After a long needed hibernation

[OPPAINTER]

Here's my rig, same old same old. Swiftech stile, Bad to the bone watercooling!

[i found nemo]

no pic oppainter lol

[MrIcee]

WOWSA WOWSA

Eric my dear friend, such a long time Buddy Feels like old times with both of us in a forum thread ! So great to see and hear from you. You need to fix your pic above so we can see what you're packing. I'm digging the new AMD stuff, waiting for the 4000 series Ryzen and the Big Navi videocard to come out later this year to do some more 3D benching on water. Been having a ball, but no xtreme cooling for awhile, maybe after I retire in 2 years. Had multiple heart attacks last year and quintuple open heart bypass surgery and needed something to do, so turned back to the watercooled rig to play again. Hope to see you here again, drop me a PM anytime my friend !

Best regards,
Randi

[mdzcpa]

I am late to this thread, but wow, holy cow! Randi and Eric in the same thread here on XtremeSystems in 2020. I hope you guys circle back to say hello. I miss all my dear friends from the overclocking glory days.

-mike

[chispy]

Here's my rig, same old same old. Swiftech stile, Bad to the bone watercooling!

Here Erick , you can use Imgur to upload and host any image you want for free - https://imgur.com/upload , then click on the link for forum links and paste the url on the forum thread bar where it says upload image at the top of your post. Some image hosting services just dont work , but Imgur works 100% , i hope this helps.

[OPPAINTER]

I Love you Randi,,
Long time no talk too.

Eric

[OPPAINTER]

bEER RUN, iLL BE BACK

opp

[chew*]

Latest AMD driver, 2050 core / 1200 mem on the Radeon VII still stock CPU. Look like my CPU is functioning properly now

that's more like it charged.

[demonkevy666]

Hey guys long time since I posted took a while to read through.

There seems to be built in purposeful bottlenecks on Ryzen and some things where dropped on the 3000 series. I was looking up the full specs of infinity fabric and it's read rate is double that of it's write. It's read rage is 51.2(MT) vs write or 25.6(MT). However the chiplets/core complex seem to over this with two writing in tandem( I think I may be wrong). I read AVX512 was dropped because excessive heat, power and lower clocks. If I'm correct infinity fabric is only 256 bit wide. I would hope along with speed increase and bit rate increase in coming generations for lower latency high bandwidth throughput.

[chew*]

Hey guys long time since I posted took a while to read through.

There seems to be built in purposeful bottlenecks on Ryzen and some things where dropped on the 3000 series. I was looking up the full specs of infinity fabric and it's read rate is double that of it's write. It's read rage is 51.2(MT) vs write or 25.6(MT). However the chiplets/core complex seem to over this with two writing in tandem( I think I may be wrong). I read AVX512 was dropped because excessive heat, power and lower clocks. If I'm correct infinity fabric is only 256 bit wide. I would hope along with speed increase and bit rate increase in coming generations for lower latency high bandwidth throughput.

chiplets getting dropped for starters, that should reduce latency in itself. The real bottlenecks come in 3900x/3950x competing for memory access on 2 dies which is semi mitigated by running DR dims but still a bandaid. 3900x/3950x really needed a TR Dimm layout aka 2 channels designated per die layout but then that would defeat the HEDT product line.

[demonkevy666]

I highly doubt chiplets will get dropped. Too easy to bin from. Although laptop silicon is monolithic. Also has less L3 cache, For lower latceny.
I did a bit more research..... erereerer nope I'm still lost in this bottleneck. I looked again at the threadripper deep dive on techpowerup. The 3960x bandwidth starts just slightly faster than current AM4 chip with it a about 66.5 GBs up to 105 GBs something like almost 60% increase. The only thing that changed really drastically was minimum frame rate went up to 30% on average. The overclock was still about The same 8% average increase with memory tuning. The latency can come close to AM4 In ns. I read some people where having better performance with ram speeds on threadripper in 2:1 with speeds above 4200. Back to the laptop's monolithic die the smaller caches have it about a 10ns drop off current 3000 series. Unfortunately the review I looked at only had a 1660 gpu in it. Couldn't really see much improvement with 50ns latency vs 62ns. I think the split amount of L3 cache has an added latency.

