i suppose you have a point but honestly the value market for threaded performance is what amd is doing best right now. in most catagories intel hasnt seen much of a loss in threaded value. the real standout is the 8700k vs 9700k which you would easily get the 8700k if you wanted thread count. but its only one generation back and intel would be competing in a saturated market including with their own products.
I concur with everyone. 16 threads are probably not used as efficiently for games as they could be. Yet someone else makes a good point that 16 threads applies to more than games.
Personally, I've combined the two. If I'm stuck on a game on my 9900K, I press the windows key, fire up, chrome or firefox, and search for the answer, opening as many tabs as I need without any slowdown when I check back and forth between the game and the browser. Is the good speed because of the 16 threads or just because the 9900K is a fast chip? I don't know. Maybe someone can chime in on this?
the extra threads only matter when the 2 threads of a core can share the cache and execute in a way that actually runs faster than a single thread would without sharing its resources. its complicated but the extra threads pay off the most when alot of computation needs to be done at the same time all in the same way. video encoding is a good example since it requires alot of the same type of computation and can be split into many similar routines.
if you are running alot of programs that work like this then the threads can pay off giving nearly double the performance but games will always have some code that needs 1 thread taking its own core to complete the fastest. and much of what could be hyperthreaded often isnt due to the cost of programming. as far as running many programs it depends on how well they each hyperthread. each program will need its own core. but the programs that can hyperthread will take up less time on each core.
There are a few games that make use of cores and threads and scale well, but you really have to identify those and see if you're playing one of them and if it will be worth it in the end. However I would always prefer HT over no HT.
The whole peformance advantage of SMT is when 1 CPU is blocked due to memory stalls, which CPUs do a whole load of prediction to try to avoid happening as it costs 100's of cycles and is generally invisible at the OS level. Sometimes 2 threads on one 1 core can run slower, nearer to memory speeds, so saving power by interleaving memory access, rather than have a single turbo core sucking juice whilst busy waiting on memory accesses. Software developers aren't given time required to deliver quality and fix real bugs, so Intel segmenting the HT market for monetisation reasons, actually killed off mainstream benefits as hardly anyone writes code for 20% of the market
I used to do this on a single CPU, it seems in the PC world most journalists/readers forget that 99% of threads spend almost all of their real wall time idle, blocked. You DO NOT need a core or even SMT to have benefits from multi-programming which was originally invented to maximise usage of expensive mainframe CPUs. Games and highly efficient programs often operate on data flows, which maximises performance due to cache effects, whereas naive approaches like spawning 100's of threads, stall on cache misses. The massively parallel part of games is delegated to a specialised co-processor, it's the Amdahl's law remnants which are hard to parallelise which are focussing gamers on single-thread performance
It's always "better" to have a whole core, rather than merely the 5% extra registers and logic stuff that HT needs. But if you're trying to maximise processing from a die for performance per $ (or watts consumed) then HT shines, because it allows a CPU stalled on memory for 100's of cycles to do something useful. Intel's market segmentation has disincentivised software developers from writing the code that maximises the benefits of HT (like sibling threads pre-loading caches) because that work only benefits a part of the market part of the time, so is just NOT done, like boring non remunatarive pursuits like quality coding and bug fixing. HT is a feature that like 64bit addressing makes sense to software producers when > 90% of their target market supports it. Unlike optional math instructions that can be patched in transparently to isolated routines on the fly at runtime, it affects the whole architecture of a program.
The majority of Consumers don't even know what Ryzen is, heck they don't even know what AMD is. They only know that the processors in ALL the computers they have ever seen or bought are made by Intel and AT MOST they have a vague understanding that i7 is better that i3 because 7 is moar than 5. Whenever I'm talking to my work colleagues, all "engineers", about AMD Ryzen CPU's or graphic cards they look at me like I'm some sort of freak.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
135 Comments
View All Comments
29a - Tuesday, April 23, 2019 - link
People do a lot more with computers than play games.Opencg - Tuesday, April 23, 2019 - link
i suppose you have a point but honestly the value market for threaded performance is what amd is doing best right now. in most catagories intel hasnt seen much of a loss in threaded value. the real standout is the 8700k vs 9700k which you would easily get the 8700k if you wanted thread count. but its only one generation back and intel would be competing in a saturated market including with their own products.shusse82 - Tuesday, April 23, 2019 - link
I concur with everyone. 16 threads are probably not used as efficiently for games as they could be. Yet someone else makes a good point that 16 threads applies to more than games.Personally, I've combined the two. If I'm stuck on a game on my 9900K, I press the windows key, fire up, chrome or firefox, and search for the answer, opening as many tabs as I need without any slowdown when I check back and forth between the game and the browser.
