Strunkenbold
-
Posts
2207 -
Joined
-
Last visited
-
Days Won
39
Content Type
Profiles
Forums
Events
Blogs
Posts posted by Strunkenbold
-
-
-
-
On 12/27/2020 at 1:30 AM, Mr.Scott said:
I remember when benchmarking was synonymous to overclocking hardware. Now it swings toward hacking software more so than overclocking hardware.
Have to draw the line somewhere........and stick to it.
For me the line is when you change the way of the intended benchmark process / calculation. I remember PCMark05 was crazy, because all subtests were changed in a way that the calculations the CPU / GPU had to do were actually way fewer than intended by altering files making them basically free, processing wise, or using newer programs like Media Encoder or newer browser. Even good old LOD "tweak" and 3D Overrider was used. And actually we could discuss here if LOD and disabling Tess were ever a good idea because in this case, they absolutely change the way a calculation was intended. But the thing is, that the bot wasn't stringent on this. Changing graphics to a state where you can only see blur is okay but using media encoder or some kind of cache is not. For me its obvious that we lived for years in a grey area where people were confused whats wrong and whats right. Thankfully Alby is cleaning all this mess now.
And PCM05 relied too much on external files. When they developed the bench, the relied on a specific version of the browser or the media encoder and settings of Windows without checking them before the launch. I mean, what's the point of disabling settings in Internet Explorer and you get awarded by receiving a higher score? The concept of this bench was just plain wrong and it actually has to get wiped from the bot.
As _mat_ already said, benchmarks relying on external files could change their behavior just because you install update for program xyz and receive other vb runtimes and you just don't even recognize. But just because you use an dll file where some functions are just a little better optimized, the calculation itself stay the same, so for me thats legit. And when I look at the rankings, I see big differences even with very old results with the same clock speeds. And no, thats not because they use older wprime.
Its nice that he now implemented showing which dll file you used. Even better would be to have something similar for GPU-PI. Because there its also only necessary to put the opencl compiler dll in the folder. Installed driver doesn't matter.
-
3
-
-
Not to mention, benching with Benchmate under Win10 makes things a lot easier. You get all the tweaks and uploading results to the bot is one click as benchmate already fills everything for you.
-
1
-
-
The reason for my proposal were bugs with Rage 128 cards, basically GPU-Z always identified wrong specs. Its showing wrong bus width, means 32 bit vs. 16 bit and sometimes detected DDR Ram. While all cards have sdram or sgram. Also clock speeds weren't detected.
The difference can be seen here, although I would need to recheck with a new version of GPU-Z
https://hwbot.org/submission/2974796_strunkenbold_3dmark2001_se_rage_128_vr_(64bit)_792_marks
A VR cant have more than 64 bit or having DDR Ram. GPU-Z was simply completely bugged and ruined our rankings. Its hard to tell if they changed something in between or if if they were correct in early versions. Rankings look strange, results varying from 500 to 2000 points in 3DMark01 in the same category.
-
22 hours ago, _mat_ said:
I remember the time well, I nearly turned my back on HWBOT and needed to take a few months off.
Good you did not. I think you saved the bot, atleast for benching AMD.
The separation was tricky. Difference between old and new was a bit too much. However, half part of benching GPU-PI is finding the right opencl dll. I actually now found a very old one where a difference between 3.2 and 3.3 is not there and seem to be quite efficient. Scores are here:
https://hwbot.org/submission/4665246_strunkenbold_gpupi_for_cpu___1b_ryzen_9_3900x_1min_36sec_681ms
https://hwbot.org/submission/4665253_strunkenbold_gpupi_v3.3_for_cpu___1b_ryzen_9_3900x_1min_36sec_569mshttps://hwbot.org/submission/4665260_strunkenbold_gpupi_for_cpu___100m_ryzen_9_3900x_6sec_719ms
https://hwbot.org/submission/4665256_strunkenbold_gpupi_v3.3_for_cpu___100m_ryzen_9_3900x_6sec_663ms-
1
-
-
Danke, ist korrigiert.
-
On 1/26/2021 at 10:53 AM, MiladKherad said:
please add mb GA-EG41-S2H
thanks
-
Seems like you included a wrong screenshot. 3DMark result is also hidden.
-
Locked for further requests. We have a support section for this.
-
1
-
-
I tried dll from blockchain driver, which is very old, and Adrenaline 20.12 without modifications no luck so far.
-
55 minutes ago, Mr.Scott said:
People will fuck those up six ways from sunday. Gotta make it simpler.
And honestly, core model on those cards makes no difference from a benching standpoint.
For sure, thats probably also the reason Alex did no changes yet. And it might sound like a good idea to condense different GPU cores with little or no impact on OC in general for the sake of simplification of the db but I hope that we could find a general solution for the future where hardware gets matched to same hardware in the db, spec-wise, despite different names.
6 minutes ago, Ashley98 said:From research.
7800GS 16P G70/G71 (G71 is very rare)
7800GS 20P G70/71
7800GS 24P G71
Thes rest don't exist.
Oh and the unlocking part there is a very very rare chance they do , haven't seen it myself on any of my cards should be G71 cards only.