Imo 10ns down ain't going cut it when intel is around half of 62 or less.

Do We know the penalty for a L3 cache hit in the other CCD? (L3 cache it is not split by CCX is it?)

We know three things about Zen 3. 1 it's CCX is now going to be an octocore. 2 it's L3 cache is one big pool instead two smaller split up. 3 aren't they increasing the FPU by 50%.
I'm starting to believe the load stores aren't big enough even for Zen 2. Like there waiting for new data, that's just not there.

Zen 2 seems to a wall on frame rates on lot of games regardless of clock speed.

[chew*]

Just some stuff I did awhile back concerning the Gigabyte Aorus Master rev 1.0 and Crosshair VIII Hero.

I might be late to the party but the older I get the less of a beta tester i want to be That said the platforms have matured a bit since launch.

All testing done at 1.35vddr or less.

Take this info and results on the Aorus Master with a large grain of salt as there is a 1.1/1.2 rev and I have no clue what they changed.

Single Rank 2x8gb passes fine stock volts actually I'm 1.34vddr with tuned out timings.





If using SR and 2x8gb the board will perform great and the UMC clocking is easy with all my cpu's doing 3800 with 2x8gb.

DR 2x16gb can get through this fine however when it comes to prime no matter what I attempted it would result in a hard crash even with just xmp loaded and 3600 spds. Maybe its the board revision, maybe its lack of bank group options in bios. No clue.

I can guarantee its not the memory as I'm going to follow that screen up with 32gb on the Crosshair VIII Hero.




Asus 32gb testing.

My findings on ASUS were that the UMC takes a small hit from 3800 to 3733 with DR installed. I did not see this happen with the gigabyte however since the gigabyte failed the rest except HCI its a moot point that it can do 3800 except in the case of "benchmarking".

In the real world however the ASUS can pass stability albeit 1 tick down on mem clocks so if 32gb is your goals I can suggest the Hero confidently and still sleep at night.







Any questions feel free to ask.

[FlawleZ]

Is your 3800X stable at 4.5Ghz and 1.27v?

[chew*]

Is your 3800X stable at 4.5Ghz and 1.27v?

I would say definitely not. my definition of stable = prime 95 without hitting 95c and throttling oh and it has to actually pass but usually it breaks rule #1 first.

I would also say that no launch silicon really was which mine is. Maybe the new glass can but I can't vouch for what I have not tested.

What my 3800x can do is run auto with PBO and hit up to 4650 with AMD holding the reigns. In games I avg 4.4. It really all depends on the actual cpu and the cooling its provided.

A good piece of glass will still suck on subpar cooling. A bad piece of glass won't suck as bad on decent cooling.

[FlawleZ]

These Ryzen 3000 cpus especially I think 3800X with the single offset chipset and higher TDP definitely need better than average cooling to have any chance of a decent overclock. Where your average AIO like an H100i could get you decent results on previous CPUs (mine would.keep an i7 3970X @4.4Ghz comfortable), they just don't cut it with Ryzen 3000.

[chew*]

These Ryzen 3000 cpus especially I think 3800X with the single offset chipset and higher TDP definitely need better than average cooling to have any chance of a decent overclock. Where your average AIO like an H100i could get you decent results on previous CPUs (mine would.keep an i7 3970X @4.4Ghz comfortable), they just don't cut it with Ryzen 3000.

I have a liquid freezer II 280mm on my 3800x. The heat from rad is non existent yet the chip still hits 84c ( prime 95 ). That tells me that the die is not transferring the heat well to IHS into water block. I don't think it's an interface issue either as it's been proven that delidding gains almost nothing. I just think it's a surface area issue. The die is small its a concentrated amount of heat on a very small portion of IHS. A larger die would transfer heat better. With the 3900x however the heat transfers more and the exhaust out of radiator is warm. It's got double the dies pumping heat into IHS.

