havli
-
Posts
413 -
Joined
-
Last visited
-
Days Won
3
Content Type
Profiles
Forums
Events
Blogs
Posts posted by havli
-
-
Looking good, although some combinations are very hard to get running.
For example:
486 - is a pure pain to run some benches on... last time I tried it took 5 hours to install win 2k and in the end it didn't work anyway.
socket 4 - noone has it and even if there are few people, I really doubt these boards even can be overclocked (other than P60 @ 66) without modding PLL.
VIA s370 CPUs are so slow and useless that noone have them
3dmark 01 - requires MMX IIRC, so no socket 5. Also V2 single is limited to 800x600, while SLI can do 1024x768.
In fact all the 3D stages are so much CPU limited, they are not really 3D anymore. Maybe rebalance it a little? Like:
3DM99 = socket 5/7 non-K6 + Riva 128/ZX
3DM00 = Klamath + Rage 128 non-Pro
3DM00 = Katmai + Voodoo3
3DM01 = TBird + GF256
-
My suggestion:
3Dmark 2006 (or later?) On the socket7 (or another very weak processor, even possible with the frequency limitation). Without limiting the video card!
Or something like that.
I really doubt 06 would run on non-SSE CPU.
-
Well HEDT is explicitly banned, however server platforms are not. So I think as long as you avoid HEDT chipsets and lga 2011-3, LGA 2066, and TR4 it should technically be allowed. Actually that means 2p is technically allowed in the rules they made. We should probably have Websmile clarify on that even though we know his answer.
Server platforms are banned by default... and it has been discussed many times already.
-
My intention was to avoid K6... because this inevitably leads to >600 MHz K6-2+/3+. Which we already had in several competitions. On the other hand P MMX is always getting overlooked.
-
Not so long ago serious security bug was discovered. It took some time but now I can present the new version of X265 which implements fix to this vulnerability and also adds some other improvements. Here is the changelog for version 2.1.0:
1. Fixed security issue allowing score manipulation.
2. Skylake, Kaby Lake, Skylake-X no longer requires HPET when running Windows 8 / 10.
3. Open Hardware Monitor updated to version 0.8.0.2 Alpha, added support for Kaby Lake, Skylake-X (not tested), Ryzen (not tested)
4. CPU-Z upgraded to version 1.80.1
5. Fixed data files path saving and name suggesting
6. Added/fixed profiles for CPU Feature Override (15h gen1, 15h gen2/3, 15h gen4, Zen, Kaby Lake, Cofee Lake)
7. Added legacy mode for compatibility with Athlon, Pentium III and possibly other old CPUs (pre-SSE2)
8. Dropped support for batch testing
----------------
To put some comments to the changes:
1. The score measuring is no longer connected to system time. Moving time has no effect on the score.
2. If you run X265 on these platforms using Windows 8/10, Windows Server 2012 / 2016, HPET is no longer required. This exception only applies to Skylake, Skylake-X, Kaby Lake, Kaby Lake-X. Everything else still must have HPET active.
3. This should make LCC and ambient cooling working on CPU released after Skylake. At least i7-7700K works (tested myself), Ryzen and Skylake-X might work (according to the OHM release notes)... but I don't have these platforms at hand, so can't tell for sure.
4. Should provide better HW detection for Ryzen, Kaby Lake, Skylake-X. Not sure if it supports latest 12/14/16/18-core i9 though.
5. There was a bug in older versions - when path for saving datafiles contained space (like "c:\my benchmark results"), then it wasn't loaded properly. Now it should be fixed. Just beware - if you are saving files to a network drive, make sure you open it using explorer (or anything else) first - so it will become properly initialized and connected. Otherwise saved path in X265 (pointing to network drive) won't work.
6. Nothing major, just updated few profiles. I doubt anyone use this function anyway
7. I received a bug report concerning incompatibility of X265 v2.0.0 with AMD Athlon (K7) and Pentium III CPUs. Most likely the updated version of the encoder is no longer compatible with pre-SSE2 CPUs, while the old version worked fine. For this reason I've added a legacy mode option which launches old encoder build (1.7.x) which was used in X265 Benchmark v1.2.x. This works for any CPU but performance is much lower... so unless you are running very old PC, there is no reason to use it.
8. I guess noone has ever used it... so now it is gone.
This patch brings no performance change, only bugfixes and minor new features.
----------------------------------
Download link is here, feel free to test it:
-
Hi,
I assume you are trying to submit directly from the benchmark? This error message sounds like at that moment HWBOT servers were unreachable, or perhaps your Internet connection malfunctioned. Try to submit it again, it should work. Or another possibility is to save a datafile and submit it via web browser at HWBOT site.
