FA+
Shop
Semi tech youtuber rant.
a year ago
General
To be honest, anyone with half a functioning brain-cell should've realized they're more personalities and celebrities than actual tech people, so i'm also kinda to blame a bit.
Still, i have to say the launch of the new zen 5/ryzen 9000 series in my eyes did more to damage the big three tech youtuber channels(linus tech tips, gamers nexus, and jayztwocents) reputations at actually being legit than they were than the small uplift(at leas for the 9600x and 9700x which should've been labeled non x imho) for AMD.
Well to be fair linus tech tips more or less gave up the ghost on that not to long ago when he up and sold a start-up company's prototype literal all-in-one liquid cooler block(cooling both cpu and gpu on the same block for small form factor cases) in an auction rather than give it back to the company like requested.
Anyway, everyone knows first party benchmarks(ones provided by the company that's selling the product) are to be taken with a grain of salt. On the other hand they generally are within about 5-10% of actual numbers. So when a tech youtuber, *coughgamernexuscough* 'known' for deep diving issues. Ignores that something's not quite right.
His review copy of the 9600x was doa on the 600 series mb chipsets(these were made for the 7000 series zen 4) and a 9700x that unlike other reviewers had memory stability issues. Did he step back and say 'there appears to be a compatibility issue here, we'll provide reviews but will not consider them final reviews of the product till these issues are fixed in case they are hampering performance.'
No, no he did not. instead he went on to review it, counts the do 9600x against amd, and the poor performance of the 9700x as the final reviews. Then posts a second video defending his first review about them being a dumb value for their cost... By saying they'll always compare the launch price of the 'new product' in the price-performance-watt calculations against the 'current' price of the previous gen.
That's like comparing the price of a current video game, to the price they were 30 years ago, not taking into account inflation, and basing your opinion on if they're cheaper or more expensive than ever on that.
Linus tech tips just posted the numbers, called it a poor showing, and walked away, similar to jayztwocents.
Now a small part of me thinks this is more or less an olive branch to intel(or intel telling them, they all have received sponsorship's from intel in the past 'review zen5 poorly or not get review samples for new cpu's gpu's gamers nexus even getting to tour intel facilities), considering all three when pretty hard at intel over their oxidation issue killing 13 and 14th gen chips... which they did not fix. the microcode update is a placebo. software cannot fix physical hardware defects. It doesn't even do what is claimed, lowering the voltage on the cpu to extend it's life. smaller tech youtuber channels have gone over the microcode and shown that it actually increases the voltage on average.
But what i'm leaning to is they got a TON of attention and traffic covering said intel debacle, so when amd launched zen 5, and when on the surface it doesn't look good. they just up and ran with it.
Turns out there's multiple issues going on.
Considering AMD couldn't have been able to test all variants of 600 series mb's in the wild with the new cpu. It's no surprise that some am5 mb's. Remember this is a new platform, JUST for the 7000 series till now. Have memory compatibility issues with the new zen5 arch. either just not working or poor timing settings, hampering performance for zen5 when those same sticks are considered 'compatible' by the mb maker and work on zen4 just fine.
Also there appears to be a similar issue to when intel came out with the novel, for x86, BIG.little* topology. Windows scheduler didn't know how to handle the cpu. Would put a demanding process on the 'little' or E cores. ignoring the BIG or P cores. Or right out crash when faced with starting an intensive process, like a game. Well Zen5 also has something new to the x86 arch too. instead trying to predict a single branch in code execution, so the cpu can increase performance by working on the code, before the actual process gets to that point. It has 2. So the cpu looks at two predictive execution branches, rather than one. The problem is, the os schedulers need to know about this to take advantage of it. Ie feed the caches right.
Linux can, because the linux kernel has run on multiple types of cpu's through out the years, and many of them, like ARM and i THINK power pc, not sure though. Also branch predicted past just the single branch x86 used to. So it knows how to handle it. Sites like phronix(linux only benchmark site) and the tech youtuber wendell?(exact spelling escapes me atm) which threw in linux benches of the new chips.
Well they noticed a larger than normal gap in performance between linux and windows. Normally about 1-5% is expected due to more efficient; process scheduling, i/o handling, and lack of in built drm. in some cases as much as 10% in some artificial benches like for how well all the various web-browser languages run happened in these comparisons, meaning something is going on.
Not to mention further testing revealed another scheduler bug in windows that affects amd cpu's only. Something in windows scheduler hampers performance on zen cpu's if you don't run programs as an administrator. which btw is a bad practice to do. Never run things as administrator if you can help it. in some games, running it as administrator on zen cpus increased fps by 10-15, but around 5~fps was normal.
