Ryzen: Strictly technical

Page 48 - 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.

Elixer

Lifer
May 7, 2002
10,376
762
126
Laptop first makes more sense from OEM perspective. And DIY folk en masse skips on APUs as of late.
Didn't they have a few laptop companies come and speak on the polaris launch for "design wins"?
I am assuming that they will have Ryzen + polaris SKUs available before back to school in the fall, which they can't miss.

That would really be interesting looking at the design of those chips, and seeing how much latency those have compared to past APUs.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
I just disassembled the FMA3 code, which caused the Ryzen crashes:
Code:
.text:0000000140006f90 0x140006f90:
.text:0000000140006f90 vfmadd231ps xmm4, xmm0, xmm1
.text:0000000140006f95 vfmadd231ps xmm5, xmm0, xmm1
.text:0000000140006f9a vfmadd231ps xmm6, xmm0, xmm1
.text:0000000140006f9f vfmadd231ps xmm7, xmm0, xmm1
.text:0000000140006fa4 vfmadd231ps xmm8, xmm0, xmm1
.text:0000000140006fa9 vfmadd231ps xmm9, xmm0, xmm1
.text:0000000140006fae vfmadd231ps xmm10, xmm0, xmm1
.text:0000000140006fb3 vfmadd231ps xmm11, xmm0, xmm1
.text:0000000140006fb8 vfmadd231ps xmm12, xmm0, xmm1
.text:0000000140006fbd vfmadd231ps xmm13, xmm0, xmm1
.text:0000000140006fc2 vfmadd231ps xmm14, xmm0, xmm1
.text:0000000140006fc7 vfmadd231ps xmm15, xmm0, xmm1
.text:0000000140006fcc vfmsub231ps xmm4, xmm0, xmm1
.text:0000000140006fd1 vfmsub231ps xmm5, xmm0, xmm1
.text:0000000140006fd6 vfmsub231ps xmm6, xmm0, xmm1
.text:0000000140006fdb vfmsub231ps xmm7, xmm0, xmm1
.text:0000000140006fe0 vfmsub231ps xmm8, xmm0, xmm1
.text:0000000140006fe5 vfmsub231ps xmm9, xmm0, xmm1
.text:0000000140006fea vfmsub231ps xmm10, xmm0, xmm1
.text:0000000140006fef vfmsub231ps xmm11, xmm0, xmm1
.text:0000000140006ff4 vfmsub231ps xmm12, xmm0, xmm1
.text:0000000140006ff9 vfmsub231ps xmm13, xmm0, xmm1
.text:0000000140006ffe vfmsub231ps xmm14, xmm0, xmm1
.text:0000000140007003 vfmsub231ps xmm15, xmm0, xmm1
.text:0000000140007008 vfmadd231ps xmm4, xmm0, xmm1
.text:000000014000700d vfmadd231ps xmm5, xmm0, xmm1
.text:0000000140007012 vfmadd231ps xmm6, xmm0, xmm1
.text:0000000140007017 vfmadd231ps xmm7, xmm0, xmm1
.text:000000014000701c vfmadd231ps xmm8, xmm0, xmm1
.text:0000000140007021 vfmadd231ps xmm9, xmm0, xmm1
.text:0000000140007026 vfmadd231ps xmm10, xmm0, xmm1
.text:000000014000702b vfmadd231ps xmm11, xmm0, xmm1
.text:0000000140007030 vfmadd231ps xmm12, xmm0, xmm1
.text:0000000140007035 vfmadd231ps xmm13, xmm0, xmm1
.text:000000014000703a vfmadd231ps xmm14, xmm0, xmm1
.text:000000014000703f vfmadd231ps xmm15, xmm0, xmm1
.text:0000000140007044 vfmsub231ps xmm4, xmm0, xmm1
.text:0000000140007049 vfmsub231ps xmm5, xmm0, xmm1
.text:000000014000704e vfmsub231ps xmm6, xmm0, xmm1
.text:0000000140007053 vfmsub231ps xmm7, xmm0, xmm1
.text:0000000140007058 vfmsub231ps xmm8, xmm0, xmm1
.text:000000014000705d vfmsub231ps xmm9, xmm0, xmm1
.text:0000000140007062 vfmsub231ps xmm10, xmm0, xmm1
.text:0000000140007067 vfmsub231ps xmm11, xmm0, xmm1
.text:000000014000706c vfmsub231ps xmm12, xmm0, xmm1
.text:0000000140007071 vfmsub231ps xmm13, xmm0, xmm1
.text:0000000140007076 vfmsub231ps xmm14, xmm0, xmm1
.text:000000014000707b vfmsub231ps xmm15, xmm0, xmm1
.text:0000000140007080 sub rax, 0x1
.text:0000000140007084 jne 0x140006f90
C7ugF0fW4AEHkI1.jpg
So be aware, that your compiler doesn't accidentally create such code from your sources. ;) Well, this won't happen.

