• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

AMD Ryzen 3000 Builders Thread

Page 124 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

bononos

Diamond Member
Aug 21, 2011
3,718
68
91
Yeap, in my routines, my FX8350 would finished up a routine in 19 minutes. This Athlon 3000G does it in 15 minutes. 26.7% time saved. Equates to a few hours total. It's a big deal for me. I know a Ryzen 1600 does it in 6 minutes, same file. So if the 5900x can drop this to under 4 minutes, it will be a massive time saver for me. So I'm not losing anything frankly going from FX8350 to the Athlon 3000G which is kind of sad, but 8 year tech gap makes a big impact. Just waiting on that 5900x, maybe in 6 months sigh.

Very best,
Are you still using your old AM3+ cooler or are you keeping it? It might be interesting to see how it works with a new Ryzen compared to a more current cooler. The newer cpus have a higher heat density and might do better with newer coolers that take that into account.
 
  • Like
Reactions: Tlh97 and MalVeauX

MalVeauX

Senior member
Dec 19, 2008
592
135
116
Are you still using your old AM3+ cooler or are you keeping it? It might be interesting to see how it works with a new Ryzen compared to a more current cooler. The newer cpus have a higher heat density and might do better with newer coolers that take that into account.
I'm using a Hyper 212 Black Edition, it works on AM3 and AM4, so I'm just reusing it. It's been a great cooler. Usually sits at 45C real world loaded (the FX8350) with my routines running. I don't bother benching synthetics as they're not useful to my actual software in this case (I do those for fun though sometimes). Will see how it does on the 3700X. If I have to get a new one that's fine, I'm happy to keep things cooled down... Florida and all.

Very best,
 

sdifox

No Lifer
Sep 30, 2005
83,145
8,539
126
I'm using a Hyper 212 Black Edition, it works on AM3 and AM4, so I'm just reusing it. It's been a great cooler. Usually sits at 45C real world loaded (the FX8350) with my routines running. I don't bother benching synthetics as they're not useful to my actual software in this case (I do those for fun though sometimes). Will see how it does on the 3700X. If I have to get a new one that's fine, I'm happy to keep things cooled down... Florida and all.

Very best,
3700X is a 65w part, you'll be fine.
 
  • Like
Reactions: Tlh97 and MalVeauX

MalVeauX

Senior member
Dec 19, 2008
592
135
116
3700X is a 65w part, you'll be fine.
Thanks; hoping it will not be ultra hot at full load. My goal is stability on the workstation, not overclocking, and keeping temps realistic. Right now its not bad, its chilly in Florida (for Florida), but in the summer its annoying to have a bunch of heat generators in the office. But that's how it goes.

Very best,
 

sdifox

No Lifer
Sep 30, 2005
83,145
8,539
126
Thanks; hoping it will not be ultra hot at full load. My goal is stability on the workstation, not overclocking, and keeping temps realistic. Right now its not bad, its chilly in Florida (for Florida), but in the summer its annoying to have a bunch of heat generators in the office. But that's how it goes.

Very best,

I am using the stock amd cooler, Wraith Prism.
 
  • Like
Reactions: Tlh97 and MalVeauX

bononos

Diamond Member
Aug 21, 2011
3,718
68
91
I'm using a Hyper 212 Black Edition, it works on AM3 and AM4, so I'm just reusing it. It's been a great cooler. Usually sits at 45C real world loaded (the FX8350) with my routines running. I don't bother benching synthetics as they're not useful to my actual software in this case (I do those for fun though sometimes). Will see how it does on the 3700X. If I have to get a new one that's fine, I'm happy to keep things cooled down... Florida and all.

Very best,
The Hyper 212 series is a step up from the old stock FX banshee heatsinks and should work ok with the new Ryzens.
 
  • Like
Reactions: Tlh97 and MalVeauX

DrMrLordX

Lifer
Apr 27, 2000
16,880
5,842
136
@CB434

Lots of data there! Let me chime in on a few things:

1). What are you using to measure voltage? I use CPU-z almost exclusively, since the voltage figure it reports during load is generally higher than anything reported by HWINFO64 (Ryzen Master misreports voltage since it doesn't seem to account for vdroop or LLC at all).
2). EDC is only max "burst" current; namely, the limit the chip can sustain over a very short period of time. That limit is removed during static OC, and it can be raised via PBO. Compare this to TDC, which is the maximum sustained current allowed by the chip.
3). Prime95 SmallFFTs is really rough on Zen2, as you may have noticed! It will ramp up current like a sonofagun and as result lower voltage (and clocks) to stay within the PPT limit of the chip (88W). The higher temps of the chip may also affect the voltage/clockspeed curve and result in still-lower clocks.
4). I didn't see you test CBR20 @ stock and record both the current and voltage reported during the run. What voltage is reported by CPU-z and HWINFO64 during the run?
 

