Sorry, was talking about another video he made yesterday regarding the same topic, where he showed 7800x 3d dropping to 60 fps in jedi survivor and therefore concluding that we need faster CPUs rather than just a 2-5% increase. My point was - you don't need faster CPUs, just tune your damn memory
stuff and nonesense
Honorable
- Mar 10, 2020
- 766
- 534
- 11,790
It’s ok for you to suggest an enthusiast might tune their memory. Normal mortals will be loathe to go beyond turning on XMP, EXPO. It isn’t reasonable for you to expect an inexperienced person buying a Dell to do more.
Box shifters like, for example, Dell should ship their products set to decent optimised defaults. Problem comes with a BIOS update, how do you retain the performance options? The user will see a marked drop off in performance after an update unless there is an easy way to restore the config.
Oh that's a pet peeve of mine. Mobo companies should make sure that saved profiles work after a bios update.
gggplaya
Splendid
- Jan 27, 2011
- 2,805
- 496
- 28,890
That's an undervolt, not underclock. It's undervolted and overclock to sustain an all core clock sustained over 5.5ghz. You would need to turn off all the Ecores to keep the thermals down low enough to sustain 5,7ghz P-core clocks without a good cooling loop. The person in the video you referenced has a full large custom cooling loop. According to TechPowerUP, the all core clock when e-cores are included is about 5 ghz in their testing. Like I said, Star Citizen scales well, even beyond 8 cores, so don't compare it to other games that only scale below 8 cores and will only use the P-cores. At motherboard defaults, the 13900k and 14900k are behind the 7800x3D in star citizen, they don't run that well unless you tune them. At defaults, the 7800x3D wins.
I played The Last of Us and it's not cpu limited. I have a 4080 super and it was GPU limited if I remember correctly. Perhaps it's CPU limited with a 4090 at 1080p, but on my 1440p ultrawide with a 4080 super it was not CPU limited. It's certainly GPU bottlenecked in this video with both the 4080 and 4090:
Also, that's incorrect about the 7800x3D. It's very efficient, runs at 60-65c with a 360mm AIO at full load full fan speed. Most people run it a little hotter and lower fan speeds. Mine runs <60c while gaming, however I do have a custom 360mm loop.
Next time there's a FREE FLY week on star citizen, you should try it for yourself. The next free fly is likely in November.
There are so many thing absolutely wrong with this post I don't even feel like correcting you. Sure bud, whatever you say.
thestryker
Judicious
- Apr 19, 2016
- 5,876
- 4,061
- 36,690
Nah, you need a custom loop and turn off the ecores to play games at 5.5 ghz 😆
stuff and nonesense
Honorable
- Mar 10, 2020
- 766
- 534
- 11,790
Was talking about this video he posted.
Yes game is heavy for no reason, but I've seen a 7800x 3d hitting a constant 90 fps (instead of drops to ~30 like in the video) just with ram tinkering. A faster gaming CPU will not do anything for these games, even if the 9800x 3d was say 30% faster than the 7800x 3d, that will take you from the lows of 30 fps to 40 fps. That's nothing. Ram tinkering is what will "save" you from these crappy optimized games. Or - not playing them at all I guess.
stuff and nonesense
Honorable
- Mar 10, 2020
- 766
- 534
- 11,790
Every part of a pc system has the possibility of being a bottleneck. Every system has bottlenecks.
There is nothing new in that vid. If you go back to the Athlon, it came on motherboards that inintially supported SDR. This bottleneck was soon shifted with DDR support.
The bottlenecks are often application specific, it depends on where the major processing load is within the pipeline and how it can be mitigated.
Hypothetically: super fast ram feeding a 2700x and a 4090 would do nothing good. Replace the 2700x with a 9900x and the 4090 with a gtx8800… your system is still bad.
Balancing your components so that the throughput for each allows each to perform close to their limits in the broadest set of applications is the target (wrt games) other applications don’t care about graphics… others are offloading processing to the gpu which changes the dynamic.
Work out what you want to do, spec for it, buy it enjoy it, please don’t try to score points.
Remembering my faithful beloved secret love of Core 2 Quad Q8400 literally played everything with that beast of a CPU back then with a GTX 480. Plus video editing on Windows Movie Maker.
My first love affair was a Pentium 4 heating my entire room in winter seasons. Lord of The Rings Battle of Middle Earth NOW that was a good time.
Sound of "cooling down" spikes were coming out from my PC case back then, like something sudden of peak. Then I learned GTX 480 had a issue with overheating.
My first love affair was a Pentium 4 heating my entire room in winter seasons. Lord of The Rings Battle of Middle Earth NOW that was a good time.
