Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

Page 129 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Agent-47

Senior member
Jan 17, 2017
290
249
76
Does Linux have SMT scheduling issue like windows too? Switching CPU governor from ondemand to performance in the Linux kernel does boost ryzen performance quite a bit interestingly enough.

I wont be surprized if it has the same issue. its a shame they did not do a SMT disable run.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Ryzen has consistent lower gaming performance vs 7700k on nonwindows platform as well.
http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-gaming&num=1
Does Linux have SMT scheduling issue like windows too? Switching CPU governor from ondemand to performance in the Linux kernel does boost ryzen performance quite a bit interestingly enough.

Well of course it is consistently worse than a 4.5Ghz Kaby Lake in games. Who in their right mind would expect it to be as good as - or better than - a high IPC speed demon?

6900k is where comparisons should be made - and whether or not someone who needs that level of performance can accept the gaming results.

I mean, seriously, Ryzen performs pretty closely to the 4690k in gaming - which has been a perfectly acceptable level of performance up until this moment. In some games Ryzen 7's extra cores help overcome its clock-speed deficit... but it STILL has a MINOR IPC deficit.

We all need to step back and recognize one critical point: Just a few months ago not a single one of us thought that Ryzen 8-core CPUs would even get close to the 6900k. We were, best case, hoping for Haswell IPC - and not expecting that to be reached.

Now, AMD delivers us a CPU that has Broadwell IPC (or even slightly better) in many tasks to which it is well suited and its worst-case tasks are around Haswell level (and, yes, with a few lower outliers).

We should be singing AMD's praises for accomplishing that and then delivering it to us for as little as $329... and not playing any of the gimping nonsense of which Intel is so fond.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Joker mentioned the 1700 being cooler than his 1800x too, but I assumed that was just a coincidence. He said that the 1700 required less voltage to hit 3.9 GHz.

Could the 1700s be low TDP binned chips, rather than the others being binned for higher clocks?

Higher leakage parts usually overclock better once you cool them properly. If overclocking on air, just get the 1700 and save a few bucks. On water, though, a 1700X or an 1800X might be worth the extra money.

I went with the 1700X - which has been delivered. Apparently stock for my motherboard should mid-week.
 

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
Higher leakage parts usually overclock better once you cool them properly. If overclocking on air, just get the 1700 and save a few bucks. On water, though, a 1700X or an 1800X might be worth the extra money.

I went with the 1700X - which has been delivered. Apparently stock for my motherboard should mid-week.

I got the 1700 Taichi D15 combo on the way.

Probably could have gotten a cheaper board, but I wanted a good reliable mb that would last several years with an oc.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Why the D15 & not a Corsair H110i or something like that?

D15 is plainly awesome. I'm running one now on my 2600k @ 4.5Ghz.

Considering Ryzen can't clock enough to become thermally limited (it can barely reach the power levels of my CPU... while handily outperforming it in most cases - even with my overclock) it will do very well with the D15.
 
  • Like
Reactions: Drazick

CHADBOGA

Platinum Member
Mar 31, 2009
2,135
832
136
D15 is plainly awesome. I'm running one now on my 2600k @ 4.5Ghz.

Considering Ryzen can't clock enough to become thermally limited (it can barely reach the power levels of my CPU... while handily outperforming it in most cases - even with my overclock) it will do very well with the D15.
Won't it do just as well with a H110i?
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Interesting variance in the BCLK of the Crosshair VI Hero. And it obviously affects memory timings.

What's really interesting is that Asus suggests that using the "REFCLOCK" instead of the multiplier to get better results.

Makes sense - that way you can keep all of the goodness still working.

Crazy thing is they claim that >120MHz should be stable if you set everything to PCI-e 2.0 mode.

I'll gladly keep using x4 PCI-e 2.0 for my NVMe drive if it gets the 1700X that little extra boost without losing all of the goodness of Precision Boost and XFR.

Oh, awesomely, Precision Boost does NOT have a single-core turbo. It's dual core turbo :p XFR can bump the rest up a touch, depending on temperatures and power.

One thing is for certain: Avoid putting Ryzen into OC mode (adjusting the multiplier above stock) unless you are just trying to see how far you can push it overall on all cores.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Watching the games actually being played makes the differences seem much less important. Seeing those difference on charts in the form of mins and averages makes it LOOK far, far worse than it ends up being in the real world.