MalVeauX

Senior member
Dec 19, 2008
592
135
116
Well hot dang. The 3700x showed up from NewEgg in only like 4 days when it was not supposed to be seen till after Jan. And the motherboard was also a potential Jan receipt day and it ended up showing up 2 days later. I was not prepared for this. I guess I'll be building tonight.

Now the question is... keep the Hyper 212 Black Edition RGB or use the stock Wraith Prism that came with it?

Very best,
 

sdifox

No Lifer
Sep 30, 2005
83,145
8,539
126
Well hot dang. The 3700x showed up from NewEgg in only like 4 days when it was not supposed to be seen till after Jan. And the motherboard was also a potential Jan receipt day and it ended up showing up 2 days later. I was not prepared for this. I guess I'll be building tonight.

Now the question is... keep the Hyper 212 Black Edition RGB or use the stock Wraith Prism that came with it?

Very best,
Well if you are planning to sell the 3700X when you score the 5000 series, use the hyper 212 in the mean time.
 

MalVeauX

Senior member
Dec 19, 2008
592
135
116
Well if you are planning to sell the 3700X when you score the 5000 series, use the hyper 212 in the mean time.
Probably not, will likely just migrate the 3700x into a separate machine, for games (steam box) after I get a 5900x. I doubt the 3700x will hold any value once 5000 series are readily available so might as well use it and it will work fine for the games we play for years. At least, maybe.

The only difference I'm really seeing from reading about these two coolers is basically the sound, the Hyper being quieter with the larger fan. But otherwise, splitting hairs on cooling performance. Then there's the ability to push-pull configure the Hyper and can't do that with the Wraith Prism. So I think I'll continue with the Hyper 212 for now because I can line it up exactly with the system exhaust and just do a push push out of the case. The case is a NZXT Whisper full tower.

I've never had the Wraith Prism in my hands before, I was rather surprised how big it was. It's impressive for a "stock" cooler.

Very best,
 

DrMrLordX

Lifer
Apr 27, 2000
16,880
5,842
136
I find it fascinating that nearly a year-and-a-half after its launch, there's still enough demand for Matisse that it coming up for sale is a big deal, and that it's still moving @ MSRP. Bravo, AMD! I guess?
 

sdifox

No Lifer
Sep 30, 2005
83,145
8,539
126
I find it fascinating that nearly a year-and-a-half after its launch, there's still enough demand for Matisse that it coming up for sale is a big deal, and that it's still moving @ MSRP. Bravo, AMD! I guess?

Part of it is COVID.
 

DrMrLordX

Lifer
Apr 27, 2000
16,880
5,842
136
Part of it is COVID.
That is part of it. The demand spike is enormous, and it's estimated that the demand for PCs/laptops will continue through 2021:


Anyone providing hardware for those markets has to be cautious about what happens when there is finally a slip in demand. Don't want to over-extend.
 

MalVeauX

Senior member
Dec 19, 2008
592
135
116
So I finally got around to putting my new 3000 series together, with some parts harvested from the previous workstation; despite having had the parts for over a week, holidays just got in the way.

Ended up with:

ASUS TUF Gaming X570 Plus Wifi motherboard (new)
Ryzen 3700X CPU (new)
CoolMaster Hyper 212 RGB Black Edition (re-used)
TeamGroup Dark Force Alpha DDR4 3200Mhz CL16 32Gb (16x2) RAM (new)
Crucial SSD for OS/Software (re-used)
Crucial NVMe M2 for data work (new)
nVidia GTX 750Ti (strictly for 3 displays, no gaming, no PCIe rail needs) (re-used)
Corsair 750w PSU (re-used)
NZXT Whisper (Full size tower) (re-used)
Windows 10 64bit
VMWare for Linux Distros

I updated BIOS, no problem, just to make sure it was the latest. Was an easy process via USB drive.

I enabled D.O.C.P. (XMP) and the memory was instantly recognized and it's 3200Mhz and CL16 timings were detected automatically and it's stable and happy.

FirstBoot_ASUS_TUF_x570.jpg
(this screen shot does not reflect the BIOS update, this was first boot with just the OS SSD in place and the USB stick ready to install Windows 10)

Temperatures are ok. This CPU seems to be much hotter than my previous FX8350 (which is surprising as those old CPUs ran so hot). But I'm not liking the overall CPU temperature at load. Running my software, I'm seeing peaks of 63 Celcius. It's not sustained, but it peaks there now and then. But while running the routines it's in the mid 50's C and peaks around 64.3 at the highest I've recorded with OpenHardwareMonitor so far. While this is within spec, it's hotter than I would prefer day to day.

