don't mind either mounting and actually expected the ryzen mounting to be the same old lock down bracket they've always used.. both work perfectly fine but i can completely understand why the direct to backplate mounting was a bitch for casual pc builders due to the fact that some motherboards came with the back plate stuck to the board, some it was optional, and some were just a piece of plastic with 4 screw holes on it and made it a royal pain in the ass to mount(cough asrock cough) if the board was installed first.
 
im not a fan or RGB BS because they usually cut the product back some IMO to deliver that lighting effect, in this case, AMD gave the BEST they were able to give AND give a very nice cooler, I would be T_T if I could not see the light show cause dang that is a pretty little thing ^.^

here is hoping the industry starts giving AMD the attention they deserve for all the crud and hard times they have went through in the last decade (some they made happen to themselves, much it was their competition that played the game against them with $$$$$$$$ instead of just racing the race)

thanks for the review as usual Kyle bang up job o7
 
The stock TIM is like glue. The twist back and forth to get it loose trick doesn't work most of the time with the stock cooler and stock TIM for me. I have ended up pulling the cpu out of the socket 3 or 4 times now with new Ryzen cpus. I just had this happen over the weekend with a Ryzen 3 2200G. It would not pull off the heatsink even after getting it out of the socket. I had to slide the cpu over to one edge and off that way. So far I have been very lucky and have not bent any pins but that's just dumb luck.
 
noko said:
Will testing be done with OCing the base clock? I heard that you can extend PB2 clocks by OCing the Base clock with the clock generator. Maybe a viable performance increase beyond PB2 will be using a higher base clock with PB2, getting the top performing cores over 4.4-4.5 ghz.

Interesting take on PB2, it shows favorably across many different type of workloads! Rather amazing. I wished power measurements were taken between the different settings in the review.
Click to expand...
If the 2800X does come around and is a new stepping with more room for higher clocks this could be really nice.
 
I was impressed with it when you did the first article about the CPU and even more impressed seeing how well the feature is working. With something like this, we could see GPU type overclocking, where we decrease temps, add power allowance, then add voltage and let it figure itself out!
 
Well, Kyle... I can see a benefit from going water. So maybe I should do that.

Then again, the Prism is pretty, and I wonder how it compares to my Noctua U12 - which is a great performing air cooler, but looks like the wrong end of a donkey (whoever came up with the idea of brown and beige fans, anyway).
 
dragonstongue said:
im not a fan or RGB BS because they usually cut the product back some IMO to deliver that lighting effect, in this case, AMD gave the BEST they were able to give AND give a very nice cooler, I would be T_T if I could not see the light show cause dang that is a pretty little thing ^.^

here is hoping the industry starts giving AMD the attention they deserve for all the crud and hard times they have went through in the last decade (some they made happen to themselves, much it was their competition that played the game against them with $$$$$$$$ instead of just racing the race)

thanks for the review as usual Kyle bang up job o7
Click to expand...
Really? Seriously, I have never direct heatpipe HSF hold that kind of vacuum on a AM4 mount so I truly was not expecting the issue. That said, to get a picture of the mating surface, moving it to the side destroys the footprint imprint, which is no good for pictures. Was more of a PSA than anything, but I was not expecting the direct heatpipe unit to pull it out.
 
DuronBurgerMan said:
Well, Kyle... I can see a benefit from going water. So maybe I should do that.