You can't judge actually quality from compressed youtube videos.
 

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
Ryzen is painted as a bad gaming system because a number of "influencial" review sites did their utmost best to avoid using titles that they knew AMD would show parity or even beat Intel at, and in the worst case scenario trail slightly. Games like Arma 3, Battlefield Multiplayer, Battlefront, Sniper Elite, Watchdogs, Hitman, Doom, AOTS and a few others were deliberately omitted to and games that are historically Intel driven and where AMD has not foot in were used to skew results negitively.

The real performance is nowhere near as bad but certain reviewers have shown their true colours and are now on my IP blocklist.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
What's really interesting is that Asus suggests that using the "REFCLOCK" instead of the multiplier to get better results.

Makes sense - that way you can keep all of the goodness still working.

Crazy thing is they claim that >120MHz should be stable if you set everything to PCI-e 2.0 mode.

I'll gladly keep using x4 PCI-e 2.0 for my NVMe drive if it gets the 1700X that little extra boost without losing all of the goodness of Precision Boost and XFR.

Oh, awesomely, Precision Boost does NOT have a single-core turbo. It's dual core turbo :p XFR can bump the rest up a touch, depending on temperatures and power.

One thing is for certain: Avoid putting Ryzen into OC mode (adjusting the multiplier above stock) unless you are just trying to see how far you can push it overall on all cores.
I usually prefer to have things set up with the minimal amount of tinkering - and I'm not going to OC the 1700 when I get it anyway, because I'll lose out all the Power Mananagment goodies that Stilt mentioned in the other thread. But the Asus board seems perplexing to me.

PCI-e 2.0 x16 should not matter if you are running a single GPU, even at 4K. Though I would not like to have the theoretical bandwidth for storage getting halved.

It would be interesting to see what Precision Boost and XFR can do in the 1700.
 

bystander36

Diamond Member
Apr 1, 2013
5,154
132
106
Ryzen is painted as a bad gaming system because a number of "influencial" review sites did their utmost best to avoid using titles that they knew AMD would show parity or even beat Intel at, and in the worst case scenario trail slightly. Games like Arma 3, Battlefield Multiplayer, Battlefront, Sniper Elite, Watchdogs, Hitman, Doom, AOTS and a few others were deliberately omitted to and games that are historically Intel driven and where AMD has not foot in were used to skew results negitively.

The real performance is nowhere near as bad but certain reviewers have shown their true colours and are now on my IP blocklist.
A couple of those are in THG's review:
http://media.bestofmicro.com/ext/aH...jU2MTc3L29yaWdpbmFsLzAyLnBuZw==/r_600x450.png
http://media.bestofmicro.com/ext/aH...jU2MjAzL29yaWdpbmFsLzAyLnBuZw==/r_600x450.png
 
  • Like
Reactions: DarkKnightDude

looncraz

Senior member
Sep 12, 2011
722
1,651
136
I usually prefer to have things set up with the minimal amount of tinkering - and I'm not going to OC the 1700 when I get it anyway, because I'll lose out all the Power Mananagment goodies that Stilt mentioned in the other thread. But the Asus board seems perplexing to me.

PCI-e 2.0 x16 should not matter if you are running a single GPU, even at 4K. Though I would not like to have the theoretical bandwidth for storage getting halved.

It would be interesting to see what Precision Boost and XFR can do in the 1700.

That's the beauty of base clock overclocking - the CPU won't trigger OC mode. Though I don't expect to keep much of an overclock myself. I was hoping for less performance than the 1700X offers stock, so I won't be upset no matter what :p
 

AMDisTheBEST

Senior member
Dec 17, 2015
682
90
61
Now, AMD delivers us a CPU that has Broadwell IPC (or even slightly better) in many tasks to which it is well suited and its worst-case tasks are around Haswell level (and, yes, with a few lower outliers).