Sound of "cooling down" spikes were coming out from my PC case back then, like something sudden of peak. Then I learned GTX 480 had a issue with overheating.
-Fran-
Glorious
- May 3, 2007
- 9,557
- 3,106
- 47,640
I don't recall linking this video... Are you sure it was me?
Also, I took a look at my posting history and I could not find it, so please find it for me?
Regards.
Didn't say you did. I said I watched it a couple of days ago and I thought that's the one you linked.
-Fran-
Glorious
- May 3, 2007
- 9,557
- 3,106
- 47,640
Where is the implied doubt in "Was talking about this video he posted"? 🤔
You know what. I don't care anymore xD
Regards.
tamalero
Distinguished
- Oct 25, 2006
- 1,331
- 333
- 19,670
I apologize, I got confused.
As I was thinking Ryzen 1000 series vs Ryzen series 2000.
tamalero
Distinguished
- Oct 25, 2006
- 1,331
- 333
- 19,670
All the progress of the PS4 emulator has been to make Bloodborne work.
They have archieved some incredible progress in that game in the past months.
mitch074
Splendid
- Mar 17, 2006
- 3,015
- 578
- 22,040
You're the kind of guy who would have bought a Pentium 4 back in the day... A 2.8 GHz P4 being shown up by a 2.4 GHz Athlon would have caused you an aneurysm.
"Who cares about clock for clock?" People who care about efficiency. If a 3 GHz processor has 20% higher IPC than another 3 GHz processor, then the latter would have to run at 3.6 GHz to compensate. That means more power draw (typically the system would go from 45W to 65W), thus more cooling needed, also less autonomy.
Yeah, most people have a laptop nowadays ! If you can get the same performance out of a laptop as you'd get out of a desktop without the hassle of lugging around 500g of cooler and 2 Kg of battery, why not do it ?
If you cared about efficiency then you'd end up measuring the efficiency. Obviously, you do not. Who in their right mind cares about efficiency and instead of ..... measuring efficiency, they just look at clockspeeds?
Okay, the 12900k has lower clockspeeds than the 7950x. According to your logic then, it's more efficient. Okay bud, whatever you say.
mitch074
Splendid
- Mar 17, 2006
- 3,015
- 578
- 22,040
Nope, I'm talking performance per clock cycle - because I care. Your own words. Toodles !
If you cared about efficiency then youd measure efficiency. Clockspeeds are irrelevant. I just proved it by the comparison
stuff and nonesense
Honorable
- Mar 10, 2020
- 766
- 534
- 11,790
Efficiency is given by the energy used to complete a task.
Read the section titled dynamic power dissipation. TLDR: 2 factors directly change the power consumption, the supply voltage for a transistor and its operating frequency.
Reducing the voltage supply in order to reduce power to the IC introduces a delay, a latency in the speed at which the transistors within the IC can switch.
So, consider the clock is 2t and active on the switch to supply voltage. The transistor output needs to switch to the correct logic level before the next trigger action to be useful, it needs to present its output to the next transistor input or to the IC output within the clock cycle. Designing for this is really complicated.
A low supply voltage reduces the ability of the transistor to do this as the transistor isn’t driven as hard as a transistor at a higher voltage. It takes longer to get to logic 1.
This is known as “slew rate”. Look it up.
FYI, overclockers increase the supply voltage to their chips to shorten the slew rate time. Driving too hard causes the wave form to “ring”, this causes different problems.
Designers take both these factors into account and work in a neutral area, one that is fast enough and doesn’t overdrive the chip. Slew rate is added to the ring time by the circuit designers to come up with a parameter “settling time”. Look it up.
Going to Mitch’s example of an IC that does an amount of work at 3GHz with an IPC of 1.2 and a second that performs the same task at 3.6GHz with an IPC of 1. The article shows that everything else being equal the IC that runs at 3GHz will consume less power. It will be more efficient.
Assuming everything performs in a linear manner, increasing the hypothetical 3GHz chip frequency to 3.6GHz the output would be the equivalent of the 3.6GHz chip @4.32GHz
Power consumption increases with frequency, monitor with HWinfo, lock your processor frequency, to for sake of argument 4 GHz in bios. Run cinebench singe thread for 10 minutes and look at HWinfo, CPU power numbers…. Then go back to bios, let the processor run as intended with dynamic boosting… run cinebench single threaded again for 10 minutes while monitoring with HWinfo… look at the power use.
Power consumption increases with voltage, check this with HWinfo, reduce your voltages in bios.. look at the numbers in HWinfo… increase your voltages in bios, rinse and repeat.
