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

Wow, XFR2/PB2 really came good. I remember them hyping the original XFR as this amazing dynamic boost system, and it ended up being a few extra hundred mhz in a really non dynamic way, glad they got it right this time.

Intel has to be worried, comparing 8700k to 2700x really does make AMD look good. For the same price (~$330), AMD gives you 8C16T, a great stock RGB cooler, a very effective auto overclock setting, and quality TIM. Intel makes you pay more for an unlocked processor, doesn't include any cooler at all, and cheaps out on shitty TIM.

The only question I have after reading the article is how well a decent/good tower cooler ($25-45) would do. How far would a Cryorig H7 or similar cooler allow PB2 to go, I'm guessing you could get pretty close to the custom loop, but it would be an interesting data point to see.
 
Great review Kyle and good to see the gaming results compared to the Intel CPU, AMD appear to have closed the gap !!
 
Excellent review, thanks!

I'm wondering if the 3400 memory speed might have hurt results at all? I know on the Intel's, anything beyond 2666 is minimal gains, and at some point actually start hurting performance. Not sure that is still the case, or if it would apply to an AMD system.
 
GoodBoy said:
Excellent review, thanks!

I'm wondering if the 3400 memory speed might have hurt results at all? I know on the Intel's, anything beyond 2666 is minimal gains, and at some point actually start hurting performance. Not sure that is still the case, or if it would apply to an AMD system.
Click to expand...

Ryzen+ has shown to get pretty significant gains from fast memory with tight timings. 3200 c14 seems to be the sweet spot for the money.
 
  • Like
Reactions: N4CR
like this
GoodBoy said:
I know on the Intel's, anything beyond 2666 is minimal gains
Click to expand...

Erm, no? Who told you that?

You will hear of a 'sweet spot' around 3200GHz, but that relates to cost vs performance, not performance per se. And i can definitely assure you that while there's definitely a diminishing return? Your 2666 and, say, a 3666 is night and day in many things. All the way down to opening your start menu, a simple document, or even your Window's control panel.
This can be exacerbated in badly optimised games or in throughput-intensive workloads, where there's a noticeable benefit all the way up to 4K (albeit still smaller relatively to 2xxx vs 3xxx).

The difference with Ryzens is the interconnect process. Different function and one related to RAM; but this is a function that ultimately amounts to complex-to-complex trip latencies, hence relevant to the CPU's 'actual' performance. The gains outside of that, ie raw RAM speed-related, are still there for both AMD and Intel.
 
Last edited:
Nice !

my first thought was, why even try overclocking this CPU ? Spot on !

Thanks for the hint with the power mode bug, I think this could explain my shutdowns as well with rig from sig. I will shuffle the modes and see if it fixes it ,)


Great report, true word, 100% [H]
 
Nice article, Thanks, Kyle.

As mentioned, many have reported (in the OCN C6H + C7H OC threads + reddit) nice ST + MT gains by using a combination of bclk OC + CPU vcore offsets of between -0.05 and -0.1 with these 2600 and 2700s, so I also look forward to any further OC articles on that.
 
bl4C3y3 said:
so funny when picturing just that (while reading a [H] enthusiast article about overclocking :D)
Click to expand...
Yeah, you are correct. But I wanted to give it a fully stock run.

But get a load of this on the 4.2GHz hand overclock. All I did was set the multiplier to 42 and turned off XFR2. I touched nothing else. That motherboard is friggin sweet.
 
BitMaster said:
Thanks for the hint with the power mode bug, I think this could explain my shutdowns as well with rig from sig. I will shuffle the modes and see if it fixes it ,)
Click to expand...
I fought with that for about a day before I got word from Asus. I would have never just cycled through the power plans, but it worked. Let me know your results.

DeeJayBump said:
As mentioned, many have reported (in the OCN C6H + C7H OC threads + reddit) nice ST + MT gains by using a combination of bclk OC + CPU vcore offsets of between -0.05 and -0.1 with these 2600 and 2700s, so I also look forward to any further OC articles on that.
Click to expand...
Yeah, that is actually outlined in the document I linked in the article. Scroll about 2/3 the way through that PDF.
 
What versions of Prime95 were used for stress testing?
This is very interesting. Thank you.
 
