vlatko_1

Distinguished
Jan 2, 2016
73
4
18,545
Hi people need help, I have b450 tomahawk max board and r5 3600 cpu need to know what is the benefit of turning it off or on the SMT for gaming...

also please what else option in mobo need to be disabled or enabled for best fps in gaming?
p.s. I already run xmp 1 profile...
thank you so much ❣️
 
Solution
Aaand now? after all this quality conversation what is best for games, I think after this might switch back to intel i7 11700 and good b560 mobo, and not bother with this complex amd things...
It's not difficult at all: if you've messed with your BIOS and don't know where it's at: reset CMOS. Then enable XMP again but leave clocks and voltage on AUTO. The most difficult part is install the chipset drivers. But you've done that much already right?

That will get you all the performance that really matters, any other tweaks are incredibly minor and won't be noticed so don't bother with them. Especially don't bother with disabling SMT. If you feel the need for more performance put a 5600X on that B450 Tomahawk and you'll get it...

vlatko_1

Distinguished
Jan 2, 2016
73
4
18,545
SMT is what AMD have on their processors as well as Intel but under a different moniker, Hyper Threading. It's best you leave it enabled since disabling it can affect gaming performance. To add, you're not going to see any benefits disabling it.
thanks a lot just one more question about bios settings for AMD is it best to leave all by default in bios or need to enable or disable something, like PBO etc.... for gaming? thanks one more time xD
 
thanks a lot just one more question about bios settings for AMD is it best to leave all by default in bios or need to enable or disable something, like PBO etc.... for gaming? thanks one more time xD
With the exception of RAM speed settings, you generally don't have to touch anything to get a lot out of your system. You can probably push it harder, but it amounts to like 5% more performance in ideal situations for a lot of work.
 
To add, you're not going to see any benefits disabling it.
Lower power draw=less heat, depending on his setup=more clocks=better gaming.
But it is an case to case issue.
I agree leave it on. From what I read it’s only some older games that perform better with it off.
That might be so but newer games also don't run any better if you exceed the optimal number threads (after a certain amount of threads all games, new included don't give any more benefit but do use more power on more cores)
 
Lower power draw=less heat, depending on his setup=more clocks=better gaming.
But it is an case to case issue.
For Ryzen this is sort of a lost cause anyway as each core has a different limit how fast they can clock to. Since Ryzen doesn't seem to have independent clocking, if you hit all of the cores, it's going to slow down to the fastest the slowest core can handle.

That might be so but newer games also don't run any better if you exceed the optimal number threads (after a certain amount of threads all games, new included don't give any more benefit but do use more power on more cores)
If the game has say 8 threads that can be serviced on an SMT system, then the processor only needs to fire up 4 cores, which is obviously going to consume less power than 6 cores on top of there being a performance hit due to 2 threads not being serviced.
 
For Ryzen this is sort of a lost cause anyway as each core has a different limit how fast they can clock to. Since Ryzen doesn't seem to have independent clocking, if you hit all of the cores, it's going to slow down to the fastest the slowest core can handle.
Precision boost would like to disagree on that, yes cores do have different max speeds and on lightly threaded workloads if there is enough temp/tdp headroom it will boost the best core to the highest clock it can reach.
Your not always running all cores 100% ,whenever there is a chance it will boost a single core "much" higher.
 
Precision boost would like to disagree on that, yes cores do have different max speeds and on lightly threaded workloads if there is enough temp/tdp headroom it will boost the best core to the highest clock it can reach.
Your not always running all cores 100% ,whenever there is a chance it will boost a single core "much" higher.
That's in the scenario when there's a light workload. I'm talking about a scenario when there's enough workload to peg all of the cores for a significant amount of time.

And again, I don't believe Ryzen has independent clocking for each core, so once you hit more that two cores, that best core is going to clock down.
 
That's in the scenario when there's a light workload. I'm talking about a scenario when there's enough workload to peg all of the cores for a significant amount of time.

And again, I don't believe Ryzen has independent clocking for each core, so once you hit more that two cores, that best core is going to clock down.
Ryzen clocking is independent for each core and you can see that in the core boosting behavior in HWInfo64.

