Ryzen: Strictly technical

Page 15 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

MajinCry

Platinum Member
Jul 28, 2015
2,495
571
136
Been a while since I posted here, but 'ey.

@The Stilt Would you be able to test Ryzen, on Win 7, with SMT disabled? One user had it disabled on Windows 10 @ 4.0ghz, and it scored even better than your Win 7 results.

1_16000.png
 

campdude

Guest
Jan 14, 2012
1
0
71
I seen other people posting this as well. Coreinfo says that FX 8350 is hyperthreaded.
It must be wrong lol


Logical to Physical Processor Map:
**------ Physical Processor 0 (Hyperthreaded)
--**---- Physical Processor 1 (Hyperthreaded)
----**-- Physical Processor 2 (Hyperthreaded)
------** Physical Processor 3 (Hyperthreaded)

******** Group 0
 
Last edited:

Rngwn

Member
Dec 17, 2015
143
24
36
Take that result with a grain of salt for now. I have had a look at the comment section and the reviewer said that He forgot to disable the "double buffering" on the 7700k part. Now that it is disabled, the result is here:

http://www.zolkorn.com/news/report-an-error-for-ryzen-7-1800x-vs-core-i7-7700k-mhz-by-mhz/

The clock vs. clock performance is now what it is supposed to be. "Slightly" inferior to Kabylake parts as expected.
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
May I suggest that it is best to take that result to the designated thread (here) and leave this thread for inquiries related to The Stilt's work? Not to be obtuse about it but I do think this thread deserves better than the "VS" talks that are plentiful elsewhere on this message board.
 
  • Like
Reactions: WiseUp216

imported_jjj

Senior member
Feb 14, 2009
660
430
136
I have an ... interesting request.
Can you disable as much I/O as possible on the chipset (while using the I/O from the CPU) and try to quantify what power savings it might generate?
With an undervolted CPU, might as well try to push further.
Would also give us a better idea on the power savings X300 might get us.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
I have an ... interesting request.
Can you disable as much I/O as possible on the chipset (while using the I/O from the CPU) and try to quantify what power savings it might generate?
With an undervolted CPU, might as well try to push further.
Would also give us a better idea on the power savings X300 might get us.

The PCH (platform controller hub - a.k.a. the "chipset") only uses some 6~8W of power total.

12045156519l.jpg
 

imported_jjj

Senior member
Feb 14, 2009
660
430
136
The PCH (platform controller hub - a.k.a. the "chipset") only uses some 6~8W of power total.

12045156519l.jpg


I was remembering 5W (guess X370 is quite a bit more if B350 is 5.8W) but even 1-2W in savings would be relevant for an undervolted 1700 at low and mild load. At full load in certain tasks or gaming with a GTX 1060 it could be 1% savings.
Ofc if the chipset is properly designed, it would not use much power at all when nothing is connected to it even if not disabled but i thought it's worth checking.
At idle normal use Windows so with a bunch of things open and .let's say average 5% load, an 1800X uses maybe 15W- just the CPU. Web browsing is maybe 5-10W above that.
An undervolted 1700 would do even better so if disabling X370 I/O or using X300 can save 1-2W, seems fun to me.
 

iBoMbY

Member
Nov 23, 2016
175
103
86
I seen other people posting this as well. Coreinfo says that FX 8350 is hyperthreaded.
It must be wrong lol


Logical to Physical Processor Map:
**------ Physical Processor 0 (Hyperthreaded)
--**---- Physical Processor 1 (Hyperthreaded)
----**-- Physical Processor 2 (Hyperthreaded)
------** Physical Processor 3 (Hyperthreaded)

******** Group 0

Two cores share one FPU. For the performance it's better to assume they work like SMT cores, than to ignore the fact.
 
  • Like
Reactions: looncraz

Timur Born

Senior member
Feb 14, 2016
277
139
116
Thanks for the work and sharing!

Some questions:

1) Did anyone try to setup their own Windows power profile yet, or disable it completely for testing?

2) Is there a clear trend yet as to which workloads Ryzen prefers over others? Does it do better at integer than at floating-point? If so, is it a general trend or just true for specific scenarios?

3) In the Mhz:Mhz article someone posted earlier I noticed that "heavy multitasking" workload was tested remarkably worse than what the 7700K delivered. Did anyone run a bunch of virtual machines on a 8/16 Ryzen yet, especially in comparison to Intel parts?
 
Last edited:

riggnix

Junior Member
Jul 27, 2016
23
3
41
First of all: Thank you. Great review. Really appreciated.

So changing the voltage and the BCLK do not enable OC mode, right? It would be really interesting to get some numbers about max BCLK for PCIe 2.0.
That should enable some descent performance upgrades for the 1800X and explains, why the 1700X is that much cheaper.

