Ryzen 7000 / Zen 4 Owner's Club

[Hurricane28]

11 hours ago, neurotix said:

@J7SC_Orionruns his RAM at 8.1Ghz, people do it here too. No longer an OCN reader...

 

Also I can run 8000 for benching or gaming but it won't pass my stability tests...

Je i saw that, amazing. 

 

It won't matter on my chip tho so i don't bother. On single CCX i belive 6400MHz is the max buildzoid said. 

 

Sadly the lower timing didn't work on the 6400 timings. Today i try again. 

[neurotix]

Good luck man.

[J7SC_Orion]

...finally had some time on the weekend to tune and extensively test eclk 104 / DDR5 8000 a bit more with CO. With eclk and negative boost offset, CO is a bit more of a challenge compared to my 'regular'  bclk 100 / 8000 setting. There may be a bit more room on per-core CO, but this is a highly iterative and thus time-consuming process.

 

[Hurricane28]

General Question, 

 

Does everyone here has the new AGESA 1.8.0.0? 

 

If so, what are the changes and did it improve much?

[J7SC_Orion]

22 minutes ago, Hurricane28 said:

General Question, 

 

Does everyone here has the new AGESA 1.8.0.0? 

 

If so, what are the changes and did it improve much?

 

 ...new bios for my Aorus 670E Master board hasn't dropped yet but from comments by folks with boards from other vendors (ie. MSI) the beta bios released for 1.0.8.0 allows for slightly higher FCLK (~ one step) and at least as good RAM oc.

 

Further to the Geekbench 6 results above via stress-tested eclk 104 / 8000, here are the updated Cinebench R24 results:

 

Edited October 23, 2023 by J7SC_Orion

[neurotix]

I have it but haven't tried OC'ing my memory since I updated, nor trying higher fclk. I require stability for folding@home 24/7 and Samba server for streaming to Raspberry Pis in Linux. So I stick to what I have that's proven stable, 7600MHz 36-45-45-45 and 2100fclk.

 

I *can* however remark that post time even from a cold boot is greatly decreased, from like 15sec to train the ram to like maybe 5 seconds tops total post time. With Memory Context Restore on.

 

Also @Hurricane28*shameless plug* take a look at my G.skill Flare X kit on sale in the marketplace here. Those timings I gave you (30-36-36-36-108-4-6-16) were from that kit. It's not RGB but it is Samsung B-Die DDR5 and is really good at 6000-6400 if you are after low timing at those speeds.

Edited October 23, 2023 by neurotix

[Hurricane28]

2 hours ago, neurotix said:

I have it but haven't tried OC'ing my memory since I updated, nor trying higher fclk. I require stability for folding@home 24/7 and Samba server for streaming to Raspberry Pis in Linux. So I stick to what I have that's proven stable, 7600MHz 36-45-45-45 and 2100fclk.

 

I *can* however remark that post time even from a cold boot is greatly decreased, from like 15sec to train the ram to like maybe 5 seconds tops total post time. With Memory Context Restore on.

 

Also @Hurricane28*shameless plug* take a look at my G.skill Flare X kit on sale in the marketplace here. Those timings I gave you (30-36-36-36-108-4-6-16) were from that kit. It's not RGB but it is Samsung B-Die DDR5 and is really good at 6000-6400 if you are after low timing at those speeds.

 

Thnx man, 

 

I do think i have a nice CPU, ram and motherboard combi tho cos 2133FclK is not too shabby imo. 6400MHz cl30 is need too.

 

I tried the other more tight timings but for some reason i couldn't boot anymore so i went back to other timings and been stable ever since. 

 

Still need to test 64GB tho but perhaps better on new AGESA. 

 

[neurotix]

1 minute ago, Hurricane28 said:

 

Thnx man, 

 

I do think i have a nice CPU, ram and motherboard combi tho cos 2133FclK is not too shabby imo. 6400MHz cl30 is need too.

 

I tried the other more tight timings but for some reason i couldn't boot anymore so i went back to other timings and been stable ever since. 

 

Still need to test 64GB tho but perhaps better on new AGESA. 

 

Can you post your ZenTimings of what you're running now? I'm sure they're in this thread somewhere else but I can't be bothered to read back and find them.

[ENTERPRISE]

3 minutes ago, Hurricane28 said:

 

Thnx man, 

 

I do think i have a nice CPU, ram and motherboard combi tho cos 2133FclK is not too shabby imo. 6400MHz cl30 is need too.

 

I tried the other more tight timings but for some reason i couldn't boot anymore so i went back to other timings and been stable ever since. 