3rd gen will only boost one core at a time to maximum rated clocks but will shift the boost to another core in the same core complex as the scheduler moves the thread to that core so as far as the processing load cares it's still getting the higher performing boost. The scheduler is aware of the core architecture and knows to move threads to cores with shared resources so there's limited to no latency loss, and further knows the CPU's core ranking order. That's what CPPC is all about.

I've an early silicon 3700X that only hits max rated clocks on 6 cores. When I run a light processing load with say 2-4 threads it bounces them around those 6 cores round-robin style. I'm sure none are ever maxxed at the same time, but then no utility can show that since the algorithm makes decisions 100 times a second. Thats way faster than any utility polling period (500mS what I've set HWinfo to).

You can't control each cores clocks for overclocking though...but you can control a die or CCX if I'm not mistaken. I don't do that so not too familiar with it how it works...or how it's different with 4th gen.

But as far as OP's titular question:
https://www.gamersnexus.net/guides/2835-amd-ryzen-r7-1700-smt-off-overclock-benchmarks

TL/DR: Even though these are 4 year old systems and older gaming titles it's still relevant as it shows one clear principle: it can depend on the game but it's mostly better with SMT on.
 
Last edited:
Ryzen clocking is independent for each core and you can see that in the core boosting behavior in HWInfo64.

3rd gen will only boost one core at a time to maximum rated clocks, but will shift the boost to another core in the same core complex if the scheduler moves the thread to that core. The scheduler is aware of the core architecture and knows to move threads to cores with shared resources so there's limited to no latency loss.

You can't control each cores clocks for overclocking though...but you can control a die or CCX if I'm not mistaken. I don't do that so not too familiar with it how it works...or how it's different with 4th gen.
What I mean by "independent clocking" is that one core will boost up to the limit while the other cores boosts to theirs, whatever they may be. So for the 5600X, this would mean, when every core has something to do, the preferred cores are boosted up to 4.6GHz while the others are boosted to whatever their limits are. I've not seen this behavior and from what I can tell, mixing and matching clock frequencies in cores isn't done often or at all due to the complexities with different performing cores and the need to have cross-clock domain communication.

Also I take frequency reports with a grain of salt. Unless you're measuring at the same rate as the clock speed, there's only so much you can report once per second when the processor is likely changing its frequency multiple times per second.
 
.....
Also I take frequency reports with a grain of salt. Unless you're measuring at the same rate as the clock speed, there's only so much you can report once per second when the processor is likely changing its frequency multiple times per second.

That's exactly right about Ryzen. It's highly dynamic and constantly dithering clocks within it's FIT parameters and shuffling processing loads between cores to manage heat loading so you can't look at a 'snapshot' and say anything with certainty about how it's clock speed relates to performance. You have to look at an average clock speed across a period of time or, probably more correctly, at performance metrics in real-world task type benchmarks like Cinebench20 and 23.

I'm certain 3rd gen does not boost more than one core to max rated speed simultaneous with any other but they will all hold the same high-mid clock in a heavy workload. At least that's what mine does. I thought 4th gen can all boost to max simultaneous though it may just look like it with monitoring software.
 
Last edited:

vlatko_1

Distinguished
Jan 2, 2016
73
4
18,545
Aaand now? after all this quality conversation what is best for games, I think after this might switch back to intel i7 11700 and good b560 mobo, and not bother with this complex amd things...
 
Aaand now? after all this quality conversation what is best for games, I think after this might switch back to intel i7 11700 and good b560 mobo, and not bother with this complex amd things...
It's not difficult at all: if you've messed with your BIOS and don't know where it's at: reset CMOS. Then enable XMP again but leave clocks and voltage on AUTO. The most difficult part is install the chipset drivers. But you've done that much already right?

That will get you all the performance that really matters, any other tweaks are incredibly minor and won't be noticed so don't bother with them. Especially don't bother with disabling SMT. If you feel the need for more performance put a 5600X on that B450 Tomahawk and you'll get it.

Oh yeah...and CPU cooling. Definitely get something better than the Wraith cooler it came with.
 
Solution