Aenra said:
Huz two questions, it does.
- Are we to see a repeat of this for TR+ CPUs? Assuming a sample is sent, etc. etc.?
- Mere curiosity, your reply will not reflect on anything (/readies pitchforks and torches, oils the chainsaw). Given that nothing changes, has PB2 affected your mindset regarding upgrading?
Click to expand...
No hands on yet, so who knows.
 
  • Like
Reactions: Aenra
like this
DeeJayBump said:
Yep, there's also a V03 of the doc you posted in the first post of the http://www.overclock.net/forum/11-a...09-rog-crosshair-vii-overclocking-thread.html OCN C7H Thread, FWIW. Hope there's no problem with posting that link here, I'll remove it if so.

Gains seen from under-volting CPU vcore make sense simply from the giving PB2/XFR /PBO more potential thermal headroom to play with as that's at least an element of the Performance Enhancer 3 + 4 modes that The Stilt/Asus created, which as you've demonstrated really appear to be designed to shine with more [H]/extreme cooling setups.
Click to expand...
If you check out article, in the nice big table I took time to make, you will see those addressed. Lvl3 and lvl4 were locked clicks and voltages under load.
 
Keep the heat-sink and spend the $$ you save on LL RAM then tweak the timings.
 
  • Like
Reactions: N4CR
like this
Outsideloopcomputers said:
So, without overclocking....what are we supposed to do with the rest of our lives now?
Click to expand...

Enjoy our hard earned money for once?
Save hundreds of hours staring at a Prime95 instance running?
Never fear that a month or two down in the road we might need an extra 0.02 volts because the VRM has finally stabilised?
Oops, and run another Prime95 just to make sure 0.02 is all it needed?

I mean Jeez, i do not know how i'll cope, lol :)

Joking aside, i find that beyond the hardware mounting, the installing and looking at some nice final number, the process itself is a borefest. If it's something you do for the process rather than the result, chances are you're mostly looking at Intel CPUs anyway; which is perfectly fine needless to say. It's not like R7s would have hit 35 and 40% OCs if it wasn't for PB2.
 
Last edited:
  • Like
Reactions: DF-1
like this
Outsideloopcomputers said:
THAT is the kind of review that made [H]ardOCP the BEST, back in the good old days! Great work Kyle.

So, without overclocking....what are we supposed to do with the rest of our lives now?
Click to expand...
I would not rule out OCing yet until BCLK OCing is tested out with PB2. I still have a feeling it may be minimal in the end but then Overclockers want to push the limits sometimes and get every cent worth of performance available.
 
  • Like
Reactions: Aenra
like this
DuronBurgerMan said:
Hey that's handy. Thanks! 5C less... but look like ass. Or look nice but 5C hotter. Hmmmmmmmmmm...
Click to expand...

I'd honestly take the 5C and not have to have that giant ass cooler taking up 1/2 of my case. For real though, the prism is darn impressive for what it is.
 
Outsideloopcomputers said:
THAT is the kind of review that made [H]ardOCP the BEST, back in the good old days! Great work Kyle.

So, without overclocking....what are we supposed to do with the rest of our lives now?
Click to expand...
Well there is still GPU, and with Ryzen the RAM. CPU O/C isn't that complicated. A few variables, test/repeat. RAM tweaking is truly [H]ard in every sense of the word.
On Ryzen 1000 series there are instances where lowering the chip to 3.7 GHz, to ease the stress off the IMC, then pushing incredibly tight timings on the RAM results in performance that bests the fastest straight CPU O/C's.
It is a new type of chip. We just have to change our way of thinking of O/C'ing. There is plenty there to keep you challenged.
[H] won't be loosing us anytime soon :)
 
Thank you so much for mentioning the power plan bug! I've been stuck for about a week having absolutely no idea what was wrong and why my crashes were so intermittent.

I'm not sure what I'm going to do with my weekend now, it seems I no longer need to RMA my mobo or CPU & the CPU is going to overclock itself, this is amazing.
 
Awesome article Kyle! I'm going to read it again tomorrow when I have more brain cells active with coffee.

Hmmm, something about this quote though.... "I do not recall a single time in the last twenty years that I would have suggested to our HardOCP readership that you give the CPU cooler that came in the box a try."

I seem to remember back in the dark ages, where you were thrilled that the Prescott P4 would get over 3ghz with the bleh stock intel cooler. I remember building my only P4 system based on that review. :)
 
Burticus said:
Awesome article Kyle! I'm going to read it again tomorrow when I have more brain cells active with coffee.

