intel i7-9700k/i9-9900k reviews - disappointed

Discussion in 'Intel x86 CPUs and chipsets' started by Jazper, Apr 16, 2019 at 3:21 PM.

  1. Jazper

    Jazper Member

    Joined:
    Jul 28, 2001
    Messages:
    2,515
    Location:
    Melbourne Australia
    Ok maybe I'm just a die hard techie..

    A lot of overclocking commentators think that the best way to overclock the 9700k and the 9900k is to turn off power limits, and even VRM temperature limits, then let the cpu dictate the power consumption. Having had experience with the 9700, if you leave it to it's own devices it'll draw 210W+ straight off the bat. Short term this isn't a big deal, but longer term this will nuke VRMs.

    If you look at the power draw figures on Anandtech. You'll see that at stock, the 9700k is 124W and the 9900k hits a huge 168W. Intel themselves suggest using a 130W heatsink Which is nearly 40W underspec. The i7-8700 in comparison hits 145W or so.

    Ok, so what does this mean?

    If I was a motherboard manufacturer, I'd probably design my board around a 160W processor power limit. The reason being that very few people are going to have processors hitting 160W power, as most of them won't be running an $800 processor. This means I'm not going to have a lot of tolerance for higher power parts, especially with a z370 board not originally designed for them..

    Be careful how far and how hard you push your chips.. it may bite you.



     
    Last edited: Apr 16, 2019 at 6:12 PM
    nope likes this.
  2. nope

    nope Member

    Joined:
    May 8, 2012
    Messages:
    1,798
    Location:
    nsw meth coast
    So tldr incel are selling cpus that will melt vrm/boards? lerl
    Gonna be funny seeing boards die long term tonnes of bargain bin price used 9xxx series hitting the market with no boards
    2020 is gonna be absolute anal rape for intel.
     
  3. OP
    OP
    Jazper

    Jazper Member

    Joined:
    Jul 28, 2001
    Messages:
    2,515
    Location:
    Melbourne Australia
    The situation was the same with phase 1 of ryzen, motherboard wise at least
     
    Last edited: Apr 16, 2019 at 9:08 PM
    nope likes this.
  4. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,518
    Location:
    Griffin , Brisbane
    But AMD don't make motherboards
     
  5. nope

    nope Member

    Joined:
    May 8, 2012
    Messages:
    1,798
    Location:
    nsw meth coast
    x5xx series they actually do, x3xx and x4xx used other companies chipsets
     
  6. OP
    OP
    Jazper

    Jazper Member

    Joined:
    Jul 28, 2001
    Messages:
    2,515
    Location:
    Melbourne Australia
    Intel don't make motherboards either...
     
  7. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,518
    Location:
    Griffin , Brisbane
    You're talking about chipsets, I'm talking about Motherboards - and they licence Asmedia IP btw.

    Indeed (not any more), but that's irrelevant. What is it AMD did with Ryzen 1 exactly?
     
  8. OP
    OP
    Jazper

    Jazper Member

    Joined:
    Jul 28, 2001
    Messages:
    2,515
    Location:
    Melbourne Australia
    Phase 1 ryzen boards were in exactly the same situation - Motherboard makers undergunned the vrms and power delivery. Especially the mid range boards
     
    Last edited: Apr 16, 2019 at 9:22 PM
  9. PersianImmortal

    PersianImmortal Member

    Joined:
    Jun 27, 2001
    Messages:
    8,964
    Location:
    Canberra
    Not sure what the point of this thread is, a warning to people not to listen to silly people online? What you describe is not Intel's or the CPU's fault, nor a flaw or any sort of deficiency. It's to do with irresponsible people spreading nonsense. Solution: ignore the "overclocking commentators". Most don't know and/or don't care enough about what they're saying, furthermore, true overclockers are people who don't hold on to CPUs for a long period of time, so it's never made any sense to listen to an overclocker if you want longevity out of any CPU!

    Power usage on the i7 9700K is extremely good. If "left to their own devices", the 9700K will NEVER hit 210W, you stand no chance of destroying the VRM unless you deliberately defeat several built-in protections on the chip in order to push it way beyond spec, for no good reason. It already outperforms the competition in a range of areas with stock boost, and hardly breaks a sweat doing it. I wrote about this earlier - my 9700K runs all day long at an average of around 4.5GHz, sipping an average of ~12W. Under heavy load, with thermal limits in place, the chip will achieve its 4.9GHz boost while only putting out ~50W heat load. And I'm using a relatively cheap air cooler rated at 150W and barely better than stock air.