-
The datafile looks like this (CPU related part):
As you can see, there is no field for codename or socket. As far as I know, when you submit the score it works exactly the same way as if you perform manual search on the main site. If you type 4800+, three records show up (in alphabetical order most likely)... and the first one is used for the submission.
Like I said in the previous post, there are two solutions. One is non-systemic workaround for current API which would take a lot of time to implement. And the second solution requires changes to the API itself (server side)... and this is beyond my power.
-
Thank you for the support
Actually - CPU-Z is used to detect all the HW and SW info shown in the main window. It has been like that since the very first version of x265.
The problem is - HWBOT API only accept one parameter to specify used CPU and that is the name. I could implement some kind of my own database for CPUs with conflicting names built into the x265 bench and since I know the codename (thanks to CPU-Z), then it would be possible to assemble the correct name for submit... in this case "Athlon 64 X2 4800+ (Toledo) ". Obviously this is not very efficient approach.
I have an idea how to do this properly, but it would require update to HWBOT API... and I'm not sure if this can be done.
-
With current system, I don't think it is possible to solve this. X265 reports CPU name, which is in this case "Athlon 64 X2 4800+" and HWBOT engine is trying to find the best match from the database. There is no way to tell which 4800+ it actually is. Anyway the manual edit should work (I've used it few times myself) and this issue is not that common anyway. There are just a few different CPUs with same name.
-
It is great to see another season of old school.
I'm sure there will be some very interesting rounds and stages. Maybe I can start with this proposal?
Platform: Pentium MMX + EDO RAM + any 3dfx Voodoo GPU (single or SLI)
Stage 1 = SuperPI 1M
Stage 2 = 3DMark 99
Stage 3 = Aida64 Memory Read
(optional) Stage 4 = CPU Frequency
3Dmark is CPU limited anyway, so Voodoo2 or Banshee should be more than enough, others are allowed just to extend the range of suitable HW. To make things more interesting, EDO memory must be used, no SDRAM. Feel free to use any board based on i430 series chipsets, VIA, SiS or others. Even Super 7 boards are fine as long as they have SIMM slots for EDO.
As for the server HW - perhaps some socket 603/4 action would be nice. Prestonia and Gallatin Xeons are cheap and easy to find. Boards are not so difficult to get either. Gallatin DP (socket 604) has 1 or 2 MB of L3 cache, Gallatin MP (socket 603) has 2 or 4 MB L3 cache... both are compatible with s604 boards and can be overclocked. Running something like SuperPI 32M could be fun.
-
Hard questions
I think at some point there was planned 16MB VSA-100 board, possibly it could have been called V4 4200 (but was canceled and never released, like V5 5000). The Daytona based V4 are called V4-2 4200 I believe. But I have no idea if that is oficial name or community given.
V4 4800 should be the never released 64MB / DVI / TV-out AGP VSA-100 board. And also 32MB / DVI / TV-OUT PCI VSA-100 board. IIRC TDP for V4 4500 is ~15W, so it should be very similar here, maybe 17-18W.
Some also say Rampage has 30 milion transistors, but who knows, anyway 18 milion seems to low for 4pp DX8 chip. VSA-100 is 2pp DX6/7 and has 14 milion. Die size is unknown as far as I know. Rampage chip should support only PS 1.1. Vertex shader (and TnL) was implemented in the Sage chip.
I think Spectre cards were planned +/- like this:
Spectre 1000 = 1x Rampage, 4pp, PS 1.1, 128-bit DDR, 32-64MB
Spectre 2000 = 1x Rampage, 1x Sage, 4pp, PS 1.1, VS, 128-bit DDR, 64MB
Spectre 3000 = 2x Rampage, 8pp, 1x Sage, PS 1.1, VS, 256-bit DDR, 128MB
-
That is true... unless stated otherwise, server HW is not allowed for competitions. Server HW (as far as I know) is considered anything called "Xeon, Opteron, Epyc". Not sure about GPUs - but most likely Quadro, FirePro, FireGL are not allowed as well.
-
The stage rules are too long mostly, and some people do not bother to read them, I just had this at division 1 as excellent example. Nontheless I think this is OK, I did this for all divisions but it got lost for pro oc obviously. On the server issue we will never be on same opinion, for me it is like using a truck or a traktor at a formula 1 or nascar race
Well, I didn't realize that - too long rules list wouldn't look very good either. The problem is (no offense) people are lazy to search for general rules page hidden somewhere on the main hwbot site (not even on the esports). The support menu -> rules link on esports is empty placeholder btw.
I can accept server HW is not allowed for comppetitions, ok - different game, different rules, no problem. Global points... well, most of the rankings are/will be dominated by Threadripper / i7 / i9... so whatever. But do not dare to touch HW points
:D I'm sure there are many people interested in the HW Masters league and for them benchmarking everything they can find is a must. And like half of the database is server HW... Also by banning server HW completely all the effort put into creating such a nice and complex database we have would be thrown out of the window. -
Maybe it would be simply easier to add "no server HW message" to the competition template... then it would be written for all stages of every comp = perfectly clear for everyone and no more repeating questions.
