FRONTPAGE AMD Ryzen 9 3900X and Ryzen 7 3700X CPU Review

[Johan45]

Zen2's 7nm Complications: Why not All Ryzen 3000 Cores Are Created Equal

I think most people sort of knew that going in, yes we were hoping for higher overclocks and maybe as the process matures they'll be pumping out better chips but the 3600X is still going to be bottom tier quality that only needs one core to make 4.4 GHz if the others will run 4.1-4.2 then it's an average CPU and still runs above base speed.
I also think a lot will depend on board/BIOS I have tested the same CPU in different boards and one will boost single-core 3900X to 4.6 and another only hits 4.45 same CPU only difference is the motherboard.

 

[Dolk]

I don't think we will ever see higher than 3 to 4GHz range as the norm for x86 CPUs. We are really hitting the limits of what the silicon, and architecture can do. These smaller process nodes have a large trade off of power vs frequency, and unlike previous node reduction steps, customers do not get both with each step down. I think it was starting around 22nm or 14nm where manufactures had to choose, in a steeper cut, which they wanted for their product: less power or more speed.

To get higher IPC (read: performance) changes to the architectural pipelines will be targeted. However, I think we will start to see the end of significant performance increases generation over generation. Nothing I've seen has been really showing large leaps and bounds over the last generation. Its been mostly software maturing with the current micro-architecture of AMD/Intel.

 

[UltraTaco]

Well, if software developers really start working on multicore optimizations, we will be set. Nowadays, most of the use you get out of like 12 cores is if you heavily multi task, but otherwise, you can't utilize all that power. Especially true in games. Imagine 12 cores being loaded!

Otoh, there isn't much more processing required to utilize all that power in a game..

 

[DaveB]

Well, if software developers really start working on multicore optimizations, we will be set. Nowadays, most of the use you get out of like 12 cores is if you heavily multi task, but otherwise, you can't utilize all that power. Especially true in games. Imagine 12 cores being loaded!

Otoh, there isn't much more processing required to utilize all that power in a game..

Programmers are lazy and will always reuse old code rather than generating new code. Plus most games and application programs have no real use for all those threads. This fantasy that developers will magically begin to muti-thread everything has been AMD fans mantra since Bulldozer was released in 2011 and we're still waiting for it to happen. The real significant advance for Zen 2 is the IPC increase.

I don't think we will ever see higher than 3 to 4GHz range as the norm for x86 CPUs.

Maybe for AMD but not Intel. My last 3 8th gen Intel CPUs, an i3-8350K, i5-8600K and i7-8700K all easily OC'd to 5 GHz. My last 3 AMD CPUs, a Ryzen 5 1600, Ryzen 7 1700X and Ryzen 7 2700 went to just 4 GHz give or take. This is Overclockers.com not Runatstockerspeed.com so that's how things are viewed here. All Kudos to AMD for Zen 2 IPC improvements but they've left no fun for us. Getting over AMD's typical early adopter woes isn't fun, just irritating.

 

[MaddMutt]

I'm still on the sidelines looking and saving for an upgrade around 12-25. I know I need another set of RAM. It will either be a slower 3800-4133 or a faster 4200 - 4600. I will also need a new motherboard x570 with good OC and a new CPU 3700x or a 3900x. I still have plenty of time to let things mature
I know I'm throwing a curve here BUT can someone test the i9-9900k HT cores only against the R7-3700x SMT cores.
IIRC - Johan45 and Earthdog have a i9-9900k & R7-3700x to be able to compare their results
This would be choosing to run CB15 with only the ODD Cores enabled.
This would hopefully show how much improvement AMD has put into their SMT effort
Currently Intel's CPUs clock higher BUT AMD's are = ~ at a slower speed. Pent-4 vs Athlon XP days ???
I know this is not a completely fair comparison BUT - - Intel is going to need to improve it's next CPU.
AMD's R9-3900x @ 5.0GHz and Intel's i9-7920x @ 5.9GHz
https://hwbot.org/submission/4190170_fugger_cinebench___r15_ryzen_9_3900x_3923_cb
https://hwbot.org/submission/3650961_dancop_cinebench___r15_core_i9_7920x_4000_cb/

 

[EarthDog]

I know I'm throwing a curve here BUT can someone test the i9-9900k HT cores only against the R7-3700x SMT cores.

It's not so much a curveball as it is an unhittable pitch. You cannot test HT without having real cores active. They are not physical cores on the chip, but logical. The second thread cannot exist without a core.

https://www.google.com/search?q=how...hrome.0.0l6.3810j1j7&sourceid=chrome&ie=UTF-8

There are tests on the web going over HT vs SMT efficiency and as is obvious, AMD is more efficient in that arena.