    Chart below shows the evidence to back up the statements above, obtained using HWMonitor on over 8 hours of mixed desktop usage:

    Untitled.jpg

    Only if you artificially stress it to 100% non-stop load using something like Prime95 will it hit its peak of 95WTDP, typically still operating at around 4.9GHz. Then -- if you're actually silly enough to imagine that removing all thermal limits and thermal protection system(s) on your CPU is perfectly fine AND you don't have sufficient cooling AND you do it for a very long time, then you risk damaging the motherboard. Neither CPU nor motherboard manufacturers can build their products to be 100% immune to human stupidity.
     
    Bion1c likes this.
  10. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,518
    Location:
    Griffin , Brisbane
    Well, That's been a problem for over a decade - Low end boards on any platform will always have underdone VRMs - what's important is that they run the supported CPU's at stock with no problems and no Throttling. For the most part this is no issue across all AM4 motherboards and CPU's, right up to the 2700x



    Sorry, this is just not the case.. In higher end boards, if 'left to its own devices' it will most certainly pull a lot more than 95w - not 210w, since that's just an upper limit, but what ever is required by the particular chip up to that limit to boost to full frequency, which as you can see from the Anandtech link above, and many others, is well beyond 95w even after extended periods. Just about all reviews were conducted like this - default settings on high end boards, so the performance numbers for the 9900k and 9700k from these reviews are only achievable if allowed to continue operating at the PL2 or 3 limit for the duration of the benchmark. This is also why there power consumption figures are so high.

    The only way to cap it to 95w is to intentionaly TDP limit in bios

    The 9700k isn't so bad, as you can see in the Anandtech review - but it has no Hyperthreading, so it's performance is quite a bit lower under MT workloads, and subsequently so is power consumption.

    At the end of the day what this means is if you pair one of these processors, (particularly the 9900k) with a lower end board, or some Itx boards with VRMs designed for 95w TDP, you won't get the performance expected - and by expected, I mean the numbers found in reviews. This is because either the BIOS will limit TDP, or rely on VRM thermal monitoring to restrict it.. The later resulting in a very uncontrolled throttling.


    I suggest you also have a read of this
    https://www.anandtech.com/show/13591/the-intel-core-i9-9900k-at-95w-fixing-the-power-for-sff

    Quite a sizable difference between the "default" performance, and when limited to the 95w tdp plastered on the box. Quite misleading.

    If you definition of heavy load is running super pi, sure.

    Can you post a HWinfo screenshot from the tail end of a proper MT workload - with bios defaults if you're going to make these outlandish claims?
     
    Jazper likes this.
  11. PersianImmortal

    PersianImmortal Member

    Joined:
    Jun 27, 2001
    Messages:
    8,964
    Location:
    Canberra
    1. The issue you refer to is largely a manufactured one. The review you link to still shows the 9900K and the 9700K beating the other CPUs in most every measure, even with this supposed limitation. The issue of the peak power load exceeding specced load (by about 25% mind you), only occurs under artificial 100% continuous load on every single core, for a sustained period, and that will never occur on a consumer desktop system outside of synthetic benchmarks. So in terms of longevity, sure, if you run benchmarks all day long then expect the VRM on cheap mobos to cook eventually, or you'll get a slightly lower score (safely) on higher-end mobos.

    2. We've done this whole "No HyperThreading" dance before :) The 9700K for example doesn't need HT to beat CPUs that do have it in actual, common usage scenarios, even in pure multi-core performance (e.g the real-world Geekbench benchmark, or Handbrake encoding). Most consumer AMD and Intel CPU buyers don't run graphics rendering farms or do high-end architectural design, so Cinebench for example is totally unrepresentative of their needs.

    3. Which CPU buyers are pairing the top two Intel CPUs with ultra-low-end motherboards? Should there be warnings on Ferraris not to put elcheapo tires on them?

    4. My definition of "under load" is a normal level of higher load during normal mixed usage over 8 hours. I also refer to "full load", which is running the fully multi-threaded Prime95 tool, and on that my CPU easily reaches 4.9GHz at 95W TDP before my meager cooling (primarily purchased for silence rather than overclocking), in a closed case, finds it difficult to keep the CPU cool, and it downclocks to around 4.2-4.3GHz. But can any AMD CPU run at anywhere near that level of performance at 95W and stock air? Nope, so the point is moot.
     
  12. shane41

    shane41 Member

    Joined:
    Nov 10, 2008
    Messages:
    6,079
    Location:
    on the edge
  13. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,518
    Location:
    Griffin , Brisbane
    So.. from the link

    Pov Ray
    7zip
    handbrake
    x264 Encoding


    Are all synthetic benchmarks? .

    Every single one of these showed large performance drops

    I'm sorry to inform you, but you seem well out of touch here. The benefit of Hyperthreading (and any SMT) is obvious in every one of the workloads I just quoted - from that reivew. They're not synthetic, they're not used by people in 'render farms' They used by anyone doing a bit of productivity, I personally use several Encoding, 7zip, Winrar and POV-ray. I can guarantee you i'm not an outlier.

    Despite this though, You then go on to say Cinebench is unrepresentative, but Geekbench is a "real world benchmark"? What the?.. it's the very definition of a synthetic benchmark. Cinebench technically is not.

    At the end of the day, If someone's buying an 8c16t processor - they want the extra MT performance it provides - Trying to suggest MT performance doesn't matter, to avoid the issue we're discussing is laughable.


    If a Motherboard has official support for a CPU, and provides the supporting functionality required - why should they instead choose some arbitrarily higher end motherboard? Or if they want an SFF system, and design it around a 95w tdp (which is the premise of the article more so than low end boards - but the issue is the same)

    Your car analogy falls over, because Ferrari's ship with Performance tyres.. If Intel were up front with there TDP, and stated what level of motherboard would be required to achieve the full performance, it would be a different matter.

    Sorry but your definiton is flakey. and you specifically said "Heavy load" . That's not heavy load under anyone elses definition but yours i'd imagine.

    Prove it.


    So your cooler is rated for less than 95w then? - What cooler is this?

    Yes.
     
  14. Butcher9_9

    Butcher9_9 Member

    Joined:
    Aug 5, 2006
    Messages:
    2,059
    Location:
    Perth , East Vic Park
    TDP =/ Power Draw. For Intel at least its just a rough guide on cooling requirements given a standard load . Its also just Very rough, a whole range of CPUs have the same TDP despite higher clocks/ more cores.

    As Persian has said, average/low end CPUs/Mobos are designed for the average Joe running standard work loads. Under standard conditions cooling and VRMs are never really an issue (unless you buy the worst case ). If you are looking to overclock and run synthetics all day long then get good and buy a board designed for that.
     
  15. OJR

    OJR Member

    Joined:
    Jan 19, 2013
    Messages:
    3,525
    For Intel it's a blatant lie.
     
  16. PersianImmortal

    PersianImmortal Member

    Joined:
    Jun 27, 2001
    Messages:
    8,964
    Location:
    Canberra
    No, I didn't deny that HT or SMT can boost performance, my actual words were:

    "The review you link to still shows the 9900K and the 9700K beating the other CPUs in most every measure, even with this supposed limitation."

    and

    "The 9700K for example doesn't need HT to beat CPUs that do have it in actual, common usage scenarios, even in pure multi-core performance."

    Again I think you're trying very hard to misinterpret or possibly misrepresent what I'm saying.

    Cinebench, like Prime95 is generally considered a stress test more than a benchmark - it puts an entirely unrealistic 100% continuous load on the CPU that no one outside of specialized users would ever see or need.

    Geekbench is synthetic, but is not a stress test; it tries to emulate real-world usage scenarios across single and multicore usage (in both of which the 9700K beats the 2700X).

    Yet, because Cinebench is one of the few benchmarks in which AMD's multicore/SMT is highlighted by a noticeably higher score in the multicore component, it is the go-to "benchmark" that people cite to prove the alleged superiority of AMD chips.

    Which would be accurate if you're buying your CPU either to (a) run synthetic stress tests and benchmarks on it 24/7 or (b) you do specialized content creation centered on rendering complex 3D objects. For actual usage scenarios such as gaming, encoding, archiving, etc., the 9700K has been proven to be a cool, efficient and often superior chip performance-wise to the AMD equivalent.

    My definitions, proof and plenty more details and data are contained in a lengthy article I wrote a couple of months ago on my site.

    While the AnandTech article from late November 2018 raises a valid concern, this is a non-issue that's being promoted certain quarters to perpetuate the myth of the "hot, expensive, energy-hungry" Intel CPUs versus the allegedly much better AMD chips. I debunked these myths in my article. Frankly, if my CPU peaks at 125W for a while before the Intel limiter kicks in and brings it back down to 95W, and it still performs well enough to beat the opposition, then it's a non issue. Virtually no-one goes CPU shopping with a specific TDP target in mind. In fact, it wasn't until a few years ago that TDP was measured so thoroughly, or considered as a significant variable, in reviews. What people generally look at is how hot the CPU can get, and what it takes to cool it. In my case, even during a summer heat wave, I got 4.9GHz performance at acceptable temps using a basic fan & heatsink, in a fully enclosed case, without having to switch off all of the CPU and mobo's protection systems.
     