 

Still need to test 64GB tho but perhaps better on new AGESA. 

 

If you do get 64GB, please get a 2x 32GB DIMM Kit as your chances of higher clocks are much better. 4xDIMMS adds additional stress to the CPU memory controller and your mileage would vary on how far you can push it. Just throwing it out there.

[Hurricane28]

3 minutes ago, neurotix said:

Can you post your ZenTimings of what you're running now? I'm sure they're in this thread somewhere else but I can't be bothered to read back and find them.

Sure thing man,

 

 

[Hurricane28]

1 minute ago, ENTERPRISE said:

If you do get 64GB, please get a 2x 32GB DIMM Kit as your chances of higher clocks are much better. 4xDIMMS adds additional stress to the CPU memory controller and your mileage would vary on how far you can push it. Just throwing it out there.

yes correct,. 

 

I am not planning on getting more ram tho, i just curious as to how it will run when i insert the other non rgb kit with this new kit. 

[neurotix]

32 minutes ago, Hurricane28 said:

Sure thing man,

 

 

Looks good to me. You should probably be sub-62ns timing in AIDA64? You're at cas 30 already. The other primary timings look good even though 36 or lower is ideal. Try lowering tFAW to 16 and tCWL to 24 or 20.

 

I couldn't even run 6400 on my B-Die kit, but had lower timings at 6200 before I upgraded to what I have now.

Edited October 23, 2023 by neurotix

[Hurricane28]

29 minutes ago, neurotix said:

Looks good to me. You should probably be sub-62ns timing in AIDA64? You're at cas 30 already. The other primary timings look good even though 36 or lower is ideal. Try lowering tFAW to 16 and tCWL to 24 or 20.

 

I couldn't even run 6400 on my B-Die kit, but had lower timings at 6200 before I upgraded to what I have now.

thnx man. 

 

idk what Aida bench i had, will do couple of runs tomorrow and some other timings. 

 

I think i will get mid 60s tho. 

 

I think its pretty good compared to what other people could do what i seen wit te same chip.

[neurotix]

@Hurricane28can you post your AIDA64 again? Also let Windows be booted for at least 15 minutes and you'll get more accurate results as if you do it when you first boot it's still running a lot of background tasks and phoning home etc

[Hurricane28]

46 minutes ago, neurotix said:

@Hurricane28can you post your AIDA64 again? Also let Windows be booted for at least 15 minutes and you'll get more accurate results as if you do it when you first boot it's still running a lot of background tasks and phoning home etc

 

Yes correct, and other apps also draw the latency down quite drastically. 

 

This is my best result so far: 

 

I think its pretty good. 

[neurotix]

You're at sub-60ns, that's a really incredible overclock, that's literally like the only AIDA of Zen 4 that I've ever seen with latency that low.

 

I'd set it and forget it and not worry about lowering timings further. That is an excellent overclock and should be great for gaming or anything else you do, assuming it is stable.

[Hurricane28]

1 hour ago, neurotix said:

You're at sub-60ns, that's a really incredible overclock, that's literally like the only AIDA of Zen 4 that I've ever seen with latency that low.

 

I'd set it and forget it and not worry about lowering timings further. That is an excellent overclock and should be great for gaming or anything else you do, assuming it is stable.

That good huh? Good to hear man! 

 

All thanks to you guys really! I am very happy with it, never had a single issue so far. Passed everything. 

[neurotix]

7 minutes ago, Hurricane28 said:

That good huh? Good to hear man! 

 

All thanks to you guys really! I am very happy with it, never had a single issue so far. Passed everything. 

I *think* maybe Fluxmaven, who also has a 7700x (or did) may also have posted an AIDA with that low of latency. But I don't think Flux follows or posts in this thread. The single CCD chips can get lower latency afaik. But it's looking pretty good to me.

[Hurricane28]

7 minutes ago, neurotix said:

I *think* maybe Fluxmaven, who also has a 7700x (or did) may also have posted an AIDA with that low of latency. But I don't think Flux follows or posts in this thread. The single CCD chips can get lower latency afaik. But it's looking pretty good to me.

 

Thnx, 

 

Ye i hear such a thing that there is always an penalty with duo CCX CPU's. I heard multiple people about it and latency is the most important thing for me so i guess i made an good choice. 

 

 

[J7SC_Orion]

1 minute ago, Hurricane28 said:

 

Thnx, 

 

Ye i hear such a thing that there is always an penalty with duo CCX CPU's. I heard multiple people about it and latency is the most important thing for me so i guess i made an good choice. 

 

 

 

...with dual CCDs, it is a bit of a tradeoff between bandwidth and latency, though 'I can live with it'. Per earlier post in this thread, 'nothing' will beat the latency in the lower screen 

 

 

[neurotix]

1 hour ago, Hurricane28 said:

 

Thnx, 

 

Ye i hear such a thing that there is always an penalty with duo CCX CPU's. I heard multiple people about it and latency is the most important thing for me so i guess i made an good choice. 

 

 

 

A lot of people neglect inter-CCX latency too. (The time it takes data to cross the Infinity Fabric to another CCD) Though, I am not sure what settings optimize it besides raising fclk. The old Zen 3 Ryzen DRAM Calculator had a inter-CCX latency option to test it with, but I have no idea if it would work with or would be accurate with Zen 4.

 

 

1 hour ago, J7SC_Orion said:

 

...with dual CCDs, it is a bit of a tradeoff between bandwidth and latency, though 'I can live with it'. Per earlier post in this thread, 'nothing' will beat the latency in the lower screen 

 

 

Nice. 58ns. Too bad my sticks or IMC isn't stable at your speed and timings. However, the difference for me between 7600 and 8000 was like 97GB/sec to 99GB/sec so I don't think I am losing very much performance.

 

As I've said before though, for me going from 6200 to 7600 has made a *big* difference, like 10 more FPS in Forza Horizon 5 benchmark. Higher minimums, higher framerate in Quake II RTX @3440x1440 144hz. I also love the RGB on the ram and it works with Asus' Aura Creator too.

Edited October 24, 2023 by neurotix

[Hurricane28]

55 minutes ago, neurotix said:

 

A lot of people neglect inter-CCX latency too. (The time it takes data to cross the Infinity Fabric to another CCD) Though, I am not sure what settings optimize it besides raising fclk. The old Zen 3 Ryzen DRAM Calculator had a inter-CCX latency option to test it with, but I have no idea if it would work with or would be accurate with Zen 4.

 

 

Nice. 58ns. Too bad my sticks or IMC isn't stable at your speed and timings. However, the difference for me between 7600 and 8000 was like 97GB/sec to 99GB/sec so I don't think I am losing very much performance.

 

As I've said before though, for me going from 6200 to 7600 has made a *big* difference, like 10 more FPS in Forza Horizon 5 benchmark. Higher minimums, higher framerate in Quake II RTX @3440x1440 144hz. I also love the RGB on the ram and it works with Asus' Aura Creator too.

Correct, 

 

I heard from The Stilt i believe too and an AMD engineer ( i am bad at names) also talked about this. 

 

This is why higher speed ram and high FcLk is more important on CPU's with high core count and double CCX. 

 

Past 6400MHz, i doubt i will see any difference cos of the lower cache compared to higher clocked CPU's. 

 

A friend of mine has an Intel and those are so different than AMD i mean, i know they are different brane etc. but they are so powerful and scale so well compared to AMD. 

 

He has an 14700k now and its an amazing piece of engineering to say the least. 

[neurotix]

I don't understand the whole P-Core/E-Core thing, I don't understand their RAM timings, and Z790 is a dead platform so your buddy is going to eventually need a whole new motherboard.

 

I bench and they are way better for benching, especially the older 3dmark benchmarks, 3dm11 and older. However, I wanted to support AMD and while Zen 4 is a marginal improvement over Zen 3+, Zen 5 is supposed to be fantastic so I wanted to have an upgrade route.

 

http://hwbot.org/user/neurotix

 

 

Intel finished 2nm and 1.5nm fabs in somewhere like Ohio that are due to come online in 2025, Zen 5 will be out by then, and I'll find out if I made a mistake.

 

Personally I would really like to see higher core counts on Zen chips, something like a 24c/48t for $750 and 20c/40t for $500 to replace the 32 and 24 thread offerings now.

Edited October 24, 2023 by neurotix

[J7SC_Orion]

IMO, .Intel is still in catch-up mode after an earlier period of mismanagement and scandals, and new designs and production facilities take a long time to get production-ready in such a behemoth, even with a proven engineer at the helm. The just-released Raptor Lake refresh is supposed to keep Intel in the consumer's mind until the real new designs are ready. I much prefer the AMD 670E platform for now since it has more genuine PCIe 5.0 lanes, never mind Zen 5 performance rumours and such. That said, I still have more Intel than AMD CPUs.

