ajc9988

The issue of ability to verify is going to be there. It is like you are not understanding what the backwards compatibility issue is going to mean. This is already a headache, and my point is this is going to get worse. Your response is that this is only for me. That isn't what this is about, just one person. I'm trying to sound a warning when it is easier to work on and solve an upcoming issue. I wish I was a coder. I'm not. I'm just an attorney who enjoys overclocking as a hobby. Instead, what I am good at is seeing potential problems and trying to devise solutions before an event occurs. Here, it effects ALL AMD MODERN CPUs, not just me. When those CPUs may be able to beat Intel's scores on equivalent core and thread count, then verification of the accuracy of the scores from AMD CPUs becomes a larger issue for the entire community. What that means is if you don't want to deal with the validation headaches later, you start strategically planning for it now, while there is still enough time to have a fix done, tested, and fully implemented before the event occurs. It is called foresight.

To run an analysis, if current Zen chips (2700X) can regularly get 1925@4.3GHz, you have a standardized 27.98 points per thread per GHz. If the chips next year on 7nm are at 5GHz and achieve the same points per thread per GHz, you get about 2238 points in CB15 as being fairly regular. Intel's 8 core chip, which is based on 14nm++, using the 8700K for the comparative performance per thread, would have had to be clocked at 5.2GHz to get 2212 in the CB15 leak. That means it will be way closer, and that means this could be a large problem, especially if the die shrink brings other changes to help this exceed what people are accomplishing with the 2700X aside from just the frequency boost (such as latency changes, IMC changes, cache changes, etc.).

Considering M$ is the one who created the problem (why it works fine on Win 7 not on Win 8 or 10), and we know they have no intention of fixing their problem, it is a ticking time bomb of problems coming up next year. If it is the chipset, because of backwards compatibility, if you try saying by chip like it was said with Intel before Skylake, then you are going to have a lot of people submitting Zen 2 (Ryzen 3000 series) results without the issue being resolved, which is going to be a pain. Either getting GPUPI to license freely a cut down program that just looks at timer, windows version, and does a screenshot to data file or just requiring it being open during a screenshot can address this before it becomes a problem. With the cut down program, it could even identify or have a bench drop down to select, that way for submissions. Either way, I think much of not wanting to think about it comes from assumptions that bias toward Intel. Looking at trends in the market, Intel may not have the top chips next year, instead not coming back to top until 2020 or 2021. If that occurs, you could see a shift in platform submissions, making it harder to get ahead of than it is now is the point I'm making. Edit: To be clear, if the problem is in the chipset, you would have to add the mainboard tab in CPU-Z to Ryzen screenshots to verify which chipset is being used because AMD allows backwards compatibility. That means, even if resolved in a newer chipset, people that just replace their chips in the old boards may submit it without realizing that the chipset, not the chip, is the problem. At that point, if they think the APU benches are hell, etc., wait until that comes. With Intel, they market segmented with board partners having new chipsets for most chips. That made the clear line divide easier when this was a problem on Intel. Depending on where the timer is with the AMD chips, this is going to be a lot more difficult. This is why I'm trying to get people thinking of it now, not 9-10 months from now when Ryzen 3000 hits. Right now, there is time to take whatever actions need taken, whether adding the CPU-Z tab for mainboard for Ryzen as a requirement, adding GPUPI to identify the timer in screenshots, or coming up with a program similar to what GPUPI does to capture the needed info and screenshot for submission. When those chips drop, it may become more problematic if they do have 5GHz about stock or can outperform the upcoming 8-core Intel chips. That is when points and records shift, and when headaches start. Here is my analysis of TR2 32 core for the speculative CB15 benchmark: "[A] stock TR 1950X with stock mem is between 2900 and 3100, with all core on 3.7GHz with 2133 loose timings. That is 26.18points per thread per ghz. If you take that, then a perfect linear scaling of the 3100 score would give you 6904 for the CB15 score for 32 Cores at 4.12GHz. The 6399 score is 7.3% slower than perfect linearity. At 2900 we get 24.49 points per thread per GHz, which would scale to 6458 points. So ... this [may be] just a very non-tuned rig the reporter examined. That means it is losing 1%-7.7% score due to the extra dies having no IMC and having to always jump die, which is not outside of the realm of possibility. So ... the ram [could be] slow as **** and it isn't tuned, and we are seeing that, whereas a person that can tune it could toss another 1,000 points on the score, potentially (going from people getting in the 3400-3600 range on CB15 scores, and depending on base, with having double the cores, so just a doubling of the spread from 500 to the 1,000 point number)." Try running a similar estimate with Intel's chips. It suggests the 2200 score for the Intel 8-core was about 5.2GHz.

