Shadow Of The Tomb Raider - CPU Performance and general game benchmark discussions

[tabascosauz]

I'm.....not sure exactly what this benchmark is testing? This is HWInfo's data for the duration of the run.

4.45GHz @ 64°C and 80% core usage........? I play indie games and cross-platform mobile games that tax the 5900X harder than this, and regularly run Core 0/1 up to 4.9GHz. I've not seen even mildly CPU-bound games draw less than 13W on Core 0 and Core 1. This just looks like it was running on the GPU the whole time, not even 30% like it claims.

Anyways, ran it with the requested settings, but it seems like it isn't a very reliable CPU benchmark......not unless everybody has a 6800XT or better, it seems. Ran it a couple of times and it came out to the same 213fps result.

 

[harm9963]

Ram at 3800 14-14-14-14

 

[Mbellantoni]

I'm.....not sure exactly what this benchmark is testing? This is HWInfo's data for the duration of the run.

View attachment 213521View attachment 213520View attachment 213522View attachment 213519

View attachment 213517

4.45GHz @ 64°C and 80% core usage........? I play indie games and cross-platform mobile games that tax the 5900X harder than this, and regularly run Core 0/1 up to 4.9GHz. I've not seen even mildly CPU-bound games draw less than 13W on Core 0 and Core 1. This just looks like it was running on the GPU the whole time, not even 30% like it claims.

Anyways, ran it with the requested settings, but it seems like it isn't a very reliable CPU benchmark......not unless everybody has a 6800XT or better, it seems. Ran it a couple of times and it came out to the same 213fps result.

i think most of us are using this benchmark as a baseline to find out how much of a raw performance increase different speeds/manual overclocked ram profile would make over the stock xmp just for this particular benchmark. everyones results vary with different hardware so its not really a comparison between each other. most of us have been going back and forth seeing what works and what didnt for us and giving each other tips on how we may improve.

it gives a rough idea of performance increases in other games as well granted they are high fps and cpu heavy. its a great way to share knowledge and push our overclocks further and to learn from others.

 

[Cowboystrekk]

The 3800c16 profile was tested 25 cycles tm5 stable.
The 3800c14 not tested yet, it was just for fun test but the results wasnt as expected.

You might need a bit higher iod voltage. Try upping it by 0.02V and see what happends?

 

[Mbellantoni]

@Taraquin i got around to taking your advice and switching to 2t instead of gdm 1t. i wasnt convinced at all at first but i was proven wrong. it seems it took an extremely minor hit to bandwidth but my latency is consistently around 52.4-52.6ns now. even with that small change somehow it translated to this benchmark and i got my best runs consistently. to make sure it wasnt a fluke i switched back to gdm 1t and got lesser scores. this is at 4000/2000 cl 14-15-14-21.

i also experimented with turning off c states and messed around with the p states to see if i can get my latency lower. all it really did was stabilize any latency jitter between aida test at the cost of bandwidth and "sleep voltage" on my cpu was increased. so my cpu and flck woke up a little faster from idle pretty much. but it doesnt make a difference at all during gaming since the cpu is in a woken up state at that point. even then 2 or 3 quick latency or read test before running a full aida benchmark is enough to wake the cpu up completely to get your true results. it might help in certain scenarios though but i couldnt tell you which lol.

 

[bissag]

Ram at 3800 14-14-14-14

Can you share zenTimings settings?

You might need a bit higher iod voltage. Try upping it by 0.02V and see what happends?

iod voltage already at 1.05, I try to add more?

 

[Zyll Goliat]

There is no cache advantage. 32MB of L3 per chiplet, a core can't just access the other chiplet's L3.

L1&L2 cache are dedicated per core and not shared between the cores L3 as far as I know is shared between the cores......

Now not sure how this works with new Ryzen chips but assume is pretty much the same.....

 

[Cowboystrekk]

Can you share zenTimings settings?


iod voltage already at 1.05, I try to add more?

Same latency at better timings can sometimes be caused by too low voltage. For instance I get 3ns higher latency in aida with iod volt at 1.02, if I raise to 1.03 everything is okay. You have 32gb ram which might require a bit more than 16gb. Have you tried the dram calc test?

 

[tabascosauz]

L1 cache is dedicated per core and not shared between the cores L3 as far as I know is shared between the cores......

Now not sure how this works with new Ryzen chips but assume is pretty much the same.....