Then again, the Prism is pretty, and I wonder how it compares to my Noctua U12 - which is a great performing air cooler, but looks like the wrong end of a donkey (whoever came up with the idea of brown and beige fans, anyway).
Click to expand...
Guess Nocuta made that their trademark color scheme when ass was in style?
 
keep-calm-and-lets-get-back-to-the-topic.png
 
Am waiting for my Seasonic to arrive, at which point..
I mentioned it elsewhere, i kinda like this, i mean i can actually afford to be lazy now and do nothing; without losing in performance either! :)
So while i can understand the sentiment of an "enthusiast", i don't quite share it; i mean the tweaking is fun, sure, but waiting for 4 and 6 hours Prime/OCCT runs to fail? Just so you turn a knob and rerun them? Over and over and over? For days in a row? Was never a fan of that, i can't see how anyone is in all honesty. I do it for the performance benefit, i do it as i consider it a further investment performance-wise, sure; but the actual process? Never a fan.
So TLDR, if purchasing a good mobo now alleviates all this tedium without costing me in terms of performance? One very happy customer right here.

Of course, one's not really escaping it all, i mean if you have the cooling power, you need at least try and see if you got lucky. If you can hit that 4.3 on your own, different story then. For the remaining 99% of silicon owners, i call this a welcome development and may more follow it.

(can we now try the same for RAMs please?)

And thanks for the review :)
 
As a heads up/extra data point, on my Asus Prime X370 Pro board, there is NO setting to enable/disable Precision Boost 2, XFR2, or anything. There are no levels, either. The BIOS does not allow any tinkering with this.

But it appears to be enabled anyway, because I'm getting scores broadly similar to Kyle's scores with the Prism cooler and XFR2/PB2 level 2. And I see in Ryzen master that the boost is operating properly, with 3975 all core boost in Cinebench, 4100 all core boost in games, 4200 all core boost on very light loads, and 4350 max single core boost.
 
bobzdar said:
What are your idle temps with the prism? Mine seem high, mid to upper 40'sC so I had to re-tune my fan curve to keep it quiet at idle. The prism is really loud so I don't have it start to ramp up until the processor is over 50C and it goes quickly to full speed at 60C. However, it's able to maintain temp in the low 70's for the most part when I do that, just really loud. My old 1700 would idle in the 30's with the wraith spire and really low rpm, so just curious if you're seeing high idle temps or if there's something funky with my system. I have it set to balanced power plan and it drops to 2.2ghz ish at idle. I didn't have any instability fortunately, in actuality mine has been up 24/7 since I put the chip in, haven't had to do a single reboot since I first popped it in on the 20th!
Click to expand...
You either have a bad mount, or a bad cooler, or a bad fan unit. I never even remember hearing the Prism I was using.
 
bobzdar said:
I don't think it's a bad mount as it's only hitting low 70's at full 8 core load and maintaining 3.975ghz, but could be a bad fan unit. What did you have your fan curve set to, default? I flipped the switch to high speed on mine and then set the curve using AI3. Mine is really loud when it goes to 100% and is the only reason I'm contemplating going aftermarket.
Click to expand...
I literally did NO TUNING on the system in any regard.
 
Excellent writeup!

Very nice to see AMD providing 'the whole solution' out of the box here, and aside from unwise memory purchases, AMD has laid out an effective solution that gets high performance without significant user effort!
 
Between PB2 making overclocking all but pointless and the prism cooler actually making me consider removing my Arctic 240 AIO, I feel like someone needs to take my [H] card away from me. Hats off to AMD for this one. I quite literally never thought I'd see the day.
 
After seeing how well the stock TIM AMD applied to these did, I kinda wish they didn't have the RGB/USB cables right there near it. On my cooler one of the cables came loose and was stuck in the TIM, removing the cable removed a patch in that area. Not a big deal for me as I am not using the cooler (Dark Rock TF) but it makes it harder to sell mine as 'brand new'
 
Last edited:
My 2700X, Asus C7H wifi, GSkill Flare X 3200 C14 and NH-D15S should be here today. Looking forward to getting them running after seeing this article, but it won't be until the weekend, unfortunately. I grabbed that EFI release as well. Thank you for the link to that, Kyle, even if I decide not to flash it right now. Hopefully it "officially" releases from Asus soon.
 