As you can see, it doesn't do useful calculations. It repeatedly multiplies xmm0 and xmm1 and adds, later subtracts that result from different registers:
a:=b*c+a
later
a:=b*c-a
which means: b*c is actually not relevant.

Edit: The code loops 16 billion times.
 
Last edited:

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96

I am no longer in possession of any samples but can anyone validate the scores and performance. If this is validated when the X300 arrives I will not be spending money on a fancy motherboard but a lot of money on a 3400+ kit of RAM.
 
  • Like
Reactions: lightmanek and bjt2

CatMerc

Golden Member
Jul 16, 2016
1,114
1,149
136

I am no longer in possession of any samples but can anyone validate the scores and performance. If this is validated when the X300 arrives I will not be spending money on a fancy motherboard but a lot of money on a 3400+ kit of RAM.
It looks like time for retesting ;)
 

OrangeKhrush

Senior member
Feb 11, 2017
220
343
96
It looks like time for retesting ;)

I think Ryzen needs to be tested with DDR4 3400-3600 and 4000, AMD utilizes bandwidth more than Intel, if it turns out correct that is massive gains, anywhere from 10-25FPS is massive difference. Sure people will say but the budget gamer can't pay that for RAM, but in all honesty, what budget gamer will roll a R7 and 1080ti anyway. I think a 170 dollar 1400 with a kit of 3400 RAM will cost less than a 1700X add a stable B350 motherboard and you have potentially a king of the value segment offering.
 
  • Like
Reactions: lightmanek

DisEnchantment

Golden Member
Mar 3, 2017
1,590
5,722
136
I think Ryzen needs to be tested with DDR4 3400-3600 and 4000, AMD utilizes bandwidth more than Intel, if it turns out correct that is massive gains, anywhere from 10-25FPS is massive difference. Sure people will say but the budget gamer can't pay that for RAM, but in all honesty, what budget gamer will roll a R7 and 1080ti anyway. I think a 170 dollar 1400 with a kit of 3400 RAM will cost less than a 1700X add a stable B350 motherboard and you have potentially a king of the value segment offering.
If we go by the reviews every budget gamer should have SLI 1080Ti/Titan XP with 240 Hz monitor at 1080p.
 

lolfail9001

Golden Member
Sep 9, 2016
1,056
353
96
I am no longer in possession of any samples but can anyone validate the scores and performance. If this is validated when the X300 arrives I will not be spending money on a fancy motherboard but a lot of money on a 3400+ kit of RAM.
Why would not you do it in any case? ITX builds are glorious for how tight they are with circuitry.
I think Ryzen needs to be tested with DDR4 3400-3600 and 4000
You need a freaking chilled loop to get ddr4 4000 working right now, and well ddr4 3400-3600 are hardly stable on anything but cream of the crop stuff.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
Having now spent a couple days running my main rig with Ryzen 1700X @ 3.9Ghz I have noticed a few interesting improvements over an i7-2600k @ 4.5Ghz.

For example, I used to always have a stutter when moving a Firefox tab with a video playing to the second screen. That is completely gone. Same profile, same Firefox, same drivers, same Windows installation... just completely gone.

In fact, I've noticed this in probably a dozen of little places - the system just seems absurdly smooth. It's amazing how much I became accustomed to those little stumbles and hiccups to the point I didn't even notice them.

On a more interesting note - I have examined my CCX benchmark code and behavior quite thoroughly - the peak low CCX latency is, indeed, just 20ns, but the average is quite a bit higher (100ns). I have much more work to do on the benchmark before the results will be clean..