We should be singing AMD's praises for accomplishing that and then delivering it to us for as little as $329... and not playing any of the gimping nonsense of which Intel is so fond.
Isn't broadwell just a die shrink of the haswell tho? Skylake is the actual Intel tock in IPC and that is inherited by kabylake i7 7700k. We need the next iteration of ryzen next year for amd to completely catch up.(assuming coffee lake is just another optimization instead of either tick or tock) I decide to wait until next year for a ryzen upgrade.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Just took the liberty of using AT slides to showcase what ZEN could do in servers

TDP ain't power usage. We know you like AMD but please keep it real.

Yes, Ryzen at lower clocks seems to have great performance/dollar. But Intel server chips, eg xeons also run at lower clocks, at least some models. Intels TDP is so high due to AVX2 workloads. However as we have seen actually power use without AVX2 is pretty similar between the 2 chips.

Still, yes I agree it looks great for Naples in this regard, but Naples isn't out yet and Skylake-X will be soonish. 14nm+ from Intel isn't bad at all for server/mobile use. Yes 7700k semi-rebrand wasn't great but laptop chips had pretty huge increases due to the new process.

Price of CPU however is somewhat irrelevant for TCO-cost of a server. Matters for us yes, but not much for server. I rather have AMD charge more for server chips and make more profit for future R&D. Also don't forget to count software cost to TCO. If you run software that costs you several 100k a year, even 5% better performance matters more even if the chips cost $1000 more.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
That's the beauty of base clock overclocking - the CPU won't trigger OC mode. Though I don't expect to keep much of an overclock myself. I was hoping for less performance than the 1700X offers stock, so I won't be upset no matter what :p
In that case it seems I'll have to loosen my no-tinkering policy a bit. :p
 
  • Like
Reactions: looncraz

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
It doesn't prove much though as with 2+2 the data flow between the 2CCX would be heavier than with 4+4.

Awesome they tested that. Actually I expected the difference between 2+2 and 4+0 to be a lot bigger. This shows that if windows scheduling is fixed, what gains are possible. 5% on average.

Also I don't think 2+2 has more traffic between CCX than 4x4. Rather less. Assuming the scheduler is random (which probably isn't true) chance of moving a thread to the other CCX is exactly 50%. In the 4x4 case you then have double the amount of cores that needs to access L3 of the other CCX. So if that is actually a bottleneck of throughput and not latency, then 4x4 could profit more than 2x2.

However it is a mystery how they will market these quads? all 4x0 as top sku and 2x2 as lower skus?
 

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
No clue how credible this is.

Combination of MB & Windows Power profile.

The FireStrike result with just a change in Windows Power profile is ridiculous. Why? Similar CPU Physics scores, but much higher overall GPU scores. There seems to be a problem with Windows for Ryzen <-> GPU communication. o_O

Now it start to make sense. Ryzen does super in professional workloads vs the 6900K. But it under performs relative to the 6900K in gaming. The difference with gaming? GPU communication comes into play. Pro work is all CPU only.

https://www.reddit.com/r/Amd/comments/5xdxvy/the_overclocker_magazine_that_moment_when_your/


Ryzen under the right power profile is way faster than 7700K at gaming.
Power states are the key

Nope, nope nope, 1700X faster. Issue us updating CPU uCode and not being able to go back, but that's another story

This seems to be a teething issue and will be quickly sorted by board vendors and AMD. It's really fast, even at 2133mhz it'll still beat 3200mhz 7700K


https://www.facebook.com/theoverclocker/photos/rpp.72682703453/10155037868423454/?type=3&theater
 
Last edited:
  • Like
Reactions: looncraz

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
I am afraid those are system-wide wattage with a GPU and we cannot tell how much of it is GPU and how much of it is CPU. :)

Exactly and if the GPU is CPU limited it will not run at 100% and hence use less power than with a 7700k.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Isn't broadwell just a die shrink of the haswell tho? Skylake is the actual Intel tock in IPC and that is inherited by kabylake i7 7700k. We need the next iteration of ryzen next year for amd to completely catch up.(assuming coffee lake is just another optimization instead of either tick or tock) I decide to wait until next year for a ryzen upgrade.

Broadwell has 3~5% higher IPC than Haswell and some minor tweaks.

Broadwell-E, however, is what I really meant by Broadwell - where the extra caches and other benefits take hold. Then we can see as much as a 10% difference over Haswell at the same clock, per thread.
 
  • Like
Reactions: Drazick