This temp is with the Hyper 212 Black Edition. So maybe I need to shop for a new CPU cooler that is significant superior to this? Or are these temps normal on this chip?

Anyhow, the real world results are promising so far in terms of performance. Definitely better than my ancient FX8350 that I just upgraded from on this workstation. It's all about time, so this is saving me time already.

My old system running a common routine on a piece of data, for example, a .SER container with 1500 RAW frames from a 20Mp CMOS monochrome sensor, around 30Gb, to sort the frames, align them, quality weight them, etc, the long end result is 1583.9 seconds of time (26.39 minutes).

FX8350_LunarFile_AS3.jpg

The new platform, with the same software, same data and same routine ran, resulted in a much better time. Total processing time came down to 552.4 seconds (9.2 minutes). This is roughly a 65% reduction in time, which ultimately will save me hours per session on the amount of data I typically generate and then process (anywhere from 250Gb to 350Gb is common per session).

Ryzen3700X_AllCores_LunarFile_AS3.jpg

To take it a step farther, I ran two common routines to off set the work. I ran one software to rapidly cull and quality weight the frames in the same file and output a new .SER container with less frames and only the best frames from the data set. It only took the software 177.5 seconds to do this. Then I ran this new data set, which is already sorted and quality weighted and reduced to only 250 frames in my second software routine. Then ran the same processing as above but on the smaller set, and the new processing time from start to finish was only 120.6 seconds. So combined, the two routines did the same work in 177.5 + 120.6 or 298.1 seconds (4.96 minutes). So ultimately going from 26.39 minutes to now 4.96 minutes is an 82% reduction in my time lost in just pre-processing.

PIPP_Lunar_Ryzen3700x.jpg

Ryzen3700X_AllCores_PIPP250framefile_LunarFile_AS3.jpg

So while I wanted a 3900x and then a 5900x, thanks to the market being really dumb right now, I'm not unhappy with the 3700X. It's a definite upgrade over my ancient FX8350 but most importantly it has bought me a lot of time with a much faster and more efficient use of the software I use (real world), as saving 82% of my time on the same work with the same data will save me hours of pre-processing time.

Next goals:

Sort out a new CPU cooler, if it's worth while.
Wait for a 5900x to become available.

Very best,
 

sdifox

No Lifer
Sep 30, 2005
83,145
8,539
126
So I finally got around to putting my new 3000 series together, with some parts harvested from the previous workstation; despite having had the parts for over a week, holidays just got in the way.

Ended up with:

ASUS TUF Gaming X570 Plus Wifi motherboard (new)
Ryzen 3700X CPU (new)
CoolMaster Hyper 212 RGB Black Edition (re-used)
TeamGroup Dark Force Alpha DDR4 3200Mhz CL16 32Gb (16x2) RAM (new)
Crucial SSD for OS/Software (re-used)
Crucial NVMe M2 for data work (new)
nVidia GTX 750Ti (strictly for 3 displays, no gaming, no PCIe rail needs) (re-used)
Corsair 750w PSU (re-used)
NZXT Whisper (Full size tower) (re-used)
Windows 10 64bit
VMWare for Linux Distros

I updated BIOS, no problem, just to make sure it was the latest. Was an easy process via USB drive.

I enabled D.O.C.P. (XMP) and the memory was instantly recognized and it's 3200Mhz and CL16 timings were detected automatically and it's stable and happy.

View attachment 36412
(this screen shot does not reflect the BIOS update, this was first boot with just the OS SSD in place and the USB stick ready to install Windows 10)

Temperatures are ok. This CPU seems to be much hotter than my previous FX8350 (which is surprising as those old CPUs ran so hot). But I'm not liking the overall CPU temperature at load. Running my software, I'm seeing peaks of 63 Celcius. It's not sustained, but it peaks there now and then. But while running the routines it's in the mid 50's C and peaks around 64.3 at the highest I've recorded with OpenHardwareMonitor so far. While this is within spec, it's hotter than I would prefer day to day.

This temp is with the Hyper 212 Black Edition. So maybe I need to shop for a new CPU cooler that is significant superior to this? Or are these temps normal on this chip?

Anyhow, the real world results are promising so far in terms of performance. Definitely better than my ancient FX8350 that I just upgraded from on this workstation. It's all about time, so this is saving me time already.

