Announcement

Collapse
No announcement yet.

"Intel Software Defined Silicon" Coming To Linux For Activating Extra Licensed Hardware Features

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • drakonas777
    replied
    Originally posted by birdie View Post

    Finally. Thank you. I've got nothing else to add. I sure hope all AMD lovers here have read this post and absorbed it.
    He literally said they did not raise the prices as much as people subjectively perceive due the lack of lower tier parts. But it's OK, I see your shaky-hands anger-filled condition really twists the way you think and see.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by birdie View Post
    Finally. Thank you. I've got nothing else to add. I sure hope all AMD lovers here have read this post and absorbed it.
    In the current shortage situation we don't have the opportunity to build "higher end parts in excess of demand" and the lower end parts get delayed

    First part of that sentence supply issues. Also note you were comparing chips of different levels with each other because AMD could not get supply to make the 5700X. Yes the 65W 8 core part is still missing out the 5000x lineup.

    Normally the lower binned parts are designed around yield expectations but the market demand for lower parts outstrips the fallout we get from production. As a consequence vendors will typically take any higher end parts they can fab in excess of demand and fuse them off to sell as a lower end part.

    Notice something this is why you get away with overclocking the lower end parts so much. But when you have a silicon supply nightmare like we currently do the low end parts are missing in a lot of cases because they don't have excess high end supply to bin down.

    Ryzen 7 3800X July 7, 2019 US $399 yes the 5800X $449 so up by 50 dollars. or a 12% increase. This is also meant to be the same watt part. Remember the 3800X AMD got the silicon for those with TSMC volume discount in full. The 5800X they did not get same TSMC volume discount.

    birdie majority of the low end problem with AMD cost is silicon supply as in unable get enough. Yes the performance increase a cross the board from 3800X to the 5800X was greater than the cost increase. The cost increase between the 3800X and 5800X is also mostly TSMC volume discount being clawed back then removed yes silicon supply.

    Remember Nvidia is now having to lift their GPU prices to cover samsung silicon prices on them again another silicon supply issue.

    Remember AMD has to make enough money so they have enough money give to TSMC to make their next batches. So the 5800X had to be priced as if it did not get any volume discount because TSMC was repeatably reducing discount this was the correct move.

    Intel for their CPUs is currently not under fab supply short fall. So should be a lot more competitive priced than what they are as well.

    Remember fabs are pay in advance most of the money(around 70%) and then pay final amount to get final delivery. Yes the 30% discount use to be you did big enough order that you get final delivery with no charge.
    Last edited by oiaohm; 01 October 2021, 05:56 PM.

    Leave a comment:


  • birdie
    replied
    Originally posted by bridgman View Post

    I would argue that limited production capacity is a third factor, at least in terms of the "price jump" that Birdie is talking about - lack of a 3700X replacement leading to comparing 5800X pricing against 3700X pricing.

    Normally the lower binned parts are designed around yield expectations but the market demand for lower parts outstrips the fallout we get from production. As a consequence vendors will typically take any higher end parts they can fab in excess of demand and fuse them off to sell as a lower end part.

    In the current shortage situation we don't have the opportunity to build "higher end parts in excess of demand" and the lower end parts get delayed as a consequence... and in the absence of a "5700X" to compare with a 3700X it appears we have hiked our prices more than we really did.
    Finally. Thank you. I've got nothing else to add. I sure hope all AMD lovers here have read this post and absorbed it.

    Leave a comment:


  • tildearrow
    replied
    Originally posted by birdie View Post
    And I would like to personally thank tilderarrow (a moderator of these forums) for continuing to allow people to insult me while I'm not allowed to use any words to describe anyone here because I'm gonna get banned.

    And please do check pal666 posts history. His every second post is an insult. What nice forums we have here.

    https://www.phoronix.com/forums/foru...53#post1282153



    An Intel "fanboy" has now owned two generations of AMD CPUs, 3700X and 5800X and owned RX 5600 XT a year ago but had to sell it because of buggy drivers and black screens.

    I'm out of this topic. Enough.
    And you should be thankful that you are still allowed to derail threads like that. Turning a conversation about Intel into NVIDIA and AMD.

    If "fanboy" offends you, then it's because you cannot admit it. Come on, you are an obvious Intel/NVIDIA fan. Does not mean you are not welcome, but we would be glad if you were to stop being so vocal and pushy:

    Originally posted by birdie View Post
    I guess if AMD literally fed you faeces, you'd eat them and say how tasty they are.
    A good example of fanboy behavior.
    Last edited by tildearrow; 01 October 2021, 02:05 PM.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by yump View Post
    You can already do this with the interface in /sys/devices/system/cpu/cpu*/online. If all the cores aren't stable with current BIOS settings, boot your kernel with maxcpus=1 and then turn on only the ones that work. You can also limit the clock frequency with cpupower frequency-set -u.
    Yes know those commands. But you can see it with a AMD epyc you have 64 core one and you do all those settings and it still does not end up equal performance as the 8 core version by a large magrin. To electrical bin in the fly you have intentionally added more power control in the power distribution in the silicon this is deeper than a sleep state. Remember the fact that you can do Maxcpu=1 and then wake cores up you want means the circuit for waking the cpu up when called on has to be sitting there eating power just in case at any time the cpu is called on to wake that cpu up and use it. With "Intel Software Define Silicon" only the controller of software define silicon has to stay wake so the cpu cores and other parts can just have their power fully pulled in particular modes.