I guess the OC mode is not related to XFR, but present on all chips (e.g. 1700), right?
 

CrazyElf

Member
May 28, 2013
88
21
81
Verify in what respect?
Sure, you can hit those BCLKs if you set the PCIe to run in Gen. 1 mode.

So basically what you are saying is that overclocks cause it to fall into Gen 2 or Gen 1 automatically?

Also, why is the Uncore different? I thought the Uncore was at the same speed as the Core itself? From what I understand, cache speed operates at the highest core speed. So in this case, the CPU was idling and the highest cache was at 1.8 GHz, rather than say, the maximum clock speed?


wow that Ryzen tested as quad core at 4 ghz against a 7700k at 4 ghz is quite good. In fact the inter CCX bandwidth problem is automatically solved by having a single CCX with all 4/8T. I think if AMD can use a single CCX for R5 1500x with clocks of 3.9-4.0 Ghz for 1-2 cores and 3.7 Ghz for all core turbo then it would be a very good gaming chip too especially if AMD can price it at USD 179 (instead of USD 199). It will basically be very close to core i5 7600k but at a significantly lower price. In fact Intel's non k Skylake chips would have a tough time against it.

If everything is on one CCX, then the last level cache becomes that CCX's L3 and not the DRAM, unlike with 2 CCX enabled.
 
Last edited:

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
So basically what you are saying is that overclocks cause it to fall into Gen 2 or Gen 1 automatically?

Also, why is the Uncore different? I thought the Uncore was at the same speed as the Core itself? From what I understand, cache speed operates at the highest core speed. So in this case, the CPU was idling and the highest cache was at 1.8 GHz, rather than say, the maximum clock speed?




If everything is on one CCX, then the last level cache becomes that CCX's L3 and not the DRAM, unlike with 2 CCX enabled.

No, you need to manually place the PCIe from Gen.3 (default) to Gen.2.

In your example (CPU-Z), the CPU + L1/L2/L3 operated at ~4055MHz, while the actual data fabric operated at 1800MHz.

L1/L2/L3 = CPU frequency.
Data fabric, memory controller = MEMCLK (effective) / 2.
 

CrazyElf

Member
May 28, 2013
88
21
81
Thanks that makes a lot more sense.

So:
  1. Data fabric at 50% of RAM speed
  2. Uncore is tied to core speed
  3. Base clock (or Refclock) is tied to PCIe , and can only go to about 105 MHz at 3.0, but then has to go to 2.0
Hmm ... that makes me wonder about the data fabric speed.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Thanks that makes a lot more sense.

So:
  1. Data fabric at 50% of RAM speed
  2. Uncore is tied to core speed
  3. Base clock (or Refclock) is tied to PCIe , and can only go to about 105 MHz at 3.0, but then has to go to 2.0
Hmm ... that makes me wonder about the data fabric speed.

There is no "uncore" in Ryzen.
CCX essentially operates at it's own (core speed) and the fabrics and the memory controller at half the effective MEMCLK speed.
Uncore is just the term used by CPU-Z for data fabric (DFICLK).
 

gupsterg

Junior Member
Mar 4, 2017
9
3
51
No, you need to manually place the PCIe from Gen.3 (default) to Gen.2.

From reading Elmor's guide it happens automatically on CH6 or have I got wrong end of stick?

The above values are based on the Auto-rules and are a good indication of the margins. PCI Express bandwidth values are at default 100MHz REFCLK and increases linearly when increased. From internal testing most PCI Express devices including graphics cards and storage controllers handle increased REFCLK very well. At the moment there’s no list with proven devices.

Has your experience been different on CH6
 

Pookums

Member
Mar 6, 2017
32
13
36
Stilt, since we cant raise or alter certain clock rates directly, assuming firmware could be presented to lie to the UCLK or fabric, would that be an adequate solution to their limited variability? Either by bifurcating information to the system agent reading the MEMCLK, or by creating a secondary path which is only read by the fabric and UCLK? Fixing the issue of control by Breaking traditional standards so to speak?

For example: lets say you had 3200mhz ram. This would normally limit UCLK and fabric rates to 1600mhz. With alterations, the firmware was then made to lie to the fabric, but the normal system still treated the memory as the correct 3200mhz rate. The Fabric/UCLK are meanwhile sent incorrect information modified by a multiplier(say in this instance 2x or 6400mhz False MEMCLK) resulting in the UCLK and fabric reaching parity with the True MEMCLK at a 1:1 ratio.