So, I'm going to address your points as presented. 1) Testing is needed, but I do not believe it is "excessive." This post already shows the problem, and it has already been shown that when HPET timer is used, the bug doesn't exist because it only effects two timers that windows uses and HPET is not one of them. http://hwbot.org/newsflash/2684_windows_10_affected_by_same_downclock_bug_like_windows_88.1_disallowed_for_now This is why benches that contain a timer identifier can be used on Windows 8 and 10, because it verifies that the timer being used by the system is not one that has been effected. Now, I just booted into Win 10 with a 1950X without HPET on, opened GPUPI 3.2, and it gave the unsafe timer message. I let it turn on HPET, reboot, and it read as on. I then went to CMD prompt, did bcdedit /set useplatformclock false, and then closed and opened GPUPI 3.2 again. It still read as HPET being on. Why? Because it requires a reboot to change the timer being used to turn on or off HPET. 2) The usage isn't limited. Because of drivers and some changes in the OS, even with HPET on, Win 10 can on some benches give higher scores than Win 7, whereas Win 7 can still beat Win 10 on others, just like with Intel chips. AMD is gaining more market share per year, and if their chips do reach the 5GHz and above on the GF 7nm, which is advertised that 5GHz is the target for HPC/Server chips, and 14nm targeted 3GHz, and met that, for server chips, we could see AMD grab a lot more market share in around April of next year. That means now is the time to deal with this situation, not when it arrives about 9 months from now. That also goes with a GF official saying that AMD should be able to hit around 5GHz next year. Gary Patton, CTO of GF: " Definitely. It is a big performance boost - we quoted around 40%. I don't know how that exactly will translate into frequency, but I would guess that it should be able to get up in the 5GHz range, I would expect." This effects every AMD chip currently on the market, X370, X470, X399, and the lower chipsets that allow for overclocking. All of those will share compatibility moving forward to 7nm. Until AMD does a redesign on chipset, which it may be Asmedia's fault to begin with which is their chipset supplier, and also supplies many Intel chipsets, which all chips before Skylake are effected by this on the Intel side as well, making the problem larger than you just said, it would make sense trying to get on top of this now. Also, I bring this up to have Christian Ney take a look at it and confirm the fix, just as he did for identifying the problem. As far as the community goes, we are members and contributors, but having specific people confirm it is what is needed for testing, along with identifying simple ways to verify, which I will identify in the next point. 3) The copy and paste comment on your part is malarkey. If that is the case, we need to throw out all scores not recorded with a video or a screen capture card. Is that what you are saying? Because copy and pasting theoretically can be done for any of those for screenshots. Meanwhile, I wasn't thinking a data file, although asking for GPUPI to allow a screenshot data file for other benches for Win 10 isn't a bad idea, or asking to borrow the part of their program that identifies clocks and can do the screenshot to data file and submit to GPUPI would be good as well. As to clocks, refer to my first point. Otherwise, the dangers are no higher than those seen normally in the submission process. Also, if they borrowed that part from GPUPI or asked for a custom program similar, it could be asked they add a part that throws up the version of Windows below the timer. 4) We have rules on what is needed for submissions. This is just part of that. It can be automated, it can allow for scores to be addressed, and fixes a larger, looming problem coming next year with Intel's hitting a wall and not having 10nm ready until an undetermined quarter in 2019, having 14nm++ 8-core topping at around 5.2 without extreme cooling, going against a Ryzen 7nm that is clocking 5GHz plus in multi-threaded benches, where on a per clock basis AMD's SMT outperforms HT, and these will be run on chipsets without the fix due to backwards compatibility. Meanwhile, you'll have TR2 and TR3 being able to go toe to toe or beat that Cascade-X offerings, which top at 28-cores and won't be available until early 2019, with no clue when Icelake will arrive, needless to say volume 10nm cannon, which means the problem is coming whether you like to admit it or not. Intel's former CEO, BK, even said that their goal on the server side is to limit AMD to 14-20% market share or less, which AMD was still around 1-2%, meaning Intel is conceding they are about to face huge market share losses to AMD. This is to be proactive, not reactive. 5) It is a problem Windows needs to fix. If the RTC is in the Asmedia chipset, then it CANNOT be fixed until a new chipset arrives or potentially an AM4+ and sTR4+ hits the market, which is speculative at best. I do not think the RTC is on the chips themselves, but welcome being corrected if someone can show me otherwise. This also solves the hell on the AMUs from AMD as well, due to compatibility issues with Win 7. So far, I have not seen this much thought being given to the issue by others. I addressed your concerns. If you can think of any other concerns, please, let me know as I do enjoy where this topic is headed and this may help with the entire issue as we all think through this together. Edit: Also, if you would like, you can try turning off HPET on the Intel platform you have and check if it still uses HPET before the system is rebooted if you would like. That part of the question should be platform agnostic, meaning anyone can test it.