I know it is written in general rules, but obviously not everyone is familiar with them. Banning server HW completely (HW and GL ranking) would would kill most of the fun and motivation for some people (me included). Possibly it could have some negative consequences.
-
Yeah, out of few hundreds CPUs I have, this is the only one that could be considered highly above average.
-
Yeah, something like that would be nice to see. Maybe selectable chart of user points / HW points / global points / submissions count over time... etc. That is of course in case the data are actually available in the database.
-
I tried on Kaby-X. Change time cheat works for this platform also. Score is increased, but "elapsed time" stays the same. Which means the score will have an unproportional long elapsed time...
Using pause button does nothing for efficiency/unrealistic score, but inceases "elapsed time" by an amount which seems to correspond to how long you have paused benchmark. It can of course be used to run higher clocks, but where does the efficiency come from?
Thank you for the testing (I'm all day at work today, so I don't have access to the source code of x265 bench) So far it seems the fix could be rather simple - to use the same kind of time measurement for score as the one which indicates "elapsed time". This value is already inside the benchmark and most likely with precision good enough to calculate the score (the elapsed time is truncated to 1s precision for display purposes). Possibly there are other ways, which I will also consider, of course everything will be tested using at least windows 7/8/10 and several hardware platforms to make sure it works. Obviously using the x265 encoder's internal time (and thus) fps measuring function wasn't a good choice.
Using this alternative method of time measuring would however kill the purpose of pause button... as it would make the score much lower.
Btw - overkill results should be always slower than regular single instance run depending on the overkill mode. For example if 100 fps @ regular takes 20s, then 100 fps @ 2x overkill would be ~40s, 100 fps @ 3x overkill = ~60s... etc.
-
Looks like this is how we may be able to identify these results: $@39@`s HWBOT x265 Benchmark - 1080p score: 55.7 fps with a Core i7 6700K
Frames: 1128
Time elapsed: 22 seconds
FPS: 1128 / 22 = 51.27 != 55.70
//edit: this can vary a bit depending on the milliseconds (i.e. 22.499sec). But in general, we should be able to identify a bogus run by comparing run time and FPS output for X265.
HWBOT Prime will be another story however ...
Yes, it seems the elapsed time is correct even on the cheated x265 scores - most likely different method is used to measure this time and the actual score time (which is then translated to fps). I will investigate this as soon as I can. I hope it will be possible to fix this vulnerability.
I've started to work on update of the x265 benchmark few days ago, it seems there are more problems to solve that I originally thought... this one is extremely serious.
I'm sorry such a big issue made it to the live version. During development I've put a lot of effort to make this benchmark as secure as possible (there are many anti-tampering measures in place)... but maybe the most obvious security hole slipped through and been here for the whole time. This time I must do even more thorough testing before the next version goes live.
-
At least one of them does unlock
havli`s 3DMark Vantage - Performance score: 24297 marks with a Radeon HD 6950 Or GPU-Z claims it is unlocked. I've already bought it like this and didn't test how it performs with stock bios.Anyway 1GB models are indeed rare to unlock, so 2GB are better choice.
-
From the file size it seems HWBP is using png compression, so the file size can get rather big... especially when saving competition scores with complicated background (like this one). 3.8MB score uploads just fine here, while 6.8MB does not.... so the limit is somewhere in between.
Updated version sounds like a good idea if the developer is still active. In that case I can provide some hints what would be nice to fix. Nothing game-breaking, just few GUI-related things that are a little annoying. I've run into similar problems while developing x265 and they are solved there (I hope
). -
Of course it is about OC, skill and fun... but who decides what is fun to bench and what is not? People who bench their stuff of course, and whether it is server HW or not depends on their taste. WRs are dominated by servers and always has been (obviously, it is the best available x86 HW). Maybe in the past it wasn't that obviout because IIRC HWBOT only had one benchmark suitable for such machines - wPrime 1024M. Now there are cinebenches and other MT benchmarks where servers occupy WRs.... because simply they are the best. Industry is still here because they care the most about the shiny OC events and newsflashes - where only desktop HW is used (and it is perfectly ok)... but server HW still has the place on top, noone talks about that in media but it is there and always were. Since 2010 (when I joined) amount of members here increased from maybe 30000 to > 110000 now, and all the time servers were here. Like I said I think what is pushing people away from overclocking are not servers or the HW occupying top scores... it very well may be Intel's restrictive policy concerning overclocking lowend and mainstream CPUs. If all of them were unlocked, many people would appreciate that and started using them here.
Banning subzero was only hypothetical of course - no way it could be enforced.