No I don't think they are the same in the way you think. On Ryzen 5000 they are literally separate dies under the heatspreader - there are 32MB L3 per chiplet in 5900X/5950X, but that doesn't mean a core gets access to 64MB. It doesn't get to go outside of its own chiplet, travel across the substrate (on an IF link that doesn't even exist, because the chiplets are only themselves linked to the I/O die), and go into the other chiplet's L3.

 

[Zyll Goliat]

No I don't think they are the same in the way you think. On Ryzen 5000 they are literally separate dies under the heatspreader - there are 32MB L3 per chiplet in 5900X/5950X, but that doesn't mean a core gets access to 64MB. It doesn't get to go outside of its own chiplet, travel across the substrate (on an IF link that doesn't even exist, because the chiplets are only themselves linked to the I/O die), and go into the other chiplet's L3.

Private L1/L2 caches and a shared L3 is hardly the only way to design a cache hierarchy, but it’s a common approach that multiple vendors have adopted. Giving each individual core a dedicated L1 and L2 cuts access latencies and reduces the chance of cache contention — meaning two different cores won’t overwrite vital data that the other put in a location in favor of their own workload. The common L3 cache is slower but much larger, which means it can store data for all the cores at once. Sophisticated algorithms are used to ensure that Core 0 tends to store information closest to itself, while Core 7 across the die also puts necessary data closer to itself.

 

[Selaya]

Yes, but if you look at the multi-CCD parts like the 5900X and 5950X, these have two CCDs each with 32MiB of L3, and C0-7 on CCD0 cannot access the CCD1's other 32 of L3.

 

[Zyll Goliat]

Yes, but if you look at the multi-CCD parts like the 5900X and 5950X, these have two CCDs each with 32MiB of L3, and C0-7 on CCD0 cannot access the CCD1's other 32 of L3.

Well as I said I am not so sure how all this goes with this new Ryzen CPU's but this is what I found:"AMD’s Ryzen processors based on the Zen, Zen+, and Zen 2 cores all share a common L3, but the structure of AMD’s CCX modules left the CPU functioning more like it had 2x8MB L3 caches, one for each CCX cluster, as opposed to one large, unified L3 cache like a standard Intel CPU"

 

[Selaya]

I'm not entirely sure about Zen(+) (which have 4-core CCXes, like Zen 2) but I know that Zen 2 caps out at 16MiB of L3 (up to 4x for a grand total of 64 for the 4 CCX parts, but only ever 16 usable per core).

 

[tabascosauz]

Well as I said I am not so sure how all this goes with this new Ryzen CPU's but this is what I found:"AMD’s Ryzen processors based on the Zen, Zen+, and Zen 2 cores all share a common L3, but the structure of AMD’s CCX modules left the CPU functioning more like it had 2x8MB L3 caches, one for each CCX cluster, as opposed to one large, unified L3 cache like a standard Intel CPU"

Yes, in the past L3 was further narrowed down based on CCX (because there were 2 CCX per chiplet). No longer have that problem because now 1 CCX = 1 chiplet.

But you don't have to overthink it. Half of the cores are literally not on the same piece of silicon as the other half. Simple as that. Core on one chiplet can access only the L3 on its own piece of silicon. The two chiplets look physically close but they aren't directly connected between the two of them.

i think most of us are using this benchmark as a baseline to find out how much of a raw performance increase different speeds/manual overclocked ram profile would make over the stock xmp just for this particular benchmark. everyones results vary with different hardware so its not really a comparison between each other. most of us have been going back and forth seeing what works and what didnt for us and giving each other tips on how we may improve.

it gives a rough idea of performance increases in other games as well granted they are high fps and cpu heavy. its a great way to share knowledge and push our overclocks further and to learn from others.

I completely get what you mean, but the point is that SoTTR is supposed to be a CPU-heavy test, thus fast mem should help. 4.6GHz at 10W per-core on two cores isn't CPU-heavy by any stretch of the imagination. That's the kind of temps/clocks/volts/power I'd expect while working in Premiere or Photoshop, not demanding or maxed-out or CPU-limited by any stretch of the imagination.

If that's how the benchmark really runs then forget memory profiles, a 4.5 or 4.6 all-core would absolutely dominate. Since the PB2 boost algorithm is literally sleeping on the job.

Are you (or anyone else) able to share some HWInfo logging during the test? Specifically effective clocks, power, and usage. From what I can tell running HWInfo during the run has a negligible impact on performance. Curious to see if it's something on my end.