So, I see a lot of bashing on AMD for this, but want to spell out some things on this topic to go into a little further depth and welcome being corrected with further information and sources. 1) RTC was traditionally in the motherboard or chipset (specifically the south bridge). It is one of many timers on the motherboard. 2) HPET was also traditionally integrated into the chipset or south bridge. 3) Some microprocessors incorporate the RTC on chip, but not all, as seen with ARM. 4) No one has specifically shown that the RTC is incorporated on the Zen SoC (exception: EPYC as the chipset is integrated, meaning that the HPET and RTC, if the RTC is on the chipset and not another component on the MB, are integrated and the CPU can be blamed), but I want more info on this (whitepaper or statement from AMD). 5) Windows 7 CAN read the RTC, whereas newer versions of the OS cannot, suggesting the problem is not the RTC, but rather it lies with the software that reads the RTC, which is the OS pointing to this being a M$ bull issue. 6) If the RTC, like the HPET, is on the MB, and specifically the chipset, then it is the chipset manufacturer that is to blame, and AMD so long as it is related to the chipset development, for not finding out the changes and implementations to avoid RTC since M$ shows little interest in correcting their OS, as always. 6.a) If the RTC is not on the chipset, but is on the MB, then MB vendors would share the blame in selection of the components related to the RTC. This would actually absolve AMD, in part, of the criticism. 7) There are known differences between the versions of Windows, with performance varying between 7, 8, and 10. 7.a) Since there is a difference, HPET, which does not seem to be effected like RTC, was implemented as a requirement on 8 and 10. 7.b) When HPET is turned on, it has been shown to cause latencies on some processes, but it levels the playing field to give accuracy. 7.c) HPET causes large slow downs with the Spectre and Meltdown patches, more disproportionately effecting Intel, although AMD is more effected by HPET compared to it being turned off and RTC being used. 7.d) Why, with all of these factors, does HWBot not have a program that can be loaded and is required for use with benches on win 10 for the CPUs effected, especially since it takes a reboot to switch HPET on/off? 7.e) Why are people here not making a bigger stink about M$'s role in the problem? 7.f) Instead of a separate program, why not make them include opening a bench, like GPUPI, which has the timer resolution at the top so that HPET can be confirmed in Win 10 for those submitting Win 10 benches with these CPUs? I could go on, but I think this might help the discussion along. @Leeghoofd I'd love your thoughts on this. Edit: This would be an example of what I'm talking about with having GPUPI only open to show the timer being used by the system (HPET on), which on every Ryzen is partially identifiable on Win 10 because it reads the bus speed as faster by almost 1MHz, but should be slightly slower than actually punching in 101MHz bus clock in the BIOS, but would be nice to have the HPET showing to remove all doubt. Just wanted to clarify 7.f.