Hmmm, something about this quote though.... "I do not recall a single time in the last twenty years that I would have suggested to our HardOCP readership that you give the CPU cooler that came in the box a try."

I seem to remember back in the dark ages, where you were thrilled that the Prescott P4 would get over 3ghz with the bleh stock intel cooler. I remember building my only P4 system based on that review. :)
Click to expand...

I think you're a bit mistaken. First of all the Pres-hott even at stock speeds surpassed a space heater and was a full on furnace. There's no way running a stock heatsink would have been a good idea on those things unless underclocked and undervolted.

You're more than likely thinking of the Northwood 2.4. Almost all of those things were hitting 3.2Ghz+ mainly because yields were so good and demand was so high for them that Intel was labeling 3.2Ghz parts as 2.4Ghz. In that case, a stock heatsink had a chance of doing some decent "overclocking" because you were actually running the CPU at what should have been near the actual clock speeds. However, that doesn't mean you didn't want to use much better cooling.

As for the article, I'm damn glad you did it. With the rather amazing boosting AMD built into the CPUs I think it was imperative that an article such as this was a necessity. Knowing that the stock cooler can give you this level of performance is a data point we needed to see especially for those of us needing to stretch every dollar.
 
Burticus said:
Awesome article Kyle! I'm going to read it again tomorrow when I have more brain cells active with coffee.

Hmmm, something about this quote though.... "I do not recall a single time in the last twenty years that I would have suggested to our HardOCP readership that you give the CPU cooler that came in the box a try."

I seem to remember back in the dark ages, where you were thrilled that the Prescott P4 would get over 3ghz with the bleh stock intel cooler. I remember building my only P4 system based on that review. :)
Click to expand...
Well, that does not mean I remember it. :p But yeah, I wanna see proof. And I am not spending time looking that up.
 
I pulled mine out of the socket as well. You warning to everyone about wiggling it left to right 1st is extremely valid.
 
Aenra said:
Erm, no? Who told you that?

You will hear of a 'sweet spot' around 3200GHz, but that relates to cost vs performance, not performance per se. And i can definitely assure you that while there's definitely a diminishing return? Your 2666 and, say, a 3666 is night and day in many things. All the way down to opening your start menu, a simple document, or even your Window's control panel.
This can be exacerbated in badly optimised games or in throughput-intensive workloads, where there's a noticeable benefit all the way up to 4K (albeit still smaller relatively to 2xxx vs 3xxx).

The difference with Ryzens is the interconnect process. Different function and one related to RAM; but this is a function that ultimately amounts to complex-to-complex trip latencies, hence relevant to the CPU's 'actual' performance. The gains outside of that, ie raw RAM speed-related, are still there for both AMD and Intel.
Click to expand...

Thanks for clearing that up. Can't recall where I heard that, but I thought it was some game benchmarks at varying ddr speeds...

I have 3400 in my system now running at 2666, so I need to try and speed it up from the sounds of it.
 
  • Like
Reactions: Aenra
like this
GoodBoy said:
I have 3400 in my system now running at 2666, so I need to try and speed it up from the sounds of it.
Click to expand...
Yes you do! Does it have profile on it? You can check that with CPUz. If it does, I would suggest getting the latest EFI for your board and seeing if you can simply turn on the profile.
 
Kudos to AMD for shipping there CPU with possibly the best stock cooler ever. Even the 2nd tier Spire makes Intel's cooler look like a joke.
 
GoodBoy said:
try and speed it up from the sounds of it.
Click to expand...

Sometimes XMP doesn't work, sometimes it does but with extreme over-volting; which by the way is not a sign of a 'bad' motherboard, as each chip's memory controller differs (silicon lottery) and to ensure all CPUs can cope, mobos provide a.. healthy over-volting, you know, so you cannot go sue anyone because 'x doesn't work'.
If you have the time/inclination, some thorough memory testing/voltage monitoring is really advised:

HWInfo for voltages (if something is lacking, can happen with very odd combos of Gigabyte mobos+AMD CPUs), OCCT is your alternative.
HCI memtest for memory stressing (you don't need any knowledge, it's literally click and run if you purchase the pro or deluxe version; the freebie mode does entail your having read the manual however, as it requires some manual tweaking).
*not to be confused with memtest+ or memtest86, am talking this one right here: http://hcidesign.com/memtest/download.html

Have fun and as mentioned, once stable, you'll definitely see a difference :)
 
Formula.350 said:
Great review, Kyle! I think it's fair to say that PB2 is the star of Zen+. Though, I can't help but wonder if maybe it was intentionally held back from the initial Ryzen to be included in this refresh, as a selling point for anyone who currently has a Ryzen...
Click to expand...