If the user cannot access these Rates directly on the motherboard this seems like an amenable solution to the problem. Testing would still need to be concluded to determine maximums, safe 24/7 rates and voltages. Has something like this ever been done before in the past? At the moment this seems like the only possible solution to the end user if it works.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Stilt, since we cant raise or alter certain clock rates directly, assuming firmware could be presented to lie to the UCLK or fabric, would that be an adequate solution to their limited variability? Either by bifurcating information to the system agent reading the MEMCLK, or by creating a secondary path which is only read by the fabric and UCLK? Fixing the issue of control by Breaking traditional standards so to speak?

For example: lets say you had 3200mhz ram. This would normally limit UCLK and fabric rates to 1600mhz. With alterations, the firmware was then made to lie to the fabric, but the normal system still treated the memory as the correct 3200mhz rate. The Fabric/UCLK are meanwhile sent incorrect information modified by a multiplier(say in this instance 2x or 6400mhz False MEMCLK) resulting in the UCLK and fabric reaching parity with the True MEMCLK at a 1:1 ratio.

If the user cannot access these Rates directly on the motherboard this seems like an amenable solution to the problem. Testing would still need to be concluded to determine maximums, safe 24/7 rates and voltages. Has something like this ever been done before in the past? At the moment this seems like the only possible solution to the end user if it works.

The software / firmware can only alter things that are supported by the hardware itself.
I don't think running the fabrics at different speeds is possible, besides the UMC which is confirmed to support also 1:1 ratio during debug.

Changing these would most likely require changes to the actual silicon design.
 

alcal

Junior Member
Mar 7, 2017
1
1
36
Hey Stilt, fellow Finn here. Thanks for the awesome post. I learned much more in these 15 pages than from the other tech sites combined.

I was wondering about the feasibility of a mild BCLK overclock with the following goals in mind:
  • Not entering OC mode, thus not forcing the CPU to run in P0 state
  • Slightly boosting memory and interconnect speeds
  • Increasing the max range of XFR, boosting single thread performance without sacrificing the ability to clock down when idle.
From my understanding of your discussion on OC mode, it isn't triggered by BCLK or Vcore tweaks--only multiplier. If that is correct, would we be able to extend XFR's max clock? I would assume it isn't hard capped at 4.1 on the 1800x for example, and is rather a byproduct of (BCLK)*(Multipliers)+(XFR offset).

Furthermore, if OC mode isn't triggered by BCLK adjustments and as you said earlier, 107 is the highest you can go while maintaining PCIe gen 3 stability, would it make sense to just use a BCLK of 102 for example (I think the default is 100 right?) to get a 3672mhz base / 4080mhz boost clock where one core can boost up to 4180mhz under XFR? This is of course assuming you could get the system stable at those clocks.

Sorry if its a dumb question, but I figured it would be good to know if there is any way to make XFR work with us instead of just ignoring it as most people seem to be doing.
 
Last edited:
  • Like
Reactions: lightmanek

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Hey Stilt, fellow Finn here. Thanks for the awesome post. I learned much more in these 15 pages than from the other tech sites combined.

I was wondering about the feasibility of a mild BCLK overclock with the following goals in mind:
  • Not entering OC mode, thus not forcing the CPU to run in P0 state
  • Slightly boosting memory and interconnect speeds
  • Increasing the max range of XFR, boosting single thread performance without sacrificing the ability to clock down when idle.
From my understanding of your discussion on OC mode, it isn't triggered by BCLK or Vcore tweaks--only multiplier. If that is correct, would we be able to extend XFR's max clock? I would assume it isn't hard capped at 4.1 on the 1800x for example, and is rather a byproduct of (BCLK)*(Multipliers)+(XFR offset).

Furthermore, if OC mode isn't triggered by BCLK adjustments and as you said earlier, 107 is the highest you can go while maintaining PCIe gen 3 stability, would it make sense to just use a BCLK of 102 for example (I think the default is 100 right?) to get a 3672mhz base / 4080mhz boost clock where one core can boost up to 4180mhz under XFR? This is of course assuming you could get the system stable at those clocks.

Yes, "OC-Mode" is only activated by increasing the P0 PState effective ratio. By using BCLK you can somewhat extend the frequencies, while maintaining the normal CPB & XFR operation.

However as said before, I personally do not recommend increasing the BCLK even by a single MHz. It's not the way these chips were designed to operate and increasing the BCLK even slightly can cause issues.
 

PPB

Golden Member
Jul 5, 2013
1,118
168
106
Yes, "OC-Mode" is only activated by increasing the P0 PState effective ratio. By using BCLK you can somewhat extend the frequencies, while maintaining the normal CPB & XFR operation.

However as said before, I personally do not recommend increasing the BCLK even by a single MHz. It's not the way these chips were designed to operate and increasing the BCLK even slightly can cause issues.

And would using Zen master profiles to avoid unwanted high idle power consumption be a good workaround to have OC when you need it and good power consumption when you arent running specific stressful apps?
 
Status
Not open for further replies.