Then a couple of days ago, another bug was found. Unlike all the other non 3dvache(cpu's without the 3d at the end aka 5800x instead of x3d) cpus, all 9000 series cpu's have core parking. As in when not many things are running, or your running an intensive 'mostly' single threaded task like a game. the cpu is designed to park, or soft-off non used cores so power can be put to use on the ones running the task. This allows in theory to up performance in tasks such as games as it will allow the used cores to clock higher thus more performance.
But here's the catch, In windows, unlike linux. If a cpu is present in the system without vcache or an intel cpu that explicitly advertises it has this feature when windows is installed, and ONLY when it's installed. The windows scheduler will NEVER be aware of this feature, causing it to NOT work at all. Of course, hampering performance.
Guess what these big youtubers do to save time? If you said 'just move a drive with windows installed between machines with everything they need already installed, never newly installing an os on new cpus' you win.
*BIG.little refers to a heterogeneous type of cpu where two different arches exist on the same cpu. In intel now they have performance cores aka BIG. These are your 'normal' fat cpu cores, with all the bells and whistles. designed to guzzle power, spit out heat, and clock as high as possible. aka your standard x86 core. the little or efficiency core, is a completely different kind of arch. Starting off as the old Intel 'atom' cpu's(yes the ones that used to power netbooks and that intel almost a decade ago now tried to put in phones to compete with arm) which in turn were taking the old pentium-pro arch and updating it. thentuning it with the bare amount instructions needed to run modern code. these things are tuned to use as little power as possible and clock only high enough to do the task needed.
In theory the idea is neat and has been done on arm cpus for almost 2 decades at this point? the performance demanding tasks go to the BIG cores, for maximum performance, and that's ALL they do, while the os and all background tasks get shoved to the little cores.
Still, i have to say the launch of the new zen 5/ryzen 9000 series in my eyes did more to damage the big three tech youtuber channels(linus tech tips, gamers nexus, and jayztwocents) reputations at actually being legit than they were than the small uplift(at leas for the 9600x and 9700x which should've been labeled non x imho) for AMD.
Well to be fair linus tech tips more or less gave up the ghost on that not to long ago when he up and sold a start-up company's prototype literal all-in-one liquid cooler block(cooling both cpu and gpu on the same block for small form factor cases) in an auction rather than give it back to the company like requested.
Anyway, everyone knows first party benchmarks(ones provided by the company that's selling the product) are to be taken with a grain of salt. On the other hand they generally are within about 5-10% of actual numbers. So when a tech youtuber, *coughgamernexuscough* 'known' for deep diving issues. Ignores that something's not quite right.
His review copy of the 9600x was doa on the 600 series mb chipsets(these were made for the 7000 series zen 4) and a 9700x that unlike other reviewers had memory stability issues. Did he step back and say 'there appears to be a compatibility issue here, we'll provide reviews but will not consider them final reviews of the product till these issues are fixed in case they are hampering performance.'
No, no he did not. instead he went on to review it, counts the do 9600x against amd, and the poor performance of the 9700x as the final reviews. Then posts a second video defending his first review about them being a dumb value for their cost... By saying they'll always compare the launch price of the 'new product' in the price-performance-watt calculations against the 'current' price of the previous gen.
That's like comparing the price of a current video game, to the price they were 30 years ago, not taking into account inflation, and basing your opinion on if they're cheaper or more expensive than ever on that.
Linus tech tips just posted the numbers, called it a poor showing, and walked away, similar to jayztwocents.
Now a small part of me thinks this is more or less an olive branch to intel(or intel telling them, they all have received sponsorship's from intel in the past 'review zen5 poorly or not get review samples for new cpu's gpu's gamers nexus even getting to tour intel facilities), considering all three when pretty hard at intel over their oxidation issue killing 13 and 14th gen chips... which they did not fix. the microcode update is a placebo. software cannot fix physical hardware defects. It doesn't even do what is claimed, lowering the voltage on the cpu to extend it's life. smaller tech youtuber channels have gone over the microcode and shown that it actually increases the voltage on average.
But what i'm leaning to is they got a TON of attention and traffic covering said intel debacle, so when amd launched zen 5, and when on the surface it doesn't look good. they just up and ran with it.
Turns out there's multiple issues going on.
Considering AMD couldn't have been able to test all variants of 600 series mb's in the wild with the new cpu. It's no surprise that some am5 mb's. Remember this is a new platform, JUST for the 7000 series till now. Have memory compatibility issues with the new zen5 arch. either just not working or poor timing settings, hampering performance for zen5 when those same sticks are considered 'compatible' by the mb maker and work on zen4 just fine.
