• 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.

AMD Zen 2 (Refresh) 3900XT/3800XT/3600XT

Soldato
Joined
6 Feb 2019
Posts
17,464
This latency in games is ********. At his point its all about how a game is optimised. Look at far cry 5: 30 fps more on intel side in comparison with amd. Thats rediculious. These games are horrible written. Look at other games where amd is doing great. The better question is, how come one game is more latency sensitive than the other?

Btw I didn't do research on this one. Isn't so that zen+ has better latency than zen 2? How come zen 2 is performing better than?

I really think its comes down to how wel the game engine is written.

The engine used for far cry 5 is optimized for single thread clock speed rather than cores. Other Ubisoft games like assassins creed games don't have that problem
 
Associate
Joined
15 May 2020
Posts
387
Testing still, 4.6ghz, IF wouldnt go any further than 1900mhz, but ram is at 3800mhz 16-18-16-36-GDM 1.41v, will post some benchmarks soon, just making sure its stable, its quick thats for sure, said in an earlier post, I was going to buy a 3800X anyway so thought I may as well try the XT version instead.

I'll still buy a Ryzen 4000, will probably just give this one to my Dad.

I'll edit this post with benchmarks as I go along, that will do for memtest: https://imgur.com/0BlWqD1

CPU-Z Benchmark very interesting validation: https://valid.x86.fr/lv8c7h

Cinebench R20 Single & Multi: https://imgur.com/OQYDnPo

Bear in mind im still tweaking the ram a little.

Dram Calc default test: https://imgur.com/HSOTjEl

Nice! Thanks for sharing. Sorry if I missed it earlier in the thread, but what ram are you using? Number of dimms, size, brand?
 
Soldato
Joined
17 Nov 2007
Posts
3,161
Testing still, 4.6ghz, IF wouldnt go any further than 1900mhz, but ram is at 3800mhz 16-18-16-36-GDM 1.41v, will post some benchmarks soon, just making sure its stable, its quick thats for sure, said in an earlier post, I was going to buy a 3800X anyway so thought I may as well try the XT version instead.

I'll still buy a Ryzen 4000, will probably just give this one to my Dad.

I'll edit this post with benchmarks as I go along, that will do for memtest: https://imgur.com/0BlWqD1

CPU-Z Benchmark very interesting validation: https://valid.x86.fr/lv8c7h

Cinebench R20 Single & Multi: https://imgur.com/OQYDnPo

Bear in mind im still tweaking the ram a little.

Dram Calc default test: https://imgur.com/HSOTjEl

Very nice.

Did you try tweaking the bus clock, I usually bump it to around 102-103 to sort out those 98.** readings ;)
 
Soldato
Joined
12 Feb 2014
Posts
2,803
Location
Somewhere Only We Know
Nice! Thanks for sharing. Sorry if I missed it earlier in the thread, but what ram are you using? Number of dimms, size, brand?

4x8gb sticks of crucial ballistic RGB 3600mhz CL16, it's e-die and is brilliant at overclocking as you see, up to 3800mhz with much tighter timings than stock, considering all 4 slots are filled too, dead cheap too, got both kits off the rainforest at the same time.
 
Soldato
Joined
9 Nov 2009
Posts
24,765
Location
Planet Earth
This latency in games is ********. At his point its all about how a game is optimised. Look at far cry 5: 30 fps more on intel side in comparison with amd. Thats rediculious. These games are horrible written. Look at other games where amd is doing great. The better question is, how come one game is more latency sensitive than the other?

Btw I didn't do research on this one. Isn't so that zen+ has better latency than zen 2? How come zen 2 is performing better than?

I really think its comes down to how wel the game engine is written.

But engines being latency dependent is down to how they are written though - they should not need to be so sensitive to it.

The issue is the OS schedular has to make sure the first 4 threads are on the same CCX,if not there is a latency jump if it is done on two different CCX units. The same goes with games,if the engines or the games themselves are not fully aware of the CCX layout on Zen,they will randomly use threads.

If the game engines are optimised for Intel ring bus designs,which seem to have lower latency,then latency will affect performance.

Yes,it is also because engines are poorly written,and hammer one or two threads massively,ie,this is probably why latency is important,as everything is held back by these threads. So lower latency and high clockspeeds means you can exchange data quicker on these limiting threads. These games also seem to scale well with memory bandwidth too,which again I also imagine is because,a ton of data is being processed by certain threads,so slower memory is holding back the amount of data being transferred.


That video has some very good testing of the Ryzen 3 3300X,which has a single CCX. Look at how much memory tuning makes a huge difference. Again why would running faster IF and having faster memory with lower timings help Zen2 so much?? It reduces the memory-CCX delay(latency),and improves the rate of CCX-CCX data transfer.

Apparently Zen3 is moving to an 8 core CCX/CCD. This is probably going to help games more than most other kinds of software.


Games which are better load balanced between threads,not only have closer performance between Zen and the Intel CPUs,but tend to also be mostly GPU limited at any normal resolution.This is the problem with some of the older engines and games,they are not fully aware. Zen has to brute force these games as they are not optimised. The games devs really need to be retire these engines,especially as they use them on console ports too,but to save money they just use these rubbish engines even now.

This explains why the Ryzen 3 3300X is better than the Ryzen 3 3100X as its one CCX and it also has much more L3 cache.