For the record, this isn't a dig at the OP's chosen graphics settings. 800x600 makes very little difference to CPU behaviour.

 

[Zyll Goliat]

Yes, in the past L3 was further narrowed down based on CCX (because there were 2 CCX per chiplet). No longer have that problem because now 1 CCX = 1 chiplet.

But you don't have to overthink it. Half of the cores are literally not on the same piece of silicon as the other half. Simple as that. Core on one chiplet can access only the L3 on its own piece of silicon. The two chiplets look physically close but they aren't directly connected between the two of them.

View attachment 213568

So in short on 5950X this means that actually now 8 cores sharing 32Mb of L3 and other 8 cores sharing also 32Mb because 64Mb of L3 cache is split on two chiplets right?

 

[tabascosauz]

So in short on 5950X this means that actually now 8 cores sharing 32Mb of L3 and other 8 cores sharing also 32Mb because 64Mb of L3 cache is split on two chiplets right?

Just think of 5950X as two 5800X glued together, and 5900X as two 5600X glued together. Well, not together, but to the same substrate and connected to the same IO die. Obviously a rough concept but you get the idea

CPU-Z has the right idea, it lists L3 as 2 x 32MB, not 64MB.

 

[bissag]

Same latency at better timings can sometimes be caused by too low voltage. For instance I get 3ns higher latency in aida with iod volt at 1.02, if I raise to 1.03 everything is okay. You have 32gb ram which might require a bit more than 16gb. Have you tried the dram calc test?

I tried IOD up to 1.075 and CCD up to 1.080, no change I can't break the 55ns with that, while I should get around 53ns
didn't test yet, just benchmarks

 

[Zyll Goliat]

Just think of 5950X as two 5800X glued together, and 5900X as two 5600X glued together. Well, not together, but to the same substrate and connected to the same IO die. Obviously a rough concept but you get the idea

CPU-Z has the right idea, it lists L3 as 2 x 32MB, not 64MB.

Yeah that's what I thought.....in short it's just split in two but it's still been shared between the cores(2x8c) which brings us on the beginning when I said that 5950X is probably faster because have more cache(+higher turbo) not because it has more cores at that point as other CPU also had 8c/16t even if that cache is split in two it's not changing the facts that more L3 cache is always better especially when it comes to the gaming....
Here watch this

 

[tabascosauz]

Yeah that's what I thought.....in short it's just split in two but it's still been shared between the cores which brings us on the beginning when I said that 5950X is probably faster because have more cache(+higher turbo) even if that cache is split in two polls it's not changing the facts that more L3 cache is always better when it comes to the gaming....

What?

5600X: Core 0 can access 32MB of L3
5900X: Core 0 can access 32MB of L3
end of story
there is no "more" L3 cache available to any given core

There's no magic IF connection between the two chiplets. They function independently. In fact, so independently that often it feels like 2 x 5600X and not 1 x 5900X. Windows likes to use Core 0/1 on mine for its usual demanding tasks (opening apps, gaming, ST benchmarking), but it also loves to use Core 7 (on the other chiplet) for 95% of its background processing.

 

[Felix123BU]

What?

5600X: Core 0 can access 32MB of L3
5900X: Core 0 can access 32MB of L3
end of story
there is no "more" L3 cache available to any given core

There's no magic IF connection between the two chiplets. They function independently. In fact, so independently that often it feels like 2 x 5600X and not 1 x 5900X. Windows likes to use Core 0/1 on mine for its usual demanding tasks (opening apps, gaming, ST benchmarking), but it also loves to use Core 7 (on the other chiplet) for 95% of its background processing.

Yet there are some measurable differences in cache between the 5600x, 5800x and 5900x and 5950x
See below screenshots of Aida Cache and Memory bench, taken from this thread, but repeatable over other threads findings:

5600X 5800x 5950x

Now, I am not expert on CPU architecture, but those double numbers of cache speed jump out, and in certain situations, like this test also, there is a noticeable gain between them, so my humble opinion, which could be wrong, is that there is a difference in cache behavior between single and dual CCD Ryzen 5000 CPUs, which in certain scenarios can give a uplift.

My best example is also found in this tread, 2 very close systems, both with a 6800XT as GPU, and 3800mhz tuned ram for the 5800x and stock 3200mhz ram for the 5950x, even if the 5950x is paired with slower ram and worse latency, it has a gain of ~20FPS over the 5800x in a more CPU intensive game setting.