My old system running a common routine on a piece of data, for example, a .SER container with 1500 RAW frames from a 20Mp CMOS monochrome sensor, around 30Gb, to sort the frames, align them, quality weight them, etc, the long end result is 1583.9 seconds of time (26.39 minutes).

View attachment 36415

The new platform, with the same software, same data and same routine ran, resulted in a much better time. Total processing time came down to 552.4 seconds (9.2 minutes). This is roughly a 65% reduction in time, which ultimately will save me hours per session on the amount of data I typically generate and then process (anywhere from 250Gb to 350Gb is common per session).

View attachment 36416

To take it a step farther, I ran two common routines to off set the work. I ran one software to rapidly cull and quality weight the frames in the same file and output a new .SER container with less frames and only the best frames from the data set. It only took the software 177.5 seconds to do this. Then I ran this new data set, which is already sorted and quality weighted and reduced to only 250 frames in my second software routine. Then ran the same processing as above but on the smaller set, and the new processing time from start to finish was only 120.6 seconds. So combined, the two routines did the same work in 177.5 + 120.6 or 298.1 seconds (4.96 minutes). So ultimately going from 26.39 minutes to now 4.96 minutes is an 82% reduction in my time lost in just pre-processing.

View attachment 36417

View attachment 36418

So while I wanted a 3900x and then a 5900x, thanks to the market being really dumb right now, I'm not unhappy with the 3700X. It's a definite upgrade over my ancient FX8350 but most importantly it has bought me a lot of time with a much faster and more efficient use of the software I use (real world), as saving 82% of my time on the same work with the same data will save me hours of pre-processing time.

Next goals:

Sort out a new CPU cooler, if it's worth while.
Wait for a 5900x to become available.

Very best,

me think you want to go to Threadripper.
 

MalVeauX

Senior member
Dec 19, 2008
592
135
116
me think you want to go to Threadripper.
I thought about it, but ultimately when I saw any benchmarks of a threadripper vs something like the 3900x, the 3900x was beating it up. So for cost, it just seemed like the 3900x/5900x are the way to go for non-thread ripper application. The thing is my software here is both influenced by single core performance as some routines are only single threaded and others are heavily multi-threaded, so the CPU has to be great at both, and it's not 50/50 but it's enough that threadripper doesn't seem to be the ideal CPU for this stuff, as it's single core speeds just don't match these 3000/5000 series single core performances that I've seen. I only know of one person using threadripper with this software though, and he's the developer of it and the numbers reported don't look like threadripper is ideal overall for all routines. Maybe the best one would be great, but for cost, I think I'll stick to mid-tier stuff as I'd rather buy again in 5 years.

So while waiting for a 5900x or whatever else becomes a good option (I'm not against a threadripper if it truly would out-class a 5900x in the above software), I really need to sort out a good CPU cooler. This thing is running hot it seems. Maybe it's my chip. Maybe it's just my system. Not sure.

Very best,
 

dr1337

Member
May 25, 2020
112
166
76
TeamGroup Dark Force Alpha DDR4 3200Mhz CL16 32Gb (16x2) RAM (new)
If you can you might want to switch out this ram for something faster, ryzen already likes faster memory bandwidth and it seems like your specific application would also benefit from greater memory bandwidth. This synergy carries over to zen 3 as well, and if you're willing can be often pushed even faster up to 4000mhz if you're lucky.

Also from my experience peaks of 64c is about average for zen 2 with a cooler like the 212. Unless you're seeing some kind of heat saturation or throttling I wouldn't worry about it.
 

MalVeauX

Senior member
Dec 19, 2008
592
135
116
If you can you might want to switch out this ram for something faster, ryzen already likes faster memory bandwidth and it seems like your specific application would also benefit from greater memory bandwidth. This synergy carries over to zen 3 as well, and if you're willing can be often pushed even faster up to 4000mhz if you're lucky.

Also from my experience peaks of 64c is about average for zen 2 with a cooler like the 212. Unless you're seeing some kind of heat saturation or throttling I wouldn't worry about it.
Thanks; the temps are at the upper limit, but I'm trying to look up common temps for this chip to get an idea. I'm not a worry wart or anything, but if I can get it back into the 50's or something at load, I'd like that. But as you said, if it's not causing any performance issues, it doesn't matter much other than it being a space heater.