 

[pgdeaner]

I'm still on the sidelines looking and saving for an upgrade around 12-25. I know I need another set of RAM. It will either be a slower 3800-4133 or a faster 4200 - 4600. I will also need a new motherboard x570 with good OC and a new CPU 3700x or a 3900x. I still have plenty of time to let things mature
I know I'm throwing a curve here BUT can someone test the i9-9900k HT cores only against the R7-3700x SMT cores.
IIRC - Johan45 and Earthdog have a i9-9900k & R7-3700x to be able to compare their results
This would be choosing to run CB15 with only the ODD Cores enabled.
This would hopefully show how much improvement AMD has put into their SMT effort
Currently Intel's CPUs clock higher BUT AMD's are = ~ at a slower speed. Pent-4 vs Athlon XP days ???
I know this is not a completely fair comparison BUT - - Intel is going to need to improve it's next CPU.
AMD's R9-3900x @ 5.0GHz and Intel's i9-7920x @ 5.9GHz
https://hwbot.org/submission/4190170_fugger_cinebench___r15_ryzen_9_3900x_3923_cb
https://hwbot.org/submission/3650961_dancop_cinebench___r15_core_i9_7920x_4000_cb/

I think that any tests that make sense have been done. Your time would be better spent deciding what you want and what hardware is appropriate. There is a sweet spot for memory depending on which direction you take. Focus on your build and what you want to do with it.

 

[MaddMutt]

Zen2's 7nm Complications: Why not All Ryzen 3000 Cores Are Created Equal

From this testing, AMD will only let you disable 2 cores at a time

 

[Johan45]

I'm still on the sidelines looking and saving for an upgrade around 12-25. I know I need another set of RAM. It will either be a slower 3800-4133 or a faster 4200 - 4600. I will also need a new motherboard x570 with good OC and a new CPU 3700x or a 3900x. I still have plenty of time to let things mature
I know I'm throwing a curve here BUT can someone test the i9-9900k HT cores only against the R7-3700x SMT cores.
IIRC - Johan45 and Earthdog have a i9-9900k & R7-3700x to be able to compare their results
This would be choosing to run CB15 with only the ODD Cores enabled.
This would hopefully show how much improvement AMD has put into their SMT effort
Currently Intel's CPUs clock higher BUT AMD's are = ~ at a slower speed. Pent-4 vs Athlon XP days ???
I know this is not a completely fair comparison BUT - - Intel is going to need to improve it's next CPU.
AMD's R9-3900x @ 5.0GHz and Intel's i9-7920x @ 5.9GHz
https://hwbot.org/submission/4190170_fugger_cinebench___r15_ryzen_9_3900x_3923_cb
https://hwbot.org/submission/3650961_dancop_cinebench___r15_core_i9_7920x_4000_cb/

There's SMT HT comparisons in the review which gives a result in % so SMT adds X% for AMD and HT adds X% for Intel. Cinebench is included in those results

 

[Woomack]

I think I have the micro-ATX version of that case. If so, the airflow is awful, and the images for that one don't look any better. Did you modify it at all?

I was checking airflow with an open and closed case, and the difference was about 3-4°C. I also use SFX PSU, which saves a lot of space inside and has less and shorter cables. It's also Plat 80+ unit so runs passively even under full CPU load. Just above the CPU is 120mm Noctua fan which handles all the hot air pretty good and it's the only thing I replaced.
There is a cooler size limit as with ATX PSU it will be about 70mm, with SFX I could install 110mm cooler, and there is still about 10mm space. With modified ATX to SFX adapter can install 125mm cooler.
This PC is more like 24/7 small box for additional tests which I can take with me somewhere. I will run some more tests on ATX rig soon... when I finally find some time as last 2 weeks were pain at work.

Is he working for Asus now?

I only have a loose interest in really extreme OC, like replacing power delivery levels of modification. My career started in electronic engineering but I'm way past my interest point to do anything so deep nowadays.

Shamino was working for ASUS and looks like he still is. But he quit extreme OC scene a while ago. I haven't heard about him for maybe 2 years and now looks like he back with something new. Hard to say if it's one time appearance or he will work on enthusiast products some more. Some years ago he was tuning all Intel ROG motherboards and without him some world records wouldn't be possible. He was supporting ASUS overclockers for years but was not posting his results on hwbot directly (even though some of them were his).
Weird is only that he was never touching AMD stuff and now he did. Guys like Elmor were mostly playing with AMD.


Edit:

Started to play with ASRock Extreme4. So far 4x16GB@3600 CL16/Micron IC runs fine but board sets relaxed subs or something else what is causing 2-3GB/s lower read and copy bandwidth than on 2x8GB@3600 with the same IC. Will check that later and probably post some screenshots.

Btw. for the first time I used stock AMD cooler on these new chips. Max temp in specs is 95°C ... my 3700X with stock cooler, all at auto and full load, goes up to ... 94.8°C

 

[FlanK3r]

testing of memory Micron chips on ROG VIII /propably Impact/, 5200 MHz stable Tested by user Safedisk

 

[Woomack]

Nice one. I have 8 sticks of this RAM and it runs between 4200 and 4900. I wish to see memory performance as at 4800 it was still below of what 3600-3733 1:1 could make.

Btw. one weird thing on Gigabyte. When I enable virtualization then bclk is 1MHz lower at stock and 1.6MHz lower once I set it to 102MHz (so 99MHz or 100.4MHz). Not a big problem just a bit weird.