    Last edited: Apr 17, 2019 at 2:41 PM
  17. HobartTas

    HobartTas Member

    Joined:
    Jun 22, 2006
    Messages:
    749
    I agree because I remember people would replace their 2.4 Ghz Q6600 105W with a 2.66 Ghz Q9450 95W and were surprised that the Q9450 ran hotter even though its power consumption was 10W lower but the explanation was simple because the latter CPU was basically a die-shrink and the reduction in CPU area size was greater than the reduction in power consumption and therefore the CPU emitted more heat relative to its size and it took more effort to get rid of it and a few people I knew had to replace their tower coolers with a water setup just for this reason.
     
  18. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,518
    Location:
    Griffin , Brisbane
    100% of electrical power is turned into heat. Simple as that. The only thing that varies is what sort of workloads defined TDP, and how long the processor can exceed TDP to take advantage of Cooling systems 'latency' .. that essentially, how much energy can be dumped into it before temperature limits are reached. All manufacturers do this to some extent. What's concering is long term exceeding TDP, which is what these CPU's will do by default.

    I'm not actually concerned with an AMD vs Intel debate here - even though you keep pushing for one.


    OK.. you just keep cherry picking and beating that horse if you like.

    Cinebench btw runs for about 30s on an 8c/16t CPU even the new one not much longer - Some stress test that one!


    The Only proof I see there, is a really old non AVX prime95 run @ 4.2-4.3

    your 4.9Ghz run is showing 139w package power

    Not an issue , because your CPU idles most of the life.? :)
     
  19. PersianImmortal

    PersianImmortal Member

    Joined:
    Jun 27, 2001
    Messages:
    8,964
    Location:
    Canberra
    That's precisely what makes it a great stress test. Both it and Prime95 can reveal an unstable overclock within a minute or two. A stress test isn't an endurance test, it simulates extreme stress to quickly bring out any latent instability. I don't want to run a 12 or 24 test every time I change a setting to know if my system is stable.

    There are two "modes" I use, everyday mode and performance mode, covered on Page 5. Basically, with everday mode, alll limits are in place and the CPU rarely exceeds 95W TDP, typically running at up to 75% load during normal desktop usage at 4.4-6 GHz and around 50W at most. In performance mode, which is what gamers or people who need max performance at all times should use, the TDP limit is removed but the other limits remain in place to keep the CPU safe. In that mode, I provide two examples:\

    1. An hour of general usage ranging from mild to full 100% load across all cores. This is basically "real world" usage, not synthetic benchmark/stress testing. The HWINfo below shows the results, I've highlighted the relevant bits in red:

    Hardcon19_28.jpg

    Basically, the CPU reaches the full 4.9GHz boost across all cores at 100% max CPU usage, while the CPU package power shows that at max usage, it peaks at 115W and around 70C temps (again, closed case, hot summers day, basic air cooling). I find that very impressive.


    2. This is the synthetic run you asked for, which I did at that time using the generally accepted Prime95 v26.6, and is not reflective of real-world usage at all; it is merely a 10-minute stress test run to ensure the CPU is stable. Again, relevant bits highlighted:

    Hardcon19_27.jpg

    Once again, the CPU reaches its full 4.9GHz and 100% max loads across all cores, and sustains them for the entire 10 minutes (hence average CPU usage is virtually identical to max CPU usage at ~4.9GHz). But in this artificial test, the CPU hits 139W package power max and around 80c.

    This is not something that anyone other than overclockers and people who run extremely specialized software will ever see on their systems. Scenario 1 above is the realistic one, showing that even when maxed out at100% using real-world programs, and with the TDP limit removed, the CPU barely breaches 115W, and still reaches 4.9GHz as advertised, able to be cooled by a standard air cooler.


    Remember, Intel advertises the boost on these CPUs as 4.9GHz on only one core at max performance. Specifically, as this Anandtech review points out: it will reach its advertised 4.9GHz maximum on 1 core; 4.8GHz boost on 2 cores when required; 4.7GHz when boosting on four cores; and a maximum boost of 4.6GHz on 5 or more cores.

    I'm getting 4.9GHz across all 8 cores, safely, with real-world TDP of 115W max. If I want to play it completely safe, with the TDP limit in place, I can average 4.5GHz at less than 95W, with bursts up to 4.9GHz on one or more cores when needed.

    Porn doesn't require a great deal of CPU power :)
     
    Last edited: Apr 17, 2019 at 11:28 PM
  20. demiurge3141

    demiurge3141 Member

    Joined:
    Aug 19, 2005
    Messages:
    1,374
    Location:
    Melbourne 3073
    I gave up after this:

    Yes, Intel gets a $36 cooler but add a $100 cooler to Ryzen which already came with good stock cooler. These are simply not arguments in good faith.
     

Share This Page