I highly doubt that. I suspect it's the other way around. AMD knew they had to get Zen out the door and return to competition, and they pushed the product out before they completed all the possible boost and latency tuning they could've done, given more time. Especially given single core IPC and boost improvements with PBR2 would have partly addressed Zen's Achilles heel: lightly threaded, latency-sensitive performance.
 
DuronBurgerMan said:
I highly doubt that. I suspect it's the other way around. AMD knew they had to get Zen out the door and return to competition, and they pushed the product out before they completed all the possible boost and latency tuning they could've done, given more time. Especially given single core IPC and boost improvements with PBR2 would have partly addressed Zen's Achilles heel: lightly threaded, latency-sensitive performance.
Click to expand...
You're right, that's a bit more logical, especially for PB2. There was some polishing left fo Ryzen at launch, no doubt, not just on the motherboard side of things. It seemed clear from the start that they wanted PB to function like it does now. They always talked it up as being dynamic, but it just didn't operate that way and was little more than the Boost on their past chips (and Intel's for that matter). It'll be very interesting to see what they change with Zen2 in regards to PB, that is, if anything! It certainly seems to be quite awesome now in its current form!

The ability to run memory faster and their other smaller tweaks that have gotten latencies down, I definitely chalk that up to the typical post-release revisions that we've always seen occur. There's always going to be some hind sight for things they should've done differently, as well as new discoveries of things that can be changed for better results.

Regardless, I just find PB2 to be a deal maker for Ryzen+, and had it been working the same as the original Ryzen... this would be labeled a very lackluster update to the CPU. Granted, there's the improvements and the higher achievable memory speeds, but it seems like a lot of the performance brought by Ryzen+ is thanks to what PB2 is now capable of. I just hope the general consumers will be able to appreciate it as much as they should (y)
 
Formula.350 said:
Regardless, I just find PB2 to be a deal maker for Ryzen+, and had it been working the same as the original Ryzen... this would be labeled a very lackluster update to the CPU. Granted, there's the improvements and the higher achievable memory speeds, but it seems like a lot of the performance brought by Ryzen+ is thanks to what PB2 is now capable of. I just hope the general consumers will be able to appreciate it as much as they should (y)
Click to expand...
The weird part is probably that the amount of sensors between the 1000 and 2000 series did not change but the PB algorithm did.
 
Skylinestar said:
My biggest fear when removing the cooler to sell off the cpu.
Click to expand...

I've been telling this to people for years. Use a hair dryer to heat the heatsink before removing it. Undo the clips of course and heat it up good, then gently twist side to side. Always pops right off without pulling the CPU from the socket. :)
 
  • Like
Reactions: Aenra
like this
Pieter3dnow said:
The weird part is probably that the amount of sensors between the 1000 and 2000 series did not change but the PB algorithm did.
Click to expand...
And lessons have probably been learned from Intel's mistakes, so it's probably not stored anywhere accessible like Intel ME to allow it to be updated. As much as I'd like for it to be, so that Ryzen 1 could get PB2, I'd rather it also not be, simply for security sake. Last thing we need is a malware that can tell our chip "lawl it's really -50C right now, and you're only getting 1V, crank everything up!" *explodes*. :p

Even if it could be, I don't think they'd release an update for Ryzen 1 since that'd be taking away a large selling point (at least IMO).
At least it's possible to utilize on X370 boards (y) So I'm thankful for that much!
 
freeloader1969 said:
I've been telling this to people for years. Use a hair dryer to heat the heatsink before removing it. Undo the clips of course and heat it up good, then gently twist side to side. Always pops right off without pulling the CPU from the socket. :)
Click to expand...
Or just load the CPU with a stress load for 20 minutes before taking cooler off.
 
ACC said:
Or just load the CPU with a stress load for 20 minutes before taking cooler off.
Click to expand...

That will work too but the hair dryer will get the heatsink hotter faster without having the need for the system power to be on. Either way works but I prefer my method. :)
 
You must log in or register to reply here.
Back
Top