Also, these chips are getting really hot and I wasn't expecting that, especially looking at 65W TDP chips. With a cooler that should handle ~150W, R5 3600 hits 95°C. AMD stock cooler can't really handle these processors. At auto, when CPU hits 95°C then it throttles to keep temps within specs (so 95°C).

 

[mackerel]

Heat is an interesting thing with these CPUs. I downgraded the cooler on one for cosmetic reasons. The 3600 I have was originally under a Noctua D9L. Under P95 6x128k FFT load it stabilised around 80C at 3.9 GHz. Dropping the Wraith Prism in place, it was near enough same clock (maybe flicking 25 MHz [one step] down more often) at 88C. So no meaningful throttling due to temperatures. Limiting reason in Ryzen Master was given as PPT, with current being close to the limit also. The fan of the Prism runs at higher speed though, so noisier. Note the Stealth is bundled with the 3700X, I just tried it on the 3600.

 

[Woomack]

I can set 4GHz 1.225V (R5 3600) on the D9L but it still reaches 95°C in small FFT or AIDA CPU+FPU. On Cryorig C7 Cu it went up to 100°C and the fan was at max speed. It's 115W TDP cooler so should handle this CPU.
3700X with Scythe Fuma 2 went up to 88°C at auto settings and AIDA CPU+FPU. When I compared the same settings on 3600 and 3700X then temps were similar even though there are 2 cores difference. I have 2-3 more coolers to test but somehow it looks like there is a problem with transfering the heat from dies ... or simply it runs hot.
Btw. 3700X passed 30 mins AIDA at 4.3GHz 1.37V where 3600 needed ~1.3V to run at 4.1GHz.

 

[bignazpwns]

I think the issue is the offset chiplets. And 7nm is a really focused heat spot.


I been running a carbon heatpad with good results. I seen a 4c drop in temps vs Noctua NT-H1. I think it's from the pads ability to spread the heat out from those focused spots.

Woundering if anyone else seen this with a carbon pad. Why not the best for heat transfer in the z axis. They are amazing for x and y.

 

[EarthDog]

I think the issue is the offset chiplets.

Why? I've seen you say this, but haven't really thrown out a theory (that I caught). The cold plate of any cooler touched the entire surface. What isn't directly over it perhaps are the microchannels in SOME blocks... but, I don't think that is the issue. I get a great spread on my paste and see similar results.

 

[mackerel]

I had wondered if heat spread might be a problem too, but not sure how to prove or disprove it. Maybe the heat spreader is good enough for it not to be a problem. I'm thinking for a typical tower cooler, you have heatpipes running along the CPU. With the offset CCD maybe more heat will go to one part, with the other part less active. If true, I wonder if this would show up in IR imaging. I could try this...

BTW I'm running a Prime number finding challenge elsewhere. On the 3600 with the Prism cooler (not Stealth I wrote earlier) it is running up to 90C or so. Of completed work, I've had a couple of "inconclusive" units. This indicates a potential computational instability if it is proven my system's result is the incorrect one. I think it has the HyperX 4000 ram in it at 3600. Workload is 1280k FFT and I'm running one per CCX, as this was determined to be best throughput. IIRC IF speeds impact the connectivity between cores and L3 cache, so wondering if that is a problem, or the cores themselves? Or maybe I should slap the D9L back in... the Prism is louder and hotter by some margin, even if RGB!

Edit: scratch that, I forgot I swapped the ram with Crucial 2666 before running. It probably isn't a ram/IF speed thing.

 

[Woomack]

I have no issues with spreading too. I was checking 4 coolers so far and 3 more on the way. Most of them had heat pipes covering the whole base. Used Noctua H1 and H2. The H2 spreads really well. I don't care about 3-4°C drop ... I'm missing like 20°C to be happy with temps.
The main problem is that terrible temps are on a lower chip with 6 cores and lower clock. It supposed to be a replacement for i3-i5 while it requires a quite big cooler. Not to mention that stock cooler designed for this CPU can't handle it. They could make like Intel ... design a CPU which has ridiculous specification, don't add any cooler and just wish luck to their users ... at least then no one could complain at pathetic cooling solution added to the CPU. But hey, it has a really nice RGB lighting

 

[EarthDog]

I would honestly guess the multiple chiplet design would be BETTER as only so much heat can actually escape the smaller footprint of a single die...It is more surface area to get rid of any heat. I would be curious to see what an accurate IR reader has to say about the hot spots. I'd imagine its still in the CCX and not the controller chiplet...

 

[Woomack]

ITX box / Sharkoon CA-I case (cable management looks better with PSU installed), space for gfx card ~205mm, the same MSI GTX1660Ti Ventus ... perfect fit side to side ... had to remove bottom of the case to install it
hardware inside: R5 3600@4GHz 39.25x102, Aorus X570 I Pro, mem@3733 16-16-16, M.2 PCIE 1TB, Noctua D9L, Corsair 600W SFX 80+Plat and mentioned gfx card ... one day will replace the CPU with 3700X which heats up less and OC better but I need it for other tests right now