Zen2 has a big IPC and L3 cache increase,and the L3 cache probably masks some aspect of memory-CCX latency,ie,stuff can be stored in the caches as a buffer. Zen2 has a massive L3 cache increase over Zen2.See what happens to the Ryzen 3 3100X performance with its dual CCX design,and half the normal amount of L3 cache over every other Zen2 model?

Now when the Zen2 based APUs come to desktop,it will be very interesting to see gaming performance with a dGPU. L3 cache is less,but the memory controller is on the same chip,and leaks show much better memory-CCX latency than the desktop Zen2 CPUs. It will be interesting to see if some of these games based on older engines improve on the APUs.

If you looked at Zen+ over Zen,some games showed big improvements as L1/L2 cache latencies decreased quite a bit. I run one of those poorly optimised games,ie,Fallout 4. Zen+ showed larger than average performance improvements over Zen due to its L1/L2 cache latency improvements.

But did you know that Skylake X is clock for clock,lower performance than Zen+ in the game?? Consumer Skylake CPUs are as fast or still faster than even Zen2 in the game. So despite the same Skylake cores,the change to a mesh bus and a different cache arrangement,actually increased overall latencies,which held back those Skylake cores.

There is also some other software which is held back by latencies including audio software,etc - you could see the Ryzen 3 3300X showing some decent improvements over the Ryzen 3 3100X.
 
Last edited:
Associate
Joined
15 May 2020
Posts
387
4x8gb sticks of crucial ballistic RGB 3600mhz CL16, it's e-die and is brilliant at overclocking as you see, up to 3800mhz with much tighter timings than stock, considering all 4 slots are filled too, dead cheap too, got both kits off the rainforest at the same time.

Firstly, sorry, I realised after I'd asked that you'd link to specs in your post.

I was curious as I've also been running 4 dimms (b-die at 3733 15-15-15-15-30 with GDM), which was going great until last night when it suddenly started crashing in AC Odyssey (3 times over a couple of hours), and a Karhu Memtest failed at 300 percent). No idea why, but anyway I've increased tRCDRD and tRP by one each and it's now at 6,000 percent with no errors, so maybe I was just always a bit close to the limit (although it had passed 3,000 percent before).

Anyway, great results! I'll definitely go e-die if ram requirements rise in future, it seems much better value.
 
Associate
Joined
29 Jun 2009
Posts
549
Just a word of warning these dont all do 1900mhz IF. This 3900XT maxes out at 1767mhz IF 1:1 with memory and will just about post at 1867mhz IF with memory at 3200mhz. Increasing vDDG does nothing. Swap in any of my old cpus and im back at 1900mhz 1:1 at between 985mv and 1030mv depending which cpu. Having serious buyers remorse about now - sad panda is sad (and £500 poorer) :(
 
Associate
Joined
29 Jun 2009
Posts
549
Just a word of warning these dont all do 1900mhz IF. This 3900XT maxes out at 1767mhz IF 1:1 with memory and will just about post at 1867mhz IF with memory at 3200mhz. Increasing vDDG does nothing. Swap in any of my old cpus and im back at 1900mhz 1:1 at between 985mv and 1030mv depending which cpu. Having serious buyers remorse about now - sad panda is sad (and £500 poorer) :(
Just to follow up on this, at stock cpu settings but with 3600mhz memory (all autos) in Cinebench R20 single core scores 523pts and multicore 7223pts. Boosts were 4375mhz with frequent jumps to 4549mhz in single core and 4050mhz multicore, flat and never moving. Very, very disappointing considering its under an EK velocity and temps barely break 60c. I'll try a bios update just in case, but these should work on everything.
If you got it from OcUK then return it. People have returned Intel K chips for being poor clockers before now, your XT is no different in this regard.
That's interesting, literally just say its bad silicon as the return reason? I mean from what I can see this one is worse than pretty much every 3900x out there... Though at desktop with no load the first CCX does tease with 4775mhz on one core and 4725mhz on another (the rest peak at 4700mhz). My second CCX looks like a junk 3600x though, with no load the various cores peak between 4250mhz and 4450mhz.
 
Caporegime
Joined
17 Feb 2006
Posts
29,263
Location
Cornwall
Caporegime
Joined
17 Feb 2006
Posts
29,263
Location
Cornwall
They offer a surprising amount of processing power, I don't think 250 is too far off the mark TBH.
Yawn. You're either completely delusional or just a terrible wind-up merchant.

Either way I'm way past the point of wanting to waste time with you :p

I remember when you said all AMD's chips are too cheap and they should charge way more for them. I see you haven't changed your mind.

Tbh I'm pretty sure you're just on the wind up.
 
Soldato
Joined
28 May 2007
Posts
18,197
Yawn. You're either completely delusional or just a terrible wind-up merchant.

Either way I'm way past the point of wanting to waste time with you :p

I remember when you said all AMD's chips are too cheap and they should charge way more for them. I see you haven't changed your mind.

Tbh I'm pretty sure you're just on the wind up.

I remember you saying you only upgrade every ten years, so unlike you I can understand why you have difficulty grasping the amount of performance on offer with the 3600XT and become confused by my comment that £150 was under priced for the 3600.
 
Soldato
Joined
4 Jan 2009
Posts
2,682
Location
Derby
Glad I got my 3700x 3 weeks ago now. Was contemplating a 3800xt or even a 3600xt but seems that for the £267 I paid made it great value compared to these new chips with a few % performance bump.

I suppose my 3700x @ 4375 MHz compared to the 3800xt @ 4600mhz would be worth a few quid but not nearly £100 more.

I have no doubt the prices will plummet soon.
 
Back
Top Bottom