 

I think that more than 16C/32T does not make much sense for now on a desktop due to problems' feeding the beast'...the just announced / new Threadrippers have 4 channels of DDR5 (and as few as 12 cores), the Threadripper Pros have up to 8 channel DDR5 RDIMM - handy when running 96 C / 192 T - and the Epyc server models have up to 12 channels of DDR 5 RDIMM...For work as well as play, all the AMD CPUs I have are 16C / 32 T: TR 2950X / 64 GB; 3950X / 32 GB; 5950X / 32 GB and 7950X3D / 48 GB . The 4-channel DDR 4 Samsung-B of the TR 2950X almost matches the 2x 24 GB DDR 5 of the 7950X3D in memory reads and writes...

 

Faster & wider (> # cores) CPUs need faster & more cache which needs faster RAM which needs faster hard storage, and so it goes from cycle to cycle. Another advantage AMD has developed are the smaller Zen 4C cores which are 'like' e-cores on Intel but have the same basic micro-code as the 'regular' Zen 4, so no potential scheduler issues. The Zen 4C cores can probably mix it up with some RISC products. AMD also offers a huge range for the  server world with up to 128 C / 256 T per socket, or mega-VCache versions.

 

-----

 

I think I am finally finished with the various tuning options of my 7950X3D / Aorus 670E Master setup...at least for this current bios, until the new one is finalized, published and proven and stable.  The board's memory topography is outstanding and my daily setting and two alternate variants either run DDR5 8000 or slightly above (absolute measured and posted max was 8200, but not stable). My normal daily setting has stock bclk, and FCLK set to 2167 (2175). The fastest FCLK I ever measured is > 2277 MHz (bottom right), but the fastest stable FCLK is 2193 (bottom center-left). The new AGESA bios is rumoured to increase FCLK and possibly RAM bandwidth, but we'll see. For now, fastest effective clock / single core CCD 0 came in at 5304 MHz (using MSFS 2020) while the fastest effective clock for single core CCD 1 came in at > 5890 MHz (below).

 

The first stable and tested daily alternate is 'eclk' at 104 MHz set, while the second stable and tested daily alternate is 'bclk' at 102.4 set. BTW, one has to be very, very careful with bclk oc because unlike a crash in eclk oc while trying to find the limits, too optimistic a bclk oc can wreck one's FAT.

 

I showed Cinebench R23 and R24 as well as Geekbench 6 and Aida cache and memory bench over the last couple of days so I did not include it below. Instead, below is a collage of some other benches, such as PCI Express, CrystalDiskMark, y-cruncher and SuperPi summaries. All but the bottom right 'nutty setting' are used daily in a custom water-cooled work-play setup. Fortunately, my HWBot days are behind me so I don't have to prep my shiny brand-new boards for sub-zero anymore.  I still have nightmarish visions of taking a perfectly fine new mobo and lathering it in Vaseline and/or liquid electrical tape. That said, I still have my CPU and GPU pots, just in case, never mind that phase cooler that should not have a problem with a ~ 150 W  or 160 W CPU  

 

[Hurricane28]

12 hours ago, neurotix said:

I don't understand the whole P-Core/E-Core thing, I don't understand their RAM timings, and Z790 is a dead platform so your buddy is going to eventually need a whole new motherboard.

 

I bench and they are way better for benching, especially the older 3dmark benchmarks, 3dm11 and older. However, I wanted to support AMD and while Zen 4 is a marginal improvement over Zen 3+, Zen 5 is supposed to be fantastic so I wanted to have an upgrade route.

 

http://hwbot.org/user/neurotix

 

 

Intel finished 2nm and 1.5nm fabs in somewhere like Ohio that are due to come online in 2025, Zen 5 will be out by then, and I'll find out if I made a mistake.

 

Personally I would really like to see higher core counts on Zen chips, something like a 24c/48t for $750 and 20c/40t for $500 to replace the 32 and 24 thread offerings now.

 

Its a bit weird indeed but it works. The E cores do all the light work so the P (power) cores have enough recourses to do the hard labor lol. Its basically becos they cannot make stronger cores so they have to add E cores in order to be super fast. 

 

He also disabled E cores and it still stomps, Intel chips are an engineering masterpiece to be honest. Memory is much less of a hassle according to him, but idk, he always has been on Intel so. He had AMD system x570 and it had so much issues that he sold it and went to Intel, he says its a much better platform but i doubt that cos that platform is already dead and i can have another CPU refresh lol. 

 

My latency is actually on par with his 8K MHz ram lol. Quite impressive. I do want to change something but its running so fine now that i don't dare to lol. 

 

I would like to see less latency on AMD side and better IPC. Overall performance is great and we could use more cores indeed.