I actually went down to 3200mhz RAM because that's what is supported by the CPU/boards, and everything above it is an overclock and not supported. Too many weird issues I'm reading from the 5000 series guys who are using overclocked RAM and seeing errors and stuff. This workstation needs to be stable and not flip bits on the data. I'd love more speed, but not at the cost of stability. I waffled a while over 3600mhz and up, but ultimately went with "safe" speeds. Also a lot of the memory timing and latency performances are not universal in all applications, there's no benchmark that really shows my above needs, other than basically large file (4k) video editing is as close as a benchmark gets. I haven't seen major performance increases in video rendering from faster RAM compared to simply a better CPU (or GPU in some software). RAM seems to be what will be great to speed up and tweak after your CPU/GPU/storage is maxed. I'm no expert though, just what I've seen this far.

Very best,
 
  • Like
Reactions: Tlh97 and scannall

CB434

Junior Member
Sep 11, 2009
4
2
81
@CB434

Lots of data there! Let me chime in on a few things:

1). What are you using to measure voltage? I use CPU-z almost exclusively, since the voltage figure it reports during load is generally higher than anything reported by HWINFO64 (Ryzen Master misreports voltage since it doesn't seem to account for vdroop or LLC at all).
2). EDC is only max "burst" current; namely, the limit the chip can sustain over a very short period of time. That limit is removed during static OC, and it can be raised via PBO. Compare this to TDC, which is the maximum sustained current allowed by the chip.
3). Prime95 SmallFFTs is really rough on Zen2, as you may have noticed! It will ramp up current like a sonofagun and as result lower voltage (and clocks) to stay within the PPT limit of the chip (88W). The higher temps of the chip may also affect the voltage/clockspeed curve and result in still-lower clocks.
4). I didn't see you test CBR20 @ stock and record both the current and voltage reported during the run. What voltage is reported by CPU-z and HWINFO64 during the run?
Apologies for late reply, it's a hot summer here in Australia and my air con went out, right in the middle of testing this stuff. Fortunately, I was able to get the front panel off , find the broken part.. and have a new part ordered and delivered right before Christmas, and was able to install the part and get it working again. There was a fan motor capacitor that leaked/exploded, but yeah. I'm happy to have it running again, because at this time of year.. testing/overclocking/benchmarking etc.. is already at 30C+ ambient (without aircon). It makes it not even worth trying. But with air con, I can be back to focussing on testing Ryzen stuff again.

All of the testing/data I mentioned was at 26-29C ambient.. during those earlier posts, I forgot to mention that.

1) Initially, I was using CPU-Z, as it's the only other one that will work without freezing the sensor chip on my board. I would simply open CBR20 and have CPU-Z over the top of it and watch. But then once I got OCCT going.. I started using that. As it gives the HWINFO numbers but does so in a way that doesn't screw up my Super IO chip. It helps for min, max numbers etc. They give similar numbers, and reflect the droop a similar way.. they just show a slightly different number. For example 1.286 in CPU-Z, is 1.29 or 1.30 in OCCT (HWINFO). The CPU-Z number is the correct one IMO, because it reflects what I see in the BIOS. It droops in a similar way in the BIOS as it does in CPU-Z. For example.. 1.264, 1.286, 1.303. Where as HWINFO/OCCT shows 1.27, 1.29, 1.30 (or .31 can't remeber). Perhaps droop isn't the right word, as it isn't under load when it is in the BIOS.. but it fluctuates sometimes. It does it for VDIMM too.

2) I was able to raise the EDC by changing it in PBO like you said, but it didn't seem to do much. Other than maybe ran a tad hotter. I decided to leave it at stock for now.

3) For now I think I'm just going to ignore Prime 95 altogether, but if I did want to make it stable.. I'd just decrease clocks by 200mhz and adjust/lower voltage until it's stable.

4) In my tables, I think I did test everything but forgot to write certain things down (got too excited) and then was too lazy to go back, because I was on to the more fun stuff. I'll reset the CMOS and quickly do it now..

For everything stock (RAM not even at XMP).. CBR20 = 1.286V drooping occasionally to 1.264V, but spending most of the time at 1.286V. LLC at Auto. Max voltage during idle/SC boost spikes.. 1.526V. For PBO+75mhz = 1.395V with drooping to 1.373V. LLC at Auto. Max voltage during idle/SC boost spikes.. 1.526V.

Would I be right in thinking that my manual OC at 4.3 @ 1.286V is safer long term than running this PBO setting? The PBO takes all core clocks from 4050mhz during CBR20 all core, and boosts it to around 4150mhz. It also boosts SC clocks to 4475 or 4450mhz, when measuring latency in AIDA for example. It's only brief though, it doesn't sustain those clocks. In CBR20 SC, it sustains a clock of between 4425 and 4000 on the good cores, and 4375 to 4350 on the bad cores.

Table.jpg
 

Attachments


ASK THE COMMUNITY