[demonkevy666]

I have a liquid freezer II 280mm on my 3800x. The heat from rad is non existent yet the chip still hits 84c ( prime 95 ). That tells me that the die is not transferring the heat well to IHS into water block. I don't think it's an interface issue either as it's been proven that delidding gains almost nothing. I just think it's a surface area issue. The die is small its a concentrated amount of heat on a very small portion of IHS. A larger die would transfer heat better. With the 3900x however the heat transfers more and the exhaust out of radiator is warm. It's got double the dies pumping heat into IHS.

Yeah very tiny, but it also feels like they did low leak silicon for huge performance per watt gains here, what do you think?
Then I'm reminded of the clocks ... :-/

Oh I have a question is Ryzen master better than the old amd over drive?

[chew*]

Yeah very tiny, but it also feels like they did low leak silicon for huge performance per watt gains here, what do you think?
Then I'm reminded of the clocks ... :-/

Oh I have a question is Ryzen master better than the old amd over drive?

I only personally use it to verify settings are being applied as it is the "most accurate" but not for overclocking. All it does is reboot pc and apply bios settings for you anyway.

I mean obviously they are catering to consumer before ocer but the reality is they are building for server then trickling that down to desktop. In server power consumption = money. I still don't think they have the financial capital that intel has to "cater" products to various segments of the industry where as for the most part intels server chips are unique to server and there desktop are unique to desktop and laptop is unique to laptop.

AMD built enough capital to do server and laptop unique implementations and I think what we are getting is slightly revised for desktop glass.

[FlanK3r]

Cool, I have ready for upgrade my home PC Ryzen 5 3600XT and HSKill TridentZ Neo 2x 16GB 3600 MHz memory, propably your settings could be work on my PC (Crosshair 7 Hero, X470)

[demonkevy666]

Cool, I have ready for upgrade my home PC Ryzen 5 3600XT and HSKill TridentZ Neo 2x 16GB 3600 MHz memory, propably your settings could be work on my PC (Crosshair 7 Hero, X470)

Why would you do that when the 4,000 series will release 10-8-2020, they also mentioned that RDNA2 will release 20 days later on 10-28-2020.

That's like a building a R3 3300X with Nvlinking two RTX 2070 super's sure it was cheaper than one RTX 2080 TI, with better performance for some titles. Everyone else thinks it's a bad idea....

I wonder if the rumored 10 core will have a infinity link in-between the two CCD's to reduce latency.

They currently have to go from CCD to IOD to the other CCD in the 3,000 series. There is only a single link from the CCD to the IOD, which they also seem to have to do for current CCX configuration

[chew*]

Flanker probably doing it because he already has the hardware on hand from reviews. It's also mature hardware so for a daily that's great for reliability. I'm about to deploy my 3800x and CH8 into one of my daily's.

4000 series and next gen cards will be for testing debugging oc and reviews maybe some forum support.

As far as ryzen 3 desktop. It's getting rather weird I mean they should have just called it ryzen 4 4000 series but anyway. I doubt they can remove the "chiplet" for the high core count parts.

If you did not notice they managed to eliminate chiplet quite fine in the laptop APU parts and imagine that 1/1 UMC over 2000 rather easily and latency in the 40's. I called it pretty well. Lets see if they can figure a way around it on desktop now.

[demonkevy666]

Flanker probably doing it because he already has the hardware on hand from reviews. It's also mature hardware so for a daily that's great for reliability. I'm about to deploy my 3800x and CH8 into one of my daily's.

4000 series and next gen cards will be for testing debugging oc and reviews maybe some forum support.

As far as ryzen 3 desktop. It's getting rather weird I mean they should have just called it ryzen 4 4000 series but anyway. I doubt they can remove the "chiplet" for the high core count parts.

