Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - Bzuco

Pages: [1] 2 3 ... 8
1
Okay, I can't wait for your results 🙂.
Just some motivation video(play at 7:46s) what can be done with 7950x, all core 4.8GHz, 0.995V, 131W, ~65°c, 36k cinebench points without running monitoring tools. Plenty of room to run on 5.0 or maybe 5.2GHz with his cooling.

2
General CG Discussion / Re: Hosting VR tours
« on: 2024-08-22, 17:53:15 »
Webhosting with domain in your country(support for HTML, FTP access, 1/2/...GB space ) and krpano.com

3
So as you can see you are reaching 100°C, that's the maximum operating temperature for this processor...and that is with water cooling, 2150rpm radiator fans.
You can expand Core Clocks row to see what are the frequencies of 8 Performance cores and what are on 16 Efficiency cores.

304-309W is too high consumption thanks to high default voltage set by Intel. You should be able to keep it on more resonably values for your cooling solution and location where you live... so something max. 250W.

Sadly I do not have these new generation of processors, I am on older rocket lake 11700F cpu, but the principles how to undervolt are the same.

In bios you should be able to find something like:
P-Core Ratio Apply mode - All Core
P-Core Ratio - change Auto to value you want to have for Performance cores, e.g. "50" which means 50x100MHz  = 5GHz
The same for E cores, but E cores have maximum turbo frequencies just 4.4GHz, so you will set Auto to "44", because those are efficiency cores and you want use them on max.
Next find Long Duration Power Limit(W), Short Duration Power Limit(W) and set it to max. values, or at least over 250W. Long Duration Maintained(s) set to max value.
It was also adviced to disable Enhanced Turbo, because you will be using fixed frequencies and you don't wanna CPU to affect this.

Next step is set the voltage. For this purpose you can use Intel Extreme Tuning Utility and use slider to set negative voltage offset. You can start with values -50mVolts, -75mV, -100mV, ...and always remember what is the current value, e.g. 1.243V. As you will be decreasing the voltage value you will find the value when PC freezes or some BSOD will appears. Then you need to set that value one step back(to higher voltage). Then you should run some rendering tests to see if that voltage value is safe even after few seconds/minutes of rendering. Intel XTU has built-in benchmark which is utilizinx AVX instructions, which needs slightly higher voltage.
If everything will be stable, you can mark the voltage value for this 5.0GH frequency and you will see what are the resulted temperatures and power consumption and then you can decide, if you have reserves for higher frequencies or not.
When you will be satisfied, you can set the voltage value in bios. You should be able to find something like CPU Core Voltage Mode - set it from auto to Override Mode and set the resulted voltage, e.g. 1.162V.
In most of the undervolting videos it is advised to set only negative offset voltage, e.g. -0.1V. The problem is, that offset is universal for all frequencies and your undervolting will be not so efficient as if you will use fixed frequency for all cores and exact lowest voltage.

I am not sure if it is possible to set different voltage also for E-cores. Dependence between frequency and voltage is not linear. With increasing frequency the value of voltage is behaving exponencialy. Every +0.1GHz needs a lot more and more voltage. So those E-cores can run at voltage ~1.0V or maybe even less...so you can save a lot of power for Performance cores.

This way I found the lowest functional voltage for my 11700F CPU for several frequencies (3.6GHz, 3.8, 4.0, 4.2, 4.4), and I am using 4.4 in winter and only 4.0 in summer when outdoor temperatures are higher, otherwise I would be not able to cool enough my CPU with air cooler.

EDIT: Voltage values you can see in HWiNFO (Core VIDs) are slightly different then values you will set in bios. e.g. if you set 1.170V in bios, you will see ~1.194V in HWiNFO Core VIDs.

4
So roughly 6500€.
For that price I would rather go with two identical PC for studio:
AMD 9950x, 64/128GB RAM, 500GB SSD system+software, 2TB project/cache, rtx 4060ti 16GB(fully enough for fluent 3D viewport, displaying hires textures, GPU denoising) if there are no plans for GPU rendering in the future, air cooling with 2 fans, or 240mm water cooling if office need to be quiet.
Both PC connected to network 4TB SSD with textures and models library. 2.5Gbit/s(320MB/s) network, which is enough speed for loading textures for first render. After first render, everything is in RAM memory, so no pressure for network bandwidth.
The rest of the money can be used for other necessary software licenses.

But if you really want to stay only with one PC, then threadripper is fine. Not sure if you will be able to utilize more than 128GB RAM. About RTX4090, same opinion as TomG. ECC memory only if you have plans to keep PC several days without restarting and shutting down.

5
Those two CPUs will behave the same, because they are both on the same manufacturing process 0.010 micron "Intel 7", same count and type of cores, maximum frequencies does not matter much, because the only limit for modern processors is the temperature.

You could easily reach 13-14M points in corona bench if you would do undervolting for your CPUs in bios or in Intel Extreme Tuning Utility (Intel® XTU).
Undervolting is safe procedure of finding the lowest fully functional(without freezing PC) voltage for certain frequency. Result of undervolting is much lower temperature and power consumption at certain frequency. This gives you room for increasing all core frequencies. Without this you are running CPUs at not ideal conditions. It is like driving on highway 100km/h on 3. gear instead of 5. gear.

You can do screenshot of HWiNFO during benchmarking before end, so we would know exactly what are current frequencies of all cores, what is the actual temperature and power consumption in watts.
HWiNFO has plenty of rows, but if you hit delete key on the row, which you do not need to be monitored, then you can fit all useful information on one page like on my screenshot.
Or you can simply keep something rendering on background and after one minute start record video with mobile and scroll with mouse through all rows in HWiNFO, this will be much better.

For undervolting process you can invite some person who will show you in bios where to set voltage value and where set fixed all core frequencies, where to disable power limit restriction, e.t.c. It is worth to do if you will be rendering a lot in the future, otherwise you are overpaying what you bought :).
The same story applies to graphics cards :).

6
Operating systems does not matter, only the number of actively running processes and services. If you disable few running apps(near tray clock icon) some unneeded services, then you will see maybe 3-5% boost in score(depending on benchmark app).

You should monitor frequencies/voltage/temperature on both CPUs during cinebench in HWiNFO util to better understand why there is 5000cb points difference.

All processors last 10-15 years are running on much higher voltage then is needed, which is causing high temperatures and maybe 14900k is throttling a litte bit more than 13900k.

If you will see exactly same frequencies during test, then 5000cb diff can be caused also by memories. DDR4/DDR5 have different timings and different read/write/copy speeds. In bios you can set e.g. interval how often should be the data refreshed in memory chips. During refreshing operation, data cannot be read and write by any software. If you set this parameter to less often refresh data(still safe for data), then you can see immediately performance boost several % in all benchmarks.
Recently there were microcode updates for bioses fixing errors(when operating on highest freqv.)  for 13/14 gen. , so that could also affect the performance.

So there is not exact answer what is causing 5000cb difference until you at least monitor using HWiNFO app and check memory parameters like timings, data transfers,...

7
Gallery / Exterior project in cottage area
« on: 2024-08-14, 13:31:32 »
Visuals for this project has more practical purpose(realization, cad drawings for workers, counting used materials, parking area dimmensions according turning radius of owners car, ...etc.) than creating a super realistic visualization.
Every piece of 3D object has it's own exact place and I hope 🙏 it will have also in the realisation.
I chose the comparative type of presentation.

8
Gallery / Re: H Apartment
« on: 2024-08-14, 12:04:59 »
Detailed everywhere 👌.

9
Hardware / Re: New Computer Setup CPU 5950x vs 12900K
« on: 2022-01-20, 10:07:31 »
Undervolted 5950 @4.4 - 4.8GHz.
12900k can match 5950's performance only when enormously overclocked which is useless in every aspect.

10
Hardware / Re: Hardware optimization?
« on: 2021-12-23, 13:36:51 »
Just set fixed all core frequency and find lowest stable voltage for that frequency. Then you will get maximum from your CPU with your cooling solution.
Do not use voltage offset mode and single core turbo boost frequencies...it is not worth those extra 10-15% speed in single threaded tasks.
With fixed freqv. and fixed clocks it is much easier and more accurate to find stabile lowest voltage.

Test 10min. rendering and you will see how temperatures will rise. Try to keep it under 80°C. If you reach 90+°C, reduce the frequency by 100 mhz and again, find lowest voltage....or buy two tower cooler :)

11
[Max] I need help! / Re: Cube map seams
« on: 2021-05-15, 10:52:43 »
Try to avoid using denoising and sharpening/bluring in frame buffer... then no visible seams appears.
You can check seams in PS if you cut and paste one square of cubemap and place it to correspondent another square.
If visible seams persist then it is probably issue on Layama side during resampling process...

12
So now it is more clear. The main diference is, that I mostly do cloudy day renders without sun or additional interior light. That's why I came across this noise problem when refl./refr. override is enabled. This scenario is not so common and I am assuming that optimization in rendering process in corona 6/7 was targeted for more common scenario(sun + sky)...therefore I see masive noise increase in direct light  render element.
Can be this considered as a bug or a simply unwanted regression in visual noise level in corona 6/7?

For now I will stay with corona 5, I am occasional "designer" (just few renders per year) so this is not something critical for me. Should I drop some report in bug section, or is this topic enough for devs to take a more indepth look?
Thank you again. ;-)

13
Hi, thank you for reply.

Hmm where did you find the wrong scale? I checked dimensions on every object and they are same(utility - measure  = object properties - general tab.) My system unit setup is 1 Unit = 1milimeter.

I also created just box with fliped polygons with one hole as window and the same issue - after enabling reflection/refraction override  the noise has increased in corona6+. Left and right render both 30s.

I also deleted the ENU folder to be sure that some script is not causing this issue...

14
Hi, when reflection/refraction override(black/gray/white color or tonemapped jpeg) in scene settings is enabled, masive noise is apearing using corona 6/7 latest daily. Corona 5 without issue.
Noise is visible in Direct render element, Indirect is very clean. Increasing LSM has almost no affect.
I am using only corona sky(Hosek Wilkie) in scene environment as a light source.
Tested 32s render, adaptivity off, UHD cache, GIvsAA 4, LSM 4, MSI 4, High.clamp 0, adapt. light off, random sampler new improved. Corona5 with portals, for corona6/7 no portals and env.sampler changed to adaptive.
Renders in corona 7 identical with 6.
Scene in attachment.

Any workaround? Similar issue mentioned @cecofuli in this topic https://forum.corona-renderer.com/index.php?topic=27953.msg170860#msg170860
THX.

15
How about this system I've just spotted is a better one than the dual x5650 it's about £30 more expensive?

HP Z600 Workstation 4 x Xeon X5672  3.20Ghz 12 GB RAM
Compare benchmark results here https://corona-renderer.com/benchmark/results   and you can search for additional CPU information (number of cores, max turbo multicore frequency) here http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20X5672%20-%20AT80614005922AA.html

Pages: [1] 2 3 ... 8