New R5 3600 build, memory write speed slow??

[tiggywombat]

Hello all, just built a new Ryzen 5 3600 system today with fresh install of Win 10 and all the latest drivers/BIOS on a Gigabyte B450M S2H.

Left the CPU at stock speeds, since reviews showed that overclocking gave little performance gain. But I managed to tweak my Apacer Panther 2666Mhz memory kit (2*8GB) to 3000Mhz by increasing voltage to 1.32V and loosened timings a little. System is stable after stress testing for an hour with AIDA64. However, I'm not sure why AIDA64 shows my write speed so much slower than read/copy. It should be around 40000MB/s if I'm not mistaken. I tested the memory at standard XMP settings (2666Mhz, 16-16-16-36) and the write speed was also slow. Compared with my 3000Mhz OC, the result was even worse at 21293MB/s.

Any idea why this is happening? Any input/advice would be much appreciated guys.

 

[phanbuey]

that looks about right.

Single CCX zen2 writespeed is low in AIDA.

"
We checked with AMD regarding the roughly half-speed memory write performance from our Ryzen 7 3700X in AIDA64. According to the response, this is to be expected given the application of one CCD for the eight-core Zen 2 part, as opposed to two CCD and therefore two links to the memory controller in the IO die for the 12-core chip.
Memory latency from Zen 2 is comparable to that of Zen and Zen+. This performance is, again, in spite of the segregated IO die for Zen 2 parts.
"

kitguru.net/components/cpu/luke-hill/amd-ryzen-9-3900x-ryzen-7-3700x-zen-2-cpu-review/7

 

[eidairaman1]

What specific mobos advertised higher speed?

 

[TheLostSwede]

This is one draw back of AMD's current CPU design, as pointed out above and you're seeing perfectly normal numbers.
This is my RAM tweaked to the max. Still "slow" writes.

 

[phanbuey]

This is one draw back of AMD's current CPU design, as pointed out above and you're seeing perfectly normal numbers.
This is my RAM tweaked to the max. Still "slow" writes.

Are those your stable settings?

 

[tiggywombat]

Wow, thanks for the clarification guys. I never knew about this issue until now. Guess I'll just live with it, I'm still happy with the performance overall.

 

[TheLostSwede]

Are those your stable settings?

Not with this UEFI, no, I can't go over 3600 now, but hopefully they will be with the new UEFI again, to which the download links are currently broken...

 

[phanbuey]

Wow, thanks for the clarification guys. I never knew about this issue until now. Guess I'll just live with it, I'm still happy with the performance overall.

It really doesnt mean anything for client workloads. the 3700x and in most cases the 3600 games/performs in most apps identically to my chip oc'd to 4.6ghz (i compare FPS from reviews using in game benchmarks - within 1 fps difference most time):

and it uses 1/2 the power, cost 1/3 what I paid for mine and:

so... it doesn't really matter. Write speed seems to have virtually 0 correlation on client side app performance.

 

[RAJOD1]

View attachment 130436

Hello all, just built a new Ryzen 5 3600 system today with fresh install of Win 10 and all the latest drivers/BIOS on a Gigabyte B450M S2H.

Left the CPU at stock speeds, since reviews showed that overclocking gave little performance gain. But I managed to tweak my Apacer Panther 2666Mhz memory kit (2*8GB) to 3000Mhz by increasing voltage to 1.32V and loosened timings a little. System is stable after stress testing for an hour with AIDA64. However, I'm not sure why AIDA64 shows my write speed so much slower than read/copy. It should be around 40000MB/s if I'm not mistaken. I tested the memory at standard XMP settings (2666Mhz, 16-16-16-36) and the write speed was also slow. Compared with my 3000Mhz OC, the result was even worse at 21293MB/s.

Any idea why this is happening? Any input/advice would be much appreciated guys.

I agree with you. I just built a Ryzen 5600x3D system in 2023 and the write speeds are 1/2 of the reads. Yet my crusty old x79 xeon has symmetric read writes on the same bench. 25,000 MB/sec write speed in 2023 on new system is not right. Something is up.
I've looked and still don't see a good answer old or in 2023 so lets wake this issue up again.

I see other AMD systems with symmetrical benchmark testing so I'm wondering if its the BIOS or MB issues.
Can someone answer this question because some have to drive 90 miles back to MicroCenter to return the ram or the MB or both. Looking at other AMD systems its not normal to have writes 1/2 the speed of the reads on AIDA64. If it is point me to the discussion on it.

 

[DirtyBiker]

I agree with you. I just built a Ryzen 5600x3D system in 2023 and the write speeds are 1/2 of the reads. Yet my crusty old x79 xeon has symmetric read writes on the same bench. 25,000 MB/sec write speed in 2023 on new system is not right. Something is up.
I've looked and still don't see a good answer old or in 2023 so lets wake this issue up again.