At the moment, I start 16 threads, 8 writers, 8 readers, and I set each to an affinity on a specific physical core and each with the highest priority I can manage on Linux... then the readers begin spinning on an atomic_int to have the value "2" which indicates that a writer has written a 64-bit TSC value into its data structure as well as modifying an integer to denote which core did the writing, the reader then calculates the time delta (in TSC cycles) and inserts the result in the list of results for the specific writer.

This has the problem that each core is running both reading and writing threads, so some of the results are dramatically high as the two heavy threads are switched back and forth, naturally, so I will be focusing on addressing that minor issue now that the basic test is working.
 

Blake_86

Junior Member
Mar 13, 2017
21
3
36
I think Ryzen needs to be tested with DDR4 3400-3600 and 4000, AMD utilizes bandwidth more than Intel, if it turns out correct that is massive gains, anywhere from 10-25FPS is massive difference. Sure people will say but the budget gamer can't pay that for RAM, but in all honesty, what budget gamer will roll a R7 and 1080ti anyway. I think a 170 dollar 1400 with a kit of 3400 RAM will cost less than a 1700X add a stable B350 motherboard and you have potentially a king of the value segment offering.
Yea, sadly, ryzen ram support is scandalous. I bought time ago 2 x16 tridend Z cas 15 3200 mhz, thinking to buy a good kit, only to discover today that this f**** cpu is not compatible whit dual rank memory. For me it s pure madness.
 

KompuKare

Golden Member
Jul 28, 2009
1,012
923
136
Yea, sadly, ryzen ram support is scandalous. I bought time ago 2 x16 tridend Z cas 15 3200 mhz, thinking to buy a good kit, only to discover today that this f**** cpu is not compatible whit dual rank memory. For me it s pure madness.
I though it didn't like dual rank which is not the same as being incompatible.
Rather if you use dual rank it has to run slower
AMD%20Ryzen%207%20Press%20Deck-18_575px.jpg

Someone was using Dual rank with their 1800X but they only got it to 2666:
https://forums.anandtech.com/threads/ddr4-rank-for-ryzen-help-me-understand.2500688/#post-38777117
 

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
I though it didn't like dual rank which is not the same as being incompatible.
Rather if you use dual rank it has to run slower
Someone was using Dual rank with their 1800X but they only got it to 2666:
https://forums.anandtech.com/threads/ddr4-rank-for-ryzen-help-me-understand.2500688/#post-38777117
Is dual rank memory being slower an update bios issue, or will it always be slower?

I heard that single rank has better OC'ing potential, but dual rank should at least be able to run at spec after some updates, right? At least within reason, like 3200 MHz, anyways.
 
Last edited:

Blake_86

Junior Member
Mar 13, 2017
21
3
36
I though it didn't like dual rank which is not the same as being incompatible.
Rather if you use dual rank it has to run slower
AMD%20Ryzen%207%20Press%20Deck-18_575px.jpg

Someone was using Dual rank with their 1800X but they only got it to 2666:
https://forums.anandtech.com/threads/ddr4-rank-for-ryzen-help-me-understand.2500688/#post-38777117
I cant see why a pair of ram should run slower, a lot slower, ther their spec. The same dual rank kit is fully compatible whit intel and also have no problem to run higher!
This is a duble fault by amd, creating a heavy ram dependant architecture whit a crappy ram compatibility.
I realy would to buy amd, but if it is true that is frustrating to go intel alwais and alwais quad core, it also frustrating to buy an 8 core and have to chose from perf and 32 gb of ram
 

Blake_86

Junior Member
Mar 13, 2017
21
3
36
Is dual rank memory being slower an update bios issue, or will it always be slower?

I heard that single rank has better OC'ing potential, but dual rank should at least be able to run at spec after some updates, right? At least within reason, like 3200 MHz, anyways.
The situatuon whil not change, maybe when in may amd relese some new microcode, but i have not so much hope.
After wr have to wait that the code will be dropped in newer bios relase..
 