Is the good speed because of the 16 threads or just because the 9900K is a fast chip?
I don't know. Maybe someone can chime in on this?
Opencg - Tuesday, April 23, 2019 - link
the extra threads only matter when the 2 threads of a core can share the cache and execute in a way that actually runs faster than a single thread would without sharing its resources. its complicated but the extra threads pay off the most when alot of computation needs to be done at the same time all in the same way. video encoding is a good example since it requires alot of the same type of computation and can be split into many similar routines.if you are running alot of programs that work like this then the threads can pay off giving nearly double the performance but games will always have some code that needs 1 thread taking its own core to complete the fastest. and much of what could be hyperthreaded often isnt due to the cost of programming. as far as running many programs it depends on how well they each hyperthread. each program will need its own core. but the programs that can hyperthread will take up less time on each core.
GlossGhost - Monday, May 6, 2019 - link
There are a few games that make use of cores and threads and scale well, but you really have to identify those and see if you're playing one of them and if it will be worth it in the end. However I would always prefer HT over no HT.RobATiOyP - Friday, May 10, 2019 - link
The whole peformance advantage of SMT is when 1 CPU is blocked due to memory stalls, which CPUs do a whole load of prediction to try to avoid happening as it costs 100's of cycles and is generally invisible at the OS level.Sometimes 2 threads on one 1 core can run slower, nearer to memory speeds, so saving power by interleaving memory access, rather than have a single turbo core sucking juice whilst busy waiting on memory accesses.
Software developers aren't given time required to deliver quality and fix real bugs, so Intel segmenting the HT market for monetisation reasons, actually killed off mainstream benefits as hardly anyone writes code for 20% of the market
RobATiOyP - Friday, May 10, 2019 - link
I used to do this on a single CPU, it seems in the PC world most journalists/readers forget that 99% of threads spend almost all of their real wall time idle, blocked. You DO NOT need a core or even SMT to have benefits from multi-programming which was originally invented to maximise usage of expensive mainframe CPUs.Games and highly efficient programs often operate on data flows, which maximises performance due to cache effects, whereas naive approaches like spawning 100's of threads, stall on cache misses. The massively parallel part of games is delegated to a specialised co-processor, it's the Amdahl's law remnants which are hard to parallelise which are focussing gamers on single-thread performance
RobATiOyP - Friday, May 10, 2019 - link
It's always "better" to have a whole core, rather than merely the 5% extra registers and logic stuff that HT needs. But if you're trying to maximise processing from a die for performance per $ (or watts consumed) then HT shines, because it allows a CPU stalled on memory for 100's of cycles to do something useful.Intel's market segmentation has disincentivised software developers from writing the code that maximises the benefits of HT (like sibling threads pre-loading caches) because that work only benefits a part of the market part of the time, so is just NOT done, like boring non remunatarive pursuits like quality coding and bug fixing.
HT is a feature that like 64bit addressing makes sense to software producers when > 90% of their target market supports it. Unlike optional math instructions that can be patched in transparently to isolated routines on the fly at runtime, it affects the whole architecture of a program.
240hzGamer - Tuesday, April 23, 2019 - link
Can you also tell me the lottery numbers? ThanksGastec - Friday, April 26, 2019 - link
The majority of Consumers don't even know what Ryzen is, heck they don't even know what AMD is. They only know that the processors in ALL the computers they have ever seen or bought are made by Intel and AT MOST they have a vague understanding that i7 is better that i3 because 7 is moar than 5.Whenever I'm talking to my work colleagues, all "engineers", about AMD Ryzen CPU's or graphic cards they look at me like I'm some sort of freak.