Maybe split them into 3 subs or have separate subs for the gainward cards as the original 7800GS spec was 16 pixel pipes 6 Vertex all of the other card makers either called there cards the proper names like 7900GS and 7950GT
The 24 pipe gainward is supposed to be rare anyway only 10000 ever produced or something so they shouldn't be a lot out there especially the special edition one I own with 500/700 clocks. It clocks high and smokes the proper 7950 GT
Then TPU db is wrong and needs fixing.
-
1
-
-
Im facing a little strange problem with GPU PI 3.3.3:
This doesnt complete even with stock settings. However version 3.2 has no problems. I also dont see whats wrong, since it states that "calculation finished successfully".
As there are zero Vega 64 cards with 3.3 32B ranking but 15 competitors in the 1B ranking, something looks wrong here. My Vega 56 got flashed to a Vega 64, so the problem might be name handling. But otherwise no clue, I tried different opencl dlls but that made no difference. Next is trying with the stock bios to confirm my theory.
-
9 hours ago, Leeghoofd said:
Programmer is onto a memory leak that he think is the root of this trouble. It slows down the Bot tremendously in under a week and creates these dreaded 50X errors and not Best submission bug
Fantastic news if the guy could fix this!
-
3 hours ago, yosarianilives said:
My understanding is for unlocking you just check unlocked cores, new categories are for differences out of the box.
Yes thats the way it should be done.
According TPU GPU-DB:
7800 GS (G70, 16 Shaders)
7800 GS (G70, 20 Shaders)
7800 GS (G70, 24 Shaders)
7800 GS (G71, 16 Shaders)
7800 GS (G71, 20 Shaders)
7800 GS (G71, 24 Shaders)outch! ?
-
On 10/8/2020 at 11:32 PM, _mat_ said:
Yes, that's very much possible. If I remember correctly from Frederick's short mail replies, the hardware has to be added in a special way to make it work for the submission API. So if the uploading doesn't work on the HWBOT side of things (like the server returns with an error), it might be necessary to let @Leeghoofd know.
Late reply to this but if it helps, yes all strings coming from external programs have to be matched manually to a specific hardware db entry. I usually do this once in a month but the first user with new hardware will always face a problem. And there is some hardware which is just cant be identified by name. Like "Radeon Graphics" or "Athlon XP", "Athlon 64 3200",... Those stuff get ignored. Dont know how you handle this. But I would prefer not to submit directly to the rankings but to give the user the possibility to enter his hardware manually on the submit page. There are other things like cooling, motherboard and ram which can be only added by editing.
I can only encourage all users with detection problems to start a new topic in the support section. Maybe you could alter the error message in those cases accordingly.
-
Had the same issue yesterday. Result didnt appeared in the ranking and keeps telling "its not the best submission". Very confusing for users. Should it detect when its not yet ranked, instead of saying its not the best one?
Apparently the bot needs extreme time to sync for GPUPI 3.2 CPU 1B CPU subs. All other GPUPI stuff werent a problem.
Also ranking is way off, recalculating didnt helped immediately, might be also a factor for the syncing prob.
-
10 hours ago, Mr.Scott said:
Problem with that is, they're all 7800 GS. It is up to the user to find the good ones.
Well, actually no. If specs are different, we usually create a new category. I remember that GeForce 9800 GT Ultimate debate. For example all Renoir IGP are "Radeon Graphics" but range from 320 to 512 Shaders, would it be fair to have it all in one category? On the other hand, this has been like this for so long, that while a split would be good for consistency of the database, it also may produce a rage between users relied on old behavior.
The thing is always to find the point where we overdoing this separation of hardware parts. Like different CPU revisions. Dividing something because one rev is C0 and one is E0 would be completely useless, because its usual development of a CPU over the time to get better and better. But a card which one vendor decided to put a higher tier GPU on a card with lower tier name is something different. The vendor made something unique and out of the specs nvidia intended for it. Making it in the end actually a new card outside of the usual naming scheme.
@Antinomy Your opinion?
-
1
-
1
-
-
On 11/2/2020 at 12:05 AM, klopcha said:
Guys, may be separated FX 5700 Ultra with DDR2 and DDR3.
Splitted categories and moved some results over. GDDR2 vs GDDR3 wasnt that much more speed it seems. Likely they had very low memory timings though, so that the newer chips couldnt shine.
-
Would be interesting if hwinfo can show us the s-spec everytime. In that case we could try to ask Franck if he implements something similar for CPU-Z.
-
On 1/11/2021 at 4:11 AM, klopcha said:
-
15 hours ago, Thor941 said:
mine has 3GB of memory instead of 6GB on all the other scores, does it change something ?
thx
Despite it has only half memory, rest of the specs are the same. So no, its the Mobile category I already mentioned.
We decided on some point to not create extra categories for new hardware with different mem sizes. IIRC even TNT2 categories got merged at some point. -
16 hours ago, Thor941 said:
AMD E2 3800
-
On 1/14/2021 at 9:33 PM, Thor941 said:
and please add AMD radeon HD8200 : https://www.techpowerup.com/gpuz/details/2z2af
thx
Whats the CPU ?
PLEASE ADD MOTHERBOARD THREAD:
in Support
Posted
This is a long standing problem. This list doesn't show enough items. Another way to show less is entering KT133 in chipset.