Also there appears to be a similar issue to when intel came out with the novel, for x86, BIG.little* topology. Windows scheduler didn't know how to handle the cpu. Would put a demanding process on the 'little' or E cores. ignoring the BIG or P cores. Or right out crash when faced with starting an intensive process, like a game. Well Zen5 also has something new to the x86 arch too. instead trying to predict a single branch in code execution, so the cpu can increase performance by working on the code, before the actual process gets to that point. It has 2. So the cpu looks at two predictive execution branches, rather than one. The problem is, the os schedulers need to know about this to take advantage of it. Ie feed the caches right.
Linux can, because the linux kernel has run on multiple types of cpu's through out the years, and many of them, like ARM and i THINK power pc, not sure though. Also branch predicted past just the single branch x86 used to. So it knows how to handle it. Sites like phronix(linux only benchmark site) and the tech youtuber wendell?(exact spelling escapes me atm) which threw in linux benches of the new chips.
Well they noticed a larger than normal gap in performance between linux and windows. Normally about 1-5% is expected due to more efficient; process scheduling, i/o handling, and lack of in built drm. in some cases as much as 10% in some artificial benches like for how well all the various web-browser languages run happened in these comparisons, meaning something is going on.
Not to mention further testing revealed another scheduler bug in windows that affects amd cpu's only. Something in windows scheduler hampers performance on zen cpu's if you don't run programs as an administrator. which btw is a bad practice to do. Never run things as administrator if you can help it. in some games, running it as administrator on zen cpus increased fps by 10-15, but around 5~fps was normal.
Then a couple of days ago, another bug was found. Unlike all the other non 3dvache(cpu's without the 3d at the end aka 5800x instead of x3d) cpus, all 9000 series cpu's have core parking. As in when not many things are running, or your running an intensive 'mostly' single threaded task like a game. the cpu is designed to park, or soft-off non used cores so power can be put to use on the ones running the task. This allows in theory to up performance in tasks such as games as it will allow the used cores to clock higher thus more performance.
But here's the catch, In windows, unlike linux. If a cpu is present in the system without vcache or an intel cpu that explicitly advertises it has this feature when windows is installed, and ONLY when it's installed. The windows scheduler will NEVER be aware of this feature, causing it to NOT work at all. Of course, hampering performance.
Guess what these big youtubers do to save time? If you said 'just move a drive with windows installed between machines with everything they need already installed, never newly installing an os on new cpus' you win.
*BIG.little refers to a heterogeneous type of cpu where two different arches exist on the same cpu. In intel now they have performance cores aka BIG. These are your 'normal' fat cpu cores, with all the bells and whistles. designed to guzzle power, spit out heat, and clock as high as possible. aka your standard x86 core. the little or efficiency core, is a completely different kind of arch. Starting off as the old Intel 'atom' cpu's(yes the ones that used to power netbooks and that intel almost a decade ago now tried to put in phones to compete with arm) which in turn were taking the old pentium-pro arch and updating it. thentuning it with the bare amount instructions needed to run modern code. these things are tuned to use as little power as possible and clock only high enough to do the task needed.
In theory the idea is neat and has been done on arm cpus for almost 2 decades at this point? the performance demanding tasks go to the BIG cores, for maximum performance, and that's ALL they do, while the os and all background tasks get shoved to the little cores.
then again, should've from the start tbh.
Finding and looking at the zen5 reviews that are unbaised, shows that while it didn't dethrone the 3dvcache chips in gaming. and one shouldn't have expected them to do so anyway. In many cases they come close to matching it in some cases, speaking of the arch and ipc improvements. On top of the fact at least when it comes to the higher end parts. the 9600x and 9700x are for the moment, if they don't release non x varrients or say a 9500 / 9500x low to low-mid range cpus. the 9950x and the 9900x are BEASTLY all round cpu's. whatever you throw at them other than gaming, they'll either top or near top the charts in the case of the 9950x or give you a good bang for the admittedly first adopter tax price, as in with the 9900x.
Fro me, i run gentoo linux. for the slight speed improvements and the customization of only have the features in programs i want. at the cost of everything having to compile on my machine. which is why i have 5950x atm. the 9950x near to doubles performance for all the other tasks i use this machine for besides gaming.
With gpu's i'm even a bit more tolerant. If a gpu is already going into the hundreds of fps on a certain title at the resolution you use, also well over the refresh rate of your monitor. (side note, os's REALLY should limit framerates to the refresh rates by default. this not vsync, vsync tells the gpu to wait till the monitor gives the okay for a new frame. frame limiting tells the gpu 'you shall not output more fps than this')
10 or so fps spreads are meaningless. so i consider those cards identical in performance. which does mean in many cases the 7900 xtx can, outside of ray tracing be considered equivalent to the 4090, which is still as of this post about $2,000 while the 7900xtx is less than $1000.