I see other AMD systems with symmetrical benchmark testing so I'm wondering if its the BIOS or MB issues.
Can someone answer this question because some have to drive 90 miles back to MicroCenter to return the ram or the MB or both. Looking at other AMD systems its not normal to have writes 1/2 the speed of the reads on AIDA64. If it is point me to the discussion on it.

same here I would expect better write speeds then my X58 system

 

[HD64G]

Since the IOD has two connections for 2 cpu chiplets, once the 2 chiplets are installed it gives full RAM writing bandwidth. If the cpu has only one cpu chiplet (6 or 8 cores) the RAM writing bandwidth is cut in half. Zen arch with chiplets is made so from the start.

 

[RAJOD1]

Since the IOD has two connections for 2 cpu chiplets, once the 2 chiplets are installed it gives full RAM writing bandwidth. If the cpu has only one cpu chiplet (6 or 8 cores) the RAM writing bandwidth is cut in half. Zen arch with chiplets is made so from the start.

Yes its not really a bug or issue at all on client machines. Write speeds make little if any real world difference in Games so its really a non issue just did not expect it on testing.
Same thing on 5800X3D (8 core) . The 12 cores will have symetrical read write.

 

[Bavor]

I found this topic while searching to find out why my 5800X3D CPU memory write speeds are so low in AIDA64 compared to two other CPUs in the same motherboard.

I'm not sure if its correct that single CCX zen2 write speed is low in AIDA.. My 5700G CPU has read and write speeds that are nearly identical across multiple runs of AID64. However my 5800X3D CPU has memory write speeds that are much lower than the memory read speeds across multiple runs of AIDA64. That's measured by AIDA64 on the same motherboard with the same RAM with the same timings and the same BIOS version. When I tested a Ryzen 7 5700 on the same motherboard, my memory read and write speeds were also nearly identical to each other. For some reason the only single CCX CPU that has slow write speeds is the 5800X3D.

5700G
Read: 55834
Write: 56622
Copy: 52015
Latency: 61.1

5800X3D
Read: 54238
Write: 31949
Copy: 52434
Latency: 66.7

 

[freeagent]

That is normal.

 

[RAJOD1]

I found this topic while searching to find out why my 5800X3D CPU memory write speeds are so low in AIDA64 compared to two other CPUs in the same motherboard.

I'm not sure if its correct that single CCX zen2 write speed is low in AIDA.. My 5700G CPU has read and write speeds that are nearly identical across multiple runs of AID64. However my 5800X3D CPU has memory write speeds that are much lower than the memory read speeds across multiple runs of AIDA64. That's measured by AIDA64 on the same motherboard with the same RAM with the same timings and the same BIOS version. When I tested a Ryzen 7 5700 on the same motherboard, my memory read and write speeds were also nearly identical to each other. For some reason the only single CCX CPU that has slow write speeds is the 5800X3D.

5700G
Read: 55834
Write: 56622
Copy: 52015
Latency: 61.1

5800X3D
Read: 54238
Write: 31949
Copy: 52434
Latency: 66.7

Perfectly nornal for the X3D chips. 5600x3d, 5700x3d and 5800x3d all have non symmetrical r/w. Don't let it bother you as it did me when I first tested Does not really affect real world performance in games.

 

[Bavor]

So its only the X3D chips that have non symmetrical read/write and not all single CCX chips as someone else stated above? Its just odd to me how both a 5700 and 5700G have symmetrical read/write speeds and the X3D chips don't have symmetrical read/write speeds.

 

[RAJOD1]

So its only the X3D chips that have non symmetrical read/write and not all single CCX chips as someone else stated above? Its just odd to me how both a 5700 and 5700G have symmetrical read/write speeds and the X3D chips don't have symmetrical read/write speeds.

Ive not tested them, so they might be correct its all single ccx. The bottom line is don't worry about it. They are great gaming cpus.

 

[DaemonForce]

new Ryzen 5 3600 system
fresh install of Win 10 and all the latest drivers/BIOS on a Gigabyte B450M S2H.

I managed to tweak my Apacer Panther 2666Mhz memory kit (2*8GB) to 3000Mhz by increasing voltage to 1.32V and loosened timings
AIDA64 shows my write speed so much slower than read/copy. It should be around 40000MB/s if I'm not mistaken.
I tested the memory at standard XMP settings (2666Mhz, 16-16-16-36) and the write speed was also slow. Compared with my 3000Mhz OC, the result was even worse at 21293MB/s.

Any idea why this is happening? Any input/advice would be much appreciated guys.

The clockspeed, timings and memory controller need as much agreement as possible. Trust me it's fine.

That's on a 2x16GB kit. Then there's 4x16GB: