• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Is my 5800x Timespy score low?

Associate
Joined
19 Dec 2017
Posts
720
I've been looking through the Timespy results thread this evening and see my ~10800 CPU Score is way lower than anyone else with a 5800x

I was using Auto OC to achieve that score, so I decided to look at potential factors.

I'd been running my memory at XMP (2x16GB Micron-B @ 3600Mhz 18-22-22-22-42) so I decided to try DRAM Calculator timings but couldn't get the machine to post at 3200Mhz with those settings. Loosening the timings let it boot, but it was unstable - it seems there's no room to improve ram timings.

When that didn't work, I had a play with PBO - I set all cores at -15 except the two strongest at -10 with a max boost of +50Mhz which seemed to be about the strongest I could achieve without starting to test individual cores.

This got me up to 11460 which puts me joint with the lowest 5800x score in that thread - most are 12-13k+

So the question is, is the swing of the silicon lottery this large? Do I have a problem? Could I push OC harder or is it my Memory?
 
Soldato
Joined
30 Mar 2016
Posts
2,884
Location
East Lothian
I wouldn't say that no. I've ran two Timespy benches with my 5800X and scored 10351 and 10769. I don't think I lucked out in the silicone lottery as I've seen a lot of scores of over 15000 for the multi-core Cinebench R23 benchmark and I'm closer to 14000. You could try a run in that for comparison. If you want to try the bench in CPU-Z also the scores for the 5800x should be around 650 for single core and 6500 for multi-core.

Not sure about the impact memory has on the score. I'm running 4000 Mhz at 18-22-22-42.
 
Associate
OP
Joined
19 Dec 2017
Posts
720
I wouldn't say that no. I've ran two Timespy benches with my 5800X and scored 10351 and 10769. I don't think I lucked out in the silicone lottery as I've seen a lot of scores of over 15000 for the multi-core Cinebench R23 benchmark and I'm closer to 14000. You could try a run in that for comparison. If you want to try the bench in CPU-Z also the scores for the 5800x should be around 650 for single core and 6500 for multi-core.

Not sure about the impact memory has on the score. I'm running 4000 Mhz at 18-22-22-42.

I have 658 / 6450 in CPU-z and 1626 / 15246 in R23.

I wonder if Timespy is sensitive to Ram.
 
Soldato
Joined
20 Aug 2019
Posts
3,028
Location
SW Florida
I wouldn't say that no. I've ran two Timespy benches with my 5800X and scored 10351 and 10769. I don't think I lucked out in the silicone lottery as I've seen a lot of scores of over 15000 for the multi-core Cinebench R23 benchmark and I'm closer to 14000. You could try a run in that for comparison. If you want to try the bench in CPU-Z also the scores for the 5800x should be around 650 for single core and 6500 for multi-core.

Not sure about the impact memory has on the score. I'm running 4000 Mhz at 18-22-22-42.

Your ram is probably running in decoupled mode and taking a huge latency hit. 3600 is the fastest most boards will run in 1:1 without changing other settings to make it (and the memory controller) go faster.

Even then memory controllers that can handle over 3800 is very rare.

You can probably lower your ram speed to 3600 without even tightening timings and score noticably better.
 
Associate
OP
Joined
19 Dec 2017
Posts
720
Your ram is probably running in decoupled mode and taking a huge latency hit. 3600 is the fastest most boards will run in 1:1 without changing other settings to make it (and the memory controller) go faster.

Even then memory controllers that can handle over 3800 is very rare.

You can probably lower your ram speed to 3600 without even tightening timings and score noticably better.

I’ve been meaning to change my ram for a while, so this prompted me to give it a go.

I went for 4x8GB 4000Mhz CL18 stuff in the hope that I’ll get fclk running at 2000 or if I can’t, they’ll at least do tighter timings at lower clocks.
 
Soldato
Joined
30 Mar 2016
Posts
2,884
Location
East Lothian
Your ram is probably running in decoupled mode and taking a huge latency hit. 3600 is the fastest most boards will run in 1:1 without changing other settings to make it (and the memory controller) go faster.

Even then memory controllers that can handle over 3800 is very rare.

You can probably lower your ram speed to 3600 without even tightening timings and score noticably better.
I did read some stuff about 1:1 and the latency hit if this was disturbed going past 3800. But I also read that this had increased to 4000 with the Ryzen 5000 series too somewhere, did you hear anything about that? To check if the memory is running 1:1 then 'Uncore frequency' and DRAM frequency displayed in CPU-Z should be the same isn't that right? If that's the case, then I think I am decoupled as I'm showing uncore speed of 1000 Mhz and DRAM speed of 2000 Mhz. Perhaps this is why I'm scoring about 1000 less points than most in Cinebench R23 also...
 
Last edited:
Soldato
Joined
20 Aug 2019
Posts
3,028
Location
SW Florida
Amd said they are working on a new AGESA that should make higher frequencies easier to hit. Nothing beyond stock is guaranteed though. The "silicon lottery" still applies.
 
Soldato
Joined
30 Mar 2016
Posts
2,884
Location
East Lothian
Amd said they are working on a new AGESA that should make higher frequencies easier to hit. Nothing beyond stock is guaranteed though. The "silicon lottery" still applies.
Yeah I just tried setting my FCLK to 2000 Mhz. It booted into Windows ok but it didn't like it. CPU-z scores were more than halved.

It's also a little disappointing when those powerpoint slides AMD put out about optimizing 5000 series CPUS stated that "4000 MHz is the new 3800 Mhz for Ryzen 5000 series".
 
Last edited:
Soldato
Joined
30 Mar 2016
Posts
2,884
Location
East Lothian
Guess you had lots of errors if you had lower scores.

Try your ram at 3600 and fclk at 1800.
Yeah everything was slower to load, I got a GPU drivers error followed by a sound I've not heard since my ZX Spectrum days! I have the RAM running at 3800 and flck 1900 now and has been stable so far and performing as expected. I was uncoupled before using the DCOP 4000Hmz profile; the fclk was at 1000 Mhz. I've done that much tweaking with RAM and PBO settings today that my heads in a spin! The temps are good and my single core performance is better than average with the current settings. My multi-core performance in benchmarks seems to be a bit below the average but I think that's just how I fared in the silicone lottery. I've made my peace with where I'm at, for now... :D
 
Associate
OP
Joined
19 Dec 2017
Posts
720
So I've spent an hour or so testing some combinations with the new RAM this morning.

Firsly, there doesn't seem to be a great deal of change with Cinebench and CPU-Z. CPU-Z multicore increased as the ram speed went up and timings came down, but not hugely.

Where it really seems evident is timespy, which I presume will translate to gaming (I don't have time for huge suites of benchmarks unforuntately)

So here are my results for Timespy CPU only test - all tests were done with RAM and fclk at 1:1

2x16gb @ 3600Mhz CL18: 11163
4x8gb @ 3600Mhz CL18: 11955
4x8gb @ 3800Mhz CL18: 11953
4x8gb @ 4000Mhz CL18: 12125
4x8gb @ 3800Mhz CL16: 12047

It seems to me that the jump to four sticks was the primary reason for an increase in performance, which I don't think is a surprise after people like Gamers Nexus have reported on just this. I was quite surprised the machine posted at 4000/2000fclk and while it yielded a higher score, there were signs of memory errors like glicthing windows, weird artifacts on the desktop etc. Maybe AGESA 1.2 will help when that drops.

Going back to 3800 with tighter timings seemed to yield a similar result but more stable. Timings were 16-20-16-16-35. I didn't have time to play with these, so I just took them from someone else that has experimented with Micron E-die. I've yet to check stability with memtest, I shall leave that runing later today.

There's more to be found with overclocking the CPU as well as trying lower clock speeds and even tighter timings. There are reports of people hitting CL14 @ 3600 on Micron E-die, so there's an experiment for another day.

Overall, considering I wanted to change RAM for RGB tat anyway, it seems like a success. Cost difference isn't directly comparable as I've gone from plain RAM to RGB stuff, but it was roughly £150 more for the same capacity. Not the worst value and you could get most of the gain for much less if you just went for 4x cheaper 3600 RAM.
 
Back
Top Bottom