    Done right there are advantages to Intel Software define silicon over the current cpu online and offline stuff that is putting cpus in sleep modes. Yes there are also going to be disadvantages it is noted in the Intel Software define silicon is that you can upload mode and find you need to reboot the system to switch out of it due to what was powered on or off that is will make the system unstable if you change it back the other way without power cycling. This is basically sleep mode on maxed out steroids yes it does make what it can do more powerful but that will also be a double sided sword at times.

    Sleep mode designs what we are use to with cpu_hotplug and the like what even done has to be reservable without rebooting the system. Binning controls its optional if it reservable without power cycling the cpu to in fact having to power cycle the complete motherboard as this could be changing how the chipset of the motherboard is power feeding the cpu.

    With the non reversible case while powered of some of the binning options this also explains why you would not expose these are general changeable options. Also that binning switches it may be possible to switch them in a combination that results in a cpu core being unstable unless loaded with the right microcode due to the fact cpu cannot by microcode attempt to use particular features when binned a particular way. Binning controls are a lot more low level. Turning cpus on and off is different to binning were you can be turning features on and off inside cpu cores as well. So you might have a total integer workload so you turn of the CPU floating point logic so that does not generate heat.

    Leave a comment:


  • drakonas777
    replied
    It is what it is. I understand the business side. I also understand the consumer side. Both have caveats. Yes, yes, "dirty anti-consumer practices" exist. They are called maximizing the profit and dates back to the beginning of capitalism (which I'm not against BTW). Irresponsible and borderline irrational consumption of the goods also exists. I mean do you really need to upgrade your f****ing 2080/Ti's, previous gen 8C16T CPUs, high end smartphones every single year? Do ya, really? Anyway.

    I'm not pro-AMD here and I'm not going to defend anything made by it. Basically I'm just irritated by birdie's takes - that's all. Actually, 5800X is not even worth talking about. Far more interesting examples would be 3900X and 3950X. (5900X and 5950X too at some extent). A lot of people don't realize, but these dual-chiplet models (SKUs is more accurate I guess) were originally planned to compete against Intel i7's in a <400USD segment. That's right, when AMD was planning these years ago, long before shortages and that bullshit mainstream i9 segment from team blue (which is the reason why R9s exist basically), it was assumed, that Intel is going to be competitive with new 10N+ uarchs and more cores, thus these dual chiplet SKUs were made and organizes economically (at that time) in such a way, that even at <~400 would be profitable (assuming no shortages or other radical changes in the market of course). Well, that is a lot more solid reason to possess anger against AMD (if one has psychological issues that is ) Difference is I'm not crying about that, or current AMD product segmentation/prices in almost every single AMD product related topic like birdie does.

    As I said - it is what it is. If Intel Royal Core design is real and on time, AMD might very well be a budget option again Or not. We will see.
    Last edited by drakonas777; 01 October 2021, 10:39 AM.

    Leave a comment:


  • yump
    replied
    Originally posted by oiaohm View Post
    There is also the problem that at times you could wish to prune a processor to get higher clock speeds as well. So these keys if done right could horrible work in reverse allow you to take a fully enabled single like with 64 cores then force disable it back to 8 cores so areas of silicon go power off to allow higher clock speeds with current level of cooling.

    [...]

    Like you buy a full functional chip you get the keys to bin it down to lower chips this could be really good for a development CPU.
    You can already do this with the interface in /sys/devices/system/cpu/cpu*/online. If all the cores aren't stable with current BIOS settings, boot your kernel with maxcpus=1 and then turn on only the ones that work. You can also limit the clock frequency with cpupower frequency-set -u.

    Leave a comment:


  • bridgman
    replied
    Originally posted by drakonas777 View Post
    PS. Birdie, the main reason for high AMD CPU prices is the lack of Intel competitiveness. (Second being increasingly expensive production).
    I would argue that limited production capacity is a third factor, at least in terms of the "price jump" that Birdie is talking about - lack of a 3700X replacement leading to comparing 5800X pricing against 3700X pricing.

    Normally the lower binned parts are designed around yield expectations but the market demand for lower parts outstrips the fallout we get from production. As a consequence vendors will typically take any higher end parts they can fab in excess of demand and fuse them off to sell as a lower end part.