Ozzyrulez

Junior Member
Mar 21, 2017
16
45
61
I cant see why a pair of ram should run slower, a lot slower, ther their spec. The same dual rank kit is fully compatible whit intel and also have no problem to run higher!
This is a duble fault by amd, creating a heavy ram dependant architecture whit a crappy ram compatibility.
I realy would to buy amd, but if it is true that is frustrating to go intel alwais and alwais quad core, it also frustrating to buy an 8 core and have to chose from perf and 32 gb of ram

Calm down man, it's a brand new chip. Give it some time. Slowly but surely the RAM support is getting better. It's not like AMD was just like "hey yeah let's release this and make dual rank RAM not work full speed on purpose".
 

Blake_86

Junior Member
Mar 13, 2017
21
3
36
Calm down man, it's a brand new chip. Give it some time. Slowly but surely the RAM support is getting better. It's not like AMD was just like "hey yeah let's release this and make dual rank RAM not work full speed on purpose".
The fact they made slide to inform people about this leave me with a bad feeling...
 

Grep_Linux

Junior Member
Mar 15, 2017
16
10
41
Having now spent a couple days running my main rig with Ryzen 1700X @ 3.9Ghz I have noticed a few interesting improvements over an i7-2600k @ 4.5Ghz.

For example, I used to always have a stutter when moving a Firefox tab with a video playing to the second screen. That is completely gone. Same profile, same Firefox, same drivers, same Windows installation... just completely gone.

In fact, I've noticed this in probably a dozen of little places - the system just seems absurdly smooth. It's amazing how much I became accustomed to those little stumbles and hiccups to the point I didn't even notice them...

While I'm waiting for a bracket from Arctic to finish my RyZen rig I still read these forums rabidly absorbing all the information I can. This has been a known benefit more cores for a long time, my FX8350 has been a real smooth desktop experience, it isn't a power gamer but more threads really helps with the little things. It's amazing how this doesn't get mentioned much because people focus on singular points all the time and make judgments based on them. In fact when I had a overclocked celeron at 4.2 ghz a 2 ghz two core played video better (anything 720p and above) because it'd choke the single core and time slicing with too many other tasks really made it hard to get smooth playback.
 
  • Like
Reactions: looncraz

looncraz

Senior member
Sep 12, 2011
722
1,651
136
While I'm waiting for a bracket from Arctic to finish my RyZen rig I still read these forums rabidly absorbing all the information I can. This has been a known benefit more cores for a long time, my FX8350 has been a real smooth desktop experience, it isn't a power gamer but more threads really helps with the little things. It's amazing how this doesn't get mentioned much because people focus on singular points all the time and make judgments based on them. In fact when I had a overclocked celeron at 4.2 ghz a 2 ghz two core played video better (anything 720p and above) because it'd choke the single core and time slicing with too many other tasks really made it hard to get smooth playback.

Well, that's the thing - I have aggressive core parking enabled and a 1.6GHz idle clock (custom P-states, 3.95GHz, 3.9Ghz, 1.6GHz).

I've found I get the best results by letting Windows handle the P-states and modifying the parking interval from 30ms to 15ms (and a few other minor changes to the power configuration - I'm working out what will be the best configuration for performance).

I do, however, always have two cores unparked, along with their SMT threads, so I guess there's always more threads available than with an i7 quad core.
 

looncraz

Senior member
Sep 12, 2011
722
1,651
136
I cant see why a pair of ram should run slower, a lot slower, ther their spec. The same dual rank kit is fully compatible whit intel and also have no problem to run higher!
This is a duble fault by amd, creating a heavy ram dependant architecture whit a crappy ram compatibility.
I realy would to buy amd, but if it is true that is frustrating to go intel alwais and alwais quad core, it also frustrating to buy an 8 core and have to chose from perf and 32 gb of ram

The advantage of being Intel is that you get to dictate how everything works to suit your own design failures. It's not that Intel has a better IMC as much as practically all the memory out there is designed to run on Intel's DDR4 IMC - all of their sub-timings are set appropriately.

AMD's IMC has been pushed to 3600MT/s.

Interestingly, I'm having the same results with dual rank memory as with single rank - I can't get past the 2667 multiplier no matter what, but I can overclock the reference clock and reach 2800. Both kits achieve the same timings and speeds with the same voltage. I still have more testing to do, but I have noticed that the ASUS Crosshair VI Hero runs CR2 for both kits with the 2667 multiplier.
 

SuperFist

Junior Member
Mar 25, 2017
2
1
16
Copy dism.exe, boot.wim & install.wim (sources directory) files from the Windows 7 ISO image to your hard-drive.
Download the USB driver for Ryzen: https://1drv.ms/u/s!Ag6oE4SOsCmDhGIQJdHdaXC-_w-C
Extract the package to the same directory as DISM is located.
Enter the folders containing the individual driver files and check that there is not "Unblock" button visible. If there is, you need to manually toggle it for each and every file.

- "DISM /mount-wim /wimfile:boot.wim /index:2 /mountdir:x:\xxx" (x:\xxx = a temporary path of your choice, make sure to have ~20GB of space available for install.wim).
- "DISM /image:x:\xxx /add-driver /driver:Ryzen_USB_W764\ /recurse /forceunsigned"
- "DISM /unmount-wim /mountdir:x:\xxx /commit"

Windows 7 install.wim files contain four different OS variants, regardless of the officially stated edition of the media you have (Home, Professional, Ultimate).
The index order within the install.wim is always the same, regardless of the edition: Index 1 = Home Basic, Index 2 = Home Premium, Index 3 = Professional, Index 4 = Ultimate.

So if your media is for Professional edition, you need to make the changes to Index 3. If it is a Ultimate media, then make the changes to Index 4, etc.

- "DISM /mount-wim /wimfile:install.wim /index:x /mountdir:x:\xxx" (x:\xxx = a temporary path of your choice, make sure to have ~20GB of space available for install.wim).
- "DISM /image:x:\xxx /add-driver /driver:Ryzen_USB_W764\ /recurse /forceunsigned"
- "DISM /unmount-wim /mountdir:x:\xxx /commit"

After you have added the drivers to both of the WIMs, you can install Windows 7 from a USB drive and using USB keyboard and mouse.
Make sure that you don't use USB ports provided by a 3rd party manufacturer (other than ASMedia), as there are still no drivers for those in the media.
After the installation, install the Relive chipset driver pack for Windows 7 (available at AMD.com).
I cannot get this to work for the life of me. What mobo do you have? I have followed the guide with similar DISM instructions here, but I just can't get this to work for my Gigabyte GA-AX370-Gaming K7 mobo and Samsung EVO 250GB M.2 NVMe SSD:

http://forum.notebookreview.com/thr...an-nvme-ssd-from-a-usb-3-0-thumbdrive.783921/

I understand I'm building a desktop PC and the above is a website for notebooks and laptops, but there shouldn't be that much difference. I can get to the Windows 7 install setup, but the USB keyboard does not work no matter which drivers I slipstream into the Windows 7 Professional ISO I have on my USB 3.0 flashdrive. And I've grabbed them from my Gigabyte mobo website, general AMD USB 3.0 drivers, etc. It's been 2 full days since I ran into this snag and I've even resorted to ordering a PS/2 keyboard from Amazon Prime Same Day delivery because I refuse to let this consume my weekend!
 
Last edited:

unseenmorbidity

Golden Member
Nov 27, 2016
1,395
967
96
The advantage of being Intel is that you get to dictate how everything works to suit your own design failures. It's not that Intel has a better IMC as much as practically all the memory out there is designed to run on Intel's DDR4 IMC - all of their sub-timings are set appropriately.

AMD's IMC has been pushed to 3600MT/s.

Interestingly, I'm having the same results with dual rank memory as with single rank - I can't get past the 2667 multiplier no matter what, but I can overclock the reference clock and reach 2800. Both kits achieve the same timings and speeds with the same voltage. I still have more testing to do, but I have noticed that the ASUS Crosshair VI Hero runs CR2 for both kits with the 2667 multiplier.
Linus, of LTT youtube, was having issues getting anything above 2666 too. He even had that new flare memory, and he still couldn't get it past 2666 on the Crosshair board.
 
  • Like
Reactions: looncraz

Artorius

Junior Member
Jun 5, 2016
7
0
6
Linus, of LTT youtube, was having issues getting anything above 2666 too. He even had that new flare memory, and he still couldn't get it past 2666 on the Crosshair board.
It's weird because another youtuber (JayzTwoCents, I think?) just posted a video saying that after a recent BIOS update he could do it on the same board. Linus didn't specify which BIOS he was using at that moment so there's no way to know if the problem was only that or something else.
 
Status
Not open for further replies.