 

[Det0x]

This benchmark use more then 8 threads, especially in the last city scene -> dual CCD zen3 cpus score higher then 5600x/5800x with their single ccd's thanks to more cores, not more cache (this is one of the few game benchmarks that actually use threads)
Let me leave with posting one of my old testing screenshots which clearly show thread usage in city scene:

On a single CCD zen3 (8 cores) i get can around ~260-280 cpu fps average while above 300 with dual ccd @ same memory settings and cpu clockspeed

@ tabascosauz
Your settings are just plain bad, that's why your scoring low.. Oh and if you want to use hwmonitor to check what's going on, you need to up the pooling rate / understand the numbers
(hwmonitor cant keep up with the changing threads and clockspeeds)

@ Felix123BU

My aida 5950x screenshot above ?
Win11 makes L3 look bugged in aida... If you want one from win10 you can use this one:

Or if you want faster win11 screenshot:

But do note that my latency numbers are not normal for a dual ccd 5900x/5950x (average is 55-58ns latency in aida i would say)

 

[Selaya]

Just think of 5950X as two 5800X glued together, and 5900X as two 5600X glued together. Well, not together, but to the same substrate and connected to the same IO die. Obviously a rough concept but you get the idea

CPU-Z has the right idea, it lists L3 as 2 x 32MB, not 64MB.

Wait isn't the 5900X 8+4 ... ?

 

[Felix123BU]

Wait isn't the 5900X 8+4 ... ?

I don't think anybody knows exactly if its 8+4 or 6+6, initial reports said 6+6, in theory 8+4 would be better, but who knows, probably selected based on yields and CCD characteristics. Though I am not sure how such a random CCD selection process could work, but its not like AMD did not deliver 5800X's with 2 CCD, one of them being disabled

 

[Zyll Goliat]

Yet there are some measurable differences in cache between the 5600x, 5800x and 5900x and 5950x
See below screenshots of Aida Cache and Memory bench, taken from this thread, but repeatable over other threads findings:

5600X 5800x 5950x

View attachment 213583View attachment 213584View attachment 213585

Now, I am not expert on CPU architecture, but those double numbers of cache speed jump out, and in certain situations, like this test also, there is a noticeable gain between them, so my humble opinion, which could be wrong, is that there is a difference in cache behavior between single and dual CCD Ryzen 5000 CPUs, which in certain scenarios can give a uplift.

My best example is also found in this tread, 2 very close systems, both with a 6800XT as GPU, and 3800mhz tuned ram for the 5800x and stock 3200mhz ram for the 5950x, even if the 5950x is paired with slower ram and worse latency, it has a gain of ~20FPS over the 5800x in a more CPU intensive game setting.

I don't know but seems like he still believes like those cores have dedicated L3 per core(like L1.L2) and it's not getting that L3 cache is actually characterized as a pool of fast memory that is shared between cores and even if it's split in half you still have 2x32mb which will certainly add some worst latency but you will have more cache....
P.S.On that video that I posted above you can clearly and without the doubt see advantage of more cache....sure it was Intel CPU but I doubt that it is such big difference with those Ryzens If I could guess I will said that gains are not as much as with the Intel but never the less you will still see the advantage of more L3 cache....

 

[Felix123BU]

This benchmark use more then 8 threads, especially in the last city scene -> dual CCD zen3 cpus score higher then 5600x/5800x with their single ccd's thanks to more cores, not more cache (this is one of the few game benchmarks that actually use threads)
Let me leave with posting one of my old testing screenshots which clearly show thread usage in city scene:
View attachment 213588
On a single CCD zen3 i get can around ~260-280 cpu fps average while above 300 with dual ccd @ same memory settings and cpu clockspeed

@ tabascosauz
Your settings are just plain bad, that's why your scoring low.. Oh and if you want to use hwmonitor to check what's going on, you need to up the pooling rate / understand the numbers
(hwmonitor cant keep up with the changing threads and clockspeeds)

@ Felix123BU

My aida 5950x screenshot above ?
Win11 makes L3 look bugged in aida... If you want one from win10 you can use this one:
View attachment 213590
Or if you want faster win11 screenshot:
View attachment 213591
But do note that my latency numbers are not normal for a dual ccd 5900x/5950x (average is 55-58ns latency in aida i would say)

Yes, it was from one of your Aida tests