    In the current shortage situation we don't have the opportunity to build "higher end parts in excess of demand" and the lower end parts get delayed as a consequence... and in the absence of a "5700X" to compare with a 3700X it appears we have hiked our prices more than we really did.
    Last edited by bridgman; 01 October 2021, 09:34 AM.

    Leave a comment:


  • drakonas777
    replied
    Birdie is offended by the fact that AMD is not a budget brand anymore. So he cherry picks and misrepresents the facts to cope with this situation, mostly to feel better. That's all there is.

    You can search his history. Attempts to prove that GSYNC is not a vendor lock-in and a comparison between i5 and i7 to show how Intel CPUs value got better (among other FUDish takes) really speak about his "not a fanboy speaker of truth, regularly attacked by AMD/FOSS fans" identity int this forum.

    BTW "I own X shit, ergo I can't possible be Y fanboy" is a such bullshit argument. Fanboyism is defined by the way you think and construct your arguments, not by things you own.

    PS. Birdie, the main reason for high AMD CPU prices is the lack of Intel competitiveness. (Second being increasingly expensive production). Yet you attack AMD, which legally and rightfully tries to grow from the not so distant semi-bankruptcy state in the past to the healthy company. You are defending and rationalizing that beloved x86 behemoth of yours. "AMD bad cause 5800X is not 299, the price I want" basically. All the point.
    Last edited by drakonas777; 01 October 2021, 03:02 AM.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by birdie View Post

    If Intel went by the same metric, you'd start throwing tons of shit at them in an instant.

    I was talking about 5800X and RX 6600 XT / RX480, you're now talking about something absolutely different. Top of the line 4-core Core i7 7700K was released for $350. 6-core Core i7 8700K cost ... $360 (basically the same) and in MT workloads it was almost 50% faster. And then 8-core Intel Core i7 10700K cost just $370, again massively faster than 8700K. By your reasoning Intel should have charged a lot more than they did, but they didn't. 5800X is barely 15% faster in MT workloads than 3700X, yet it cost 36% more.


    You have not looked at Intel 8 core or better offerings right. Yes the 7xxx series of Intel processors 8 core or better offering are price gouged by almost always 100% Yes that is take decent sales price and add 100%. Intel missing decent competition in area results in them putting on a 100% markup without fab issues or anything else this is just intel they need competition so their price remains decent. Yes the reason why 8700K is 360 dollar 6 core instead of being another 4 core part is competition from AMD.

    AMD has had a issue this time around with silicon supply from TSMC so cost gone up.

    Also 5800X under single thread/light threaded workloads is at times 36% faster than the 3700X. But Core i7 7700K vs a 8700K in single thread performance the difference is margin for error. The two extra cores in the 8700K help the MT a lot but the reason why there is two extra cores is the competition from AMD not going to have high priced 4 core parts more.

    The reality is MT thread workload with the 5800X vs 3700X is not where you see the big improvement with the 5800X yes 15% up lift when multi threading but the cpu is also not getting to clock max out to give that 15% When the 5800X is clock maxing out due to limited threads you are seeing a 30-40% uplift. If AMD could have bought the silicon in the 5800X for 2019 prices instead of 2020 prices their would have either been no uplift in price or a uplift in price less than MT performance gain. Even so AMD has still kept the price uplift at about the single thread performance uplift with the 5800x.


    Originally posted by birdie
    I did not cherry pick anything!! I used top CPUs for three consecutive generations: 7700K, 8700K and 10700K except 9700K
    No you have in fact cherry picked by error. You Cherry picked the CPU where Intel had decent competition at the time so you missed seeing how bad Intel is when they in fact don't have proper competition. Intel without proper competition is a pricing nightmare.

    AMD TDP is not based on watts the CPU uses.
    Thermal Design Power, or TDP, is a term used by AMD and Intel to refer in an extremely broad sense to the rate at which a CPU cooler must dissipate heat from the chip to allow it to perform as advertised. Sort of. Depending on the specific formula and product, this number often ends up a combination of science-y variables and voodoo mysticism, ultimately culminating in a figure that’s used to beat-down forum users over which processor has a lower advertised “TDP”. -

    Yes that is the AMD TDP maths horrible mess thing but we do in fact have the formulae but not always the input values that gave the TDP result.
    Yes intel TDP maths is no better.
    Intel publishes TDP figures for each of its CPUs, but the chip's actual behavior often doesn't conform to what this metric implies.


    Both AMD and Intel use totally screwy TDP maths. In fact none of your third party cooler makers in fact use the Intel or AMD TDP number instead use the thermal design documents from AMD and Intel that are not open to public. Yes not even AMD and Intel use their TDP numbers when designing or having design for them their own coolers. birdie TDP numbers are stack of screwed. AMD TDP number cannot always be used between AMD chips even due possibility that TDP forumalar can have had different key values like case temperature. Lot of ways AMD and intel TDP numbers are a marketing number not something real.

    Leave a comment:

Working...
X