I tried reporting this sometime back at the beginning of 2017. This is over a year old problem that has not been resolved. I've tried both with and without the "s". I gave up until trying it recently, in hopes that the changes made over the year would have resolved the issue. Nope.

How did you get it to submit the score? Every time I try, it says invalid verification URL. I have a score of 9161. https://www.3dmark.com/pcm7/1130485

How did you get it to submit? I got 9161 on this bench and it keep saying that it is an invalid URL.

Could you please put up a picture of your rig?

I won't be the one to report it, as that is a mighty fine score, but for future reference, after running it, you need to open CPU-Z at least two instances showing the CPU first tab and the memory in the second instance. It is also nice to open notepad and type in your handle. Finally, I don't know if it is true for Rookie or Novice league, but take a picture of your rig. Don't worry, mine looks horrible. But this shows what you are working with and whether it makes sense. So, hope this helps and good job!

Great score!

Nice score! Too bad it is invalid without the CPU-Z CPU and Memory tabs in it. Please resubmit! I don't mind being beat, but I do want people to follow the posting requirements...

Not if you use a 10,000 BTU portable AC under it! ;-) Sent from my SM-G900P using Tapatalk

Just a heads up, a 5960X is a high-end desktop (HEDT). You submitted this score to Div 1 Rd 1, but I think you should be competing/submitting to the Pro OC Round 1 which has no limit on chip used. BTW, nice score! Hit the x265 4k and submit to the Pro OC!

Nice score. Wondering if you can squeeze some more from your CPU and RAM. Don't know if you've played with those or your thermal overhead at 4.5. Just chiming in 'cause I saw your submission in 2017 Div 1 Rd 1. . . .

So, I need clarification: the rules say desktop parts only. But, the Clevo high end use desktop i7 quad core CPUs and notebook variants of the 1080 (1080N). Considering the CPU is a desktop and the scores are competitive (over 5GHz benches on 7700Ks), do they qualify for this competition?

You have now beat my desktop AGAIN! I really need to do some 5.2 benches (you beat my 5.1 benches)...

So, I have ran an excellent run in PCMark 7, but when trying to submit the score, it says I provided a bad URL. Here is the URL: NVIDIA GeForce GTX 980 Ti video card benchmark result - Intel Core i7-6700K,ASUSTeK COMPUTER INC. MAXIMUS VIII EXTREME Considering this would give me additional hardware points as it is in the top 10 submitted, I was wondering what could be done for submitting my score?

Note - just noticed it reads 16MS instead of the 160MS. Sorry for the confusion.

I have a push/pull config on a TT Water 3.0 Ultimate using Noctua iPPC 3000 fans. Temps hit 75C at 51/48 1.552V as seen in the picture.

I have a push/pull config on a TT Water 3.0 Ultimate using Noctua iPPC 3000 fans. Temps hit 75C at 51/48 1.552V as seen in the picture.

Behind you by 0.01! LOL! Coming for you (yet you are beating my desktop, kudos!)!

If I hadn't bought the last one from you, I'd pick this up. Instead, building a desktop and will wait till 7700K is binned for my next chip (most likely). Sent from my SM-G900P using Tapatalk

Does the defective temp sensor on the core trigger the thermal throttle on the CPU. Looking at throwing it in a Clevo P870DM3 (Sager and Eurocom sell this laptop rebranded, as does HIDevolution: Custom Gaming Laptop | Custom Workstation Laptop | Custom Built Laptop PC by HIDevolution )... Sent from my SM-G900P using Tapatalk

Shame on the one core or I might have grabbed it for a laptop...