Azrak said:
My 2700X, Asus C7H wifi, GSkill Flare X 3200 C14 and NH-D15S should be here today. Looking forward to getting them running after seeing this article, but it won't be until the weekend, unfortunately. I grabbed that EFI release as well. Thank you for the link to that, Kyle, even if I decide not to flash it right now. Hopefully it "officially" releases from Asus soon.
Click to expand...
That will be the official EFI release from ASUS, so it is safe. That said, between that EFI and the new chipset driver, I did see perf increases.
 
bpizzle1 said:
Between PB2 making overclocking all but pointless and the prism cooler actually making me consider removing my Arctic 240 AIO, I feel like someone needs to take my [H] card away from me. Hats off to AMD for this one. I quite literally never thought I'd see the day.
Click to expand...

It happens, mang. I ran through all the trouble of getting 4.3GHz OC to work on this CPU... achieved it... and then said fuck it, and went back to stock, because the gains were so small.
 
bpizzle1 said:
It looks like we can still fiddle around with memory timings and coax out some decent gains. I know you're probably sick of benching this thing, but you guys have any plans on doing any memory scaling testing with Ryzen+?

Edit : PB2 is actually awesome for the general public. AMD may become the platform for a lot of people that don't want or know how to OC. A lot of the more casual gamers I know fall into that category.
Click to expand...
I think that memory scaling is fairly well documented already, no sense in beating a dead horse.

I would suggest that PB2 is going to be of value to a lot of enthusiasts as well that like their cooling.
 
Nothing like the CPU getting the most out of itself. That is great to see. I wonder if we will see something similar from Intel soon.
 
Shmee said:
Nothing like the CPU getting the most out of itself. That is great to see. I wonder if we will see something similar from Intel soon.
Click to expand...

You better believe Intel already has people looking into how PB2 works.
 
  • Like
Reactions: Aenra
like this
We'll see Intel putting more envelope-searching performance management in the CPU over time. This stuff is especially important to eke out the last drop of battery performance in "race to idle" and clearly this shows available gains (as long as these chips don't start dying from flying too close to the sun).

Of course, Intel is super conservative on design and don't really design for desktop; laptops and servers, yes, but then desktop processors get the trickle down.
 
High resolution low image quality is the exact opposite of how CPU benchmarks should be run for games. Many settings in modern games impact the CPU so running low settings gives you no clue of how the CPU would perform in realistic settings. Just the Shadows setting alone in many games can have an absolutely massive impact on CPU performance. For instance in parts of Dishonored 2 I stand no chance of holding even 60fps with shadows on Ultra but dropping to high gives a 30 to 40% increase in frame rate and it is purely held back by the CPU as even running 720p makes absolutely no difference in the areas where I'm dropping below 60 because of the Shadow setting being on Ultra.
 
IdiotInCharge said:
While you're generally right, this writeup is very specific in its purpose: [H]ard|OCP is not testing gameplay performance/experience, but making strict performance comparisons with a particular CPU feature on a particular CPU and tailoring the tests to expose the effects of that feature.
Click to expand...

Such CPU tests are very useful in exposing certain differences in the CPUs. But it's also a bit of a crystal ball guesswork situation, too. We use low res game benchmarks to remove the GPU as a bottleneck in order to get a look at future CPU performance with faster GPUs. Yet that doesn't really work, fully. We don't know how quickly more cores/threads will be utilized by games in the future. If more thread adoption comes quickly, then despite Ryzen's single core disadvantage relative to Intel competition, Ryzen may actually fare better than equivalently-priced Intel CPUs in the long run. OTOH, slow adoption means the opposite, future proofing goes more to Intel than AMD because of Intel's generally superior single thread and lightly-threaded performance. It's hard to say how it will go. So I think it's good to keep tabs on both datapoints, low res gaming and high res gaming. Double check that the latter shows little difference between CPUs (if the difference is large between CPUs, we have major problems) - but use the former as a way to gauge lightly threaded performance.
 
You must log in or register to reply here.
Back
Top