If something sends hwbot to the graveyard, it will be technical issues, coding errors and such (there are bugs that hasn't been fixed for years)... not people and their preference of HW to bench on. Obviously (some) top OCers have everything gifted - how else they could post scores of ES CPUs at launch or even before? Noone except them has access to this kind of stuff. How about Gigabyte SOC LN2 boards, sometimes ES DDR4 sticks, etc? They do it as a job, of course they get all the equipment needed. I don't care how they started - they get stuff for free now.
Banning 1366 Xeon is non-deserved advantage for people refusing to see the wider picture. It is like banning Mercedes engines in F1 races, because other teams fear they could lose to them.
-
Yes, I remember the old global points system. It is much more fair today.
I can imagine those 5 people with unlimited server power. They would own all multicore WR, sure. But how different is that from lets say top 50 people in the Elite League? Some of them also have more or less unlimited resources.... which implies they would hold the WRs. Not with server HW, no... but with super binned $1700 10-core 6950X in the past and soon $2000 18-core 7980XE.
Is it really so much different? I don't think so - for mere mortal overclocker top spots are always out of reach and it doesn't matter which HW in on the first place at the moment. There are so many ranking you can compete in that everyone finds the one which fits his needs the most. But out of those 110000 people very few can hit the first place in their league, global ranking or WR.
In short - I believe noone of the "lower 100000 people" care what HW is currently dominating the rankings because they realize it is out of their reach anyway. They overclock and benchmark things because they like it, not because they wish to be the OC king. If you say server HW is killing all the fun and pushing regular people away because they can't compete.... I say let's ban LN2/DICE for the very same reason. Even sub-average LN2 score is far better than anyone can reach with air or water... and for this reason people are leaving. See how absurd this sounds? In fact people are not leaving even if they know they have no chance to get points/cups/etc - for example x265 i5 6600k ranking HWBOT x265 Benchmark - 1080p overclocking records @ HWBOT There are 30 scores of 0.1 points (not so long ago there were no points at all) with no chance of scoring any HW/global/WR points (it is i5) and yet people still bench it.
-
I still don't understand why so much hate for "server" CPUs. Isn't the whole point of HWBOT to attract as many people to our overclocking and benchmarking hobby as possible? Just let people decide what to bech, for regular submission at least. Removing all server HW would shake up all rankings a lot, even HW masters - which in my opinion is the most valuable ranking because it requirest most dedication to benchmarking.
not just some cheap 20 HW + 15 GL best scores + some comp points. Prices of i7 will never fall lower than Xeons - which is exactly the reason not to buy it. My budget is limited and naturally I will choose what is best price/perf/general usefulness. I don't like to throw money out of the window for no reason.
-
Of course I won't compete here... it is kinda difficult when I don't have the HW for it
Engineering samples are different issue (not available to buy officially) and I completely agree with current rules.
HWBOT may be consumer HW platform... and yet most WRs are held by server class CPUs.
Rampage III Extreme BCLK wall
in Intel CPU Overclocking
Posted · Edited by havli
Hi,
I'm quite new to X58... but I have some older scores with GA-EX58-UD3R, that board did around 215 MHz BCLK with more or less default settings, which I considered acceptable for "lowend" X58. Yesterday I bought R3E and expected something around 230 should be very easy and with the right settings even more Reference Frequency overclocking records @ HWBOT
Turns out I was wrong. I've spent two evenings on this already, studied all kinds of X58 OC guides and still can't get some decent BCLK.
The system is:
Xeon W3520 water cooled
R3E (BIOS 1502 - latest)
3x4 GB DDR3 dual-rank or 1x2 GB single-rank 9-9-9-24
Radeon HD 5500
single sata HDD running in IDE mode
Enermax 630W
win7 64
It seems this W3520 is not as bad as most of my CPUs
and it could actually run faster than 4.4 GHz... if the MB would cooperate. I can get 4378 MHz R15 stable (multiple times) running 21x208. havli`s Cinebench - R15 score: 702 cb with a Xeon W3520 After that I got freezing during R15 runs so I lowered CPU and NB multiplier by one notch to make sure CPU is not the limit. Still no matter what I do >210 is not stable. I can boot windows, wprime 32M on 4 threads finishes, on 8 threads crashes and of course R15 crashes within a second after rendering starts.
Tried raising PCI-E up to 120 MHz, changed GTX 465 to HD 5500 and also tried PCI Voodoo3, still the same. Tried HD 5500 in first slot and second slot -> same. Tried to switch RAM to single 2GB module... nothing. QPI slow mode also does nothing.
This CPU can do more I'm sure and also I have other 1366 Xeons with lower multi and those really need as high BCLK as possible. I'm kinda running out of ideas. Most likely it is some obvious mistake... which I don't see. If you have some tips how to get this thing going strong I would really appreciate it.