If you did not notice they managed to eliminate chiplet quite fine in the laptop APU parts and imagine that 1/1 UMC over 2000 rather easily and latency in the 40's. I called it pretty well. Lets see if they can figure a way around it on desktop now.

I keep forgetting he's also are reviewer. I haven't been around for 5+ years >_<

I'm not so sure about the Latency claims, like I said I only found one where it was shown only down to 50ns. The issues with what I've found is that the any of 4000G series benches I've found has about 1/4 the amount of L3 cache vs 3,000 series.

https://www.tomshardware.com/reviews...enoir-review/5

https://www.pcgamesn.com/amd/ryzen-5...-benchmark-360

On side note I'd like Launch bios reviews vs EOL for cpu for next generations to see how much Performance was squeezed out the said cpu.

Like luanch bios (3,000 series) vs launch bios (4,000 series) then add in The current Bios (3,000 series) vs (4,000 series) launch.

I'd even like that for GPU's and drivers on both sides.

[chew*]

In my experience especially on R1 some apps get faster some get slower and thats testing just r1 gen agesas. Like 32m kept getting slower and slower forcing myself and zeneffect to stay on very old bios versions when we were exploring zen 32m.

[demonkevy666]

I highly doubt chiplets will get dropped. Too easy to bin from. Although laptop silicon is monolithic. Also has less L3 cache, For lower latceny.
I did a bit more research..... erereerer nope I'm still lost in this bottleneck. I looked again at the threadripper deep dive on techpowerup. The 3960x bandwidth starts just slightly faster than current AM4 chip with it a about 66.5 GBs up to 105 GBs something like almost 60% increase. The only thing that changed really drastically was minimum frame rate went up to 30% on average. The overclock was still about The same 8% average increase with memory tuning. The latency can come close to AM4 In ns. I read some people where having better performance with ram speeds on threadripper in 2:1 with speeds above 4200. Back to the laptop's monolithic die the smaller caches have it about a 10ns drop off current 3000 series. Unfortunately the review I looked at only had a 1660 gpu in it. Couldn't really see much improvement with 50ns latency vs 62ns. I think the split amount of L3 cache has an added latency.

Imo 10ns down ain't going cut it when intel is around half of 62 or less.

Do We know the penalty for a L3 cache hit in the other CCD? (L3 cache it is not split by CCX is it?)

We know three things about Zen 3. 1 it's CCX is now going to be an octocore. 2 it's L3 cache is one big pool instead two smaller split up. 3 aren't they increasing the FPU by 50%.
I'm starting to believe the load stores aren't big enough even for Zen 2. Like there waiting for new data, that's just not there.

Zen 2 seems to a wall on frame rates on lot of games regardless of clock speed.

At around 9:40 on the AMD video

https://www.youtube.com/watch?v=iuiO6rqYV4o

I'm still interested in what this Bubble was in prediction ?

[FlanK3r]

Flanker probably doing it because he already has the hardware on hand from reviews. It's also mature hardware so for a daily that's great for reliability. I'm about to deploy my 3800x and CH8 into one of my daily's.

4000 series and next gen cards will be for testing debugging oc and reviews maybe some forum support.

As far as ryzen 3 desktop. It's getting rather weird I mean they should have just called it ryzen 4 4000 series but anyway. I doubt they can remove the "chiplet" for the high core count parts.

If you did not notice they managed to eliminate chiplet quite fine in the laptop APU parts and imagine that 1/1 UMC over 2000 rather easily and latency in the 40's. I called it pretty well. Lets see if they can figure a way around it on desktop now.

U right, I need Ryzen 9 5000 series for benchtable testing and fr my daily PC is Ryzen 7 3800X more than good Now Im on Ryzen 5 2600X and it is also OK for my daily PC, but ig I have Ryzen 7 3800X tested, and also I bought new case, it is good reason upgrade my system