The final part, as transistors are pushed toward their limits the source/drain/gate leakages become more problematic, leakages cause heat, they need to be compensated for by increasing voltages.. causing more heat. Leakage is always there to some degree but its effects are problematic as you try to drive a transistor at its limit. As stated earlier, designers try to run their ICs in a neutral region where changes are linear. As you push to the upper limits changes require substantially more power for small frequency gains. Power consumption in this use case is very high… LN2 cooling is your friend at the extremes.
Read the section titled dynamic power dissipation. TLDR: 2 factors directly change the power consumption, the supply voltage for a transistor and its operating frequency.
Reducing the voltage supply in order to reduce power to the IC introduces a delay, a latency in the speed at which the transistors within the IC can switch.
So, consider the clock is 2t and active on the switch to supply voltage. The transistor output needs to switch to the correct logic level before the next trigger action to be useful, it needs to present its output to the next transistor input or to the IC output within the clock cycle. Designing for this is really complicated.
A low supply voltage reduces the ability of the transistor to do this as the transistor isn’t driven as hard as a transistor at a higher voltage. It takes longer to get to logic 1.
This is known as “slew rate”. Look it up.
FYI, overclockers increase the supply voltage to their chips to shorten the slew rate time. Driving too hard causes the wave form to “ring”, this causes different problems.
Designers take both these factors into account and work in a neutral area, one that is fast enough and doesn’t overdrive the chip. Slew rate is added to the ring time by the circuit designers to come up with a parameter “settling time”. Look it up.
Going to Mitch’s example of an IC that does an amount of work at 3GHz with an IPC of 1.2 and a second that performs the same task at 3.6GHz with an IPC of 1. The article shows that everything else being equal the IC that runs at 3GHz will consume less power. It will be more efficient.
Assuming everything performs in a linear manner, increasing the hypothetical 3GHz chip frequency to 3.6GHz the output would be the equivalent of the 3.6GHz chip @4.32GHz
Power consumption increases with frequency, monitor with HWinfo, lock your processor frequency, to for sake of argument 4 GHz in bios. Run cinebench singe thread for 10 minutes and look at HWinfo, CPU power numbers…. Then go back to bios, let the processor run as intended with dynamic boosting… run cinebench single threaded again for 10 minutes while monitoring with HWinfo… look at the power use.
Power consumption increases with voltage, check this with HWinfo, reduce your voltages in bios.. look at the numbers in HWinfo… increase your voltages in bios, rinse and repeat.
The final part, as transistors are pushed toward their limits the source/drain/gate leakages become more problematic, leakages cause heat, they need to be compensated for by increasing voltages.. causing more heat. Leakage is always there to some degree but its effects are problematic as you try to drive a transistor at its limit. As stated earlier, designers try to run their ICs in a neutral region where changes are linear. As you push to the upper limits changes require substantially more power for small frequency gains. Power consumption in this use case is very high… LN2 cooling is your friend at the extremes.
Last edited:
Are there cpus that due to their architecture can hit higher clockspeeds than other cpus with less voltage? Eg. Zen 4 can hit ~2ghz extra clocks at similar voltages to zen 1. Therefore a clockspeed comparison does not help you to determine efficiency. That's all im saying.
mitch074
Splendid
- Mar 17, 2006
- 3,015
- 578
- 22,040
The engraving process also has a say on efficiency, and many other factors. Zen(+) is a uarch made for 12-14 nm, it's not really made for 7nm and lower. Reduce the process, reduce the voltage needed for transistor switching speed. However, if you compare Intel 7 to TSMC 7-6 nm, for comparable uarchs like Core 12 ans Zen 4, Zen 4 clocks lower but has similar performance than Core i 12 - consuming less power for a given task. If you reduce Core i 12 power to match Zen4, its clock speed lowers and performance decreases - it is, all in all, less efficient.
"But Intel 7 is actually 10nm". Yeah, well, that's what Intel is saying. In that case, let's compare Zen4 and Zen4c ! They have the exact same IPC, Zen4 can clock higher but Zen4c can sustain a given clock speed for less power. At identical clock speeds, Zen4c is more efficient, period. But Zen4 can clock much higher, meaning it can be much faster.
This benchmark is thus very interesting, because it tells us than Zen5 is 12% faster at the same clock speed as Zen4 - meaning, all other things being equal, it's 12% better. Considering Intel's IPC gains between 13th and 14th being zero, and Zen4 already being competitive with Core i 14 while clocking 15% lower, Intel would need to clock their CPUs up by 25-30% more to compete on performance, and power draw would be out the window.
Remind us who had problems with their CPUs degrading due to overvolting again ?
What does that have to do with anything? If you care about efficiency then you look at the efficiency, you don't look at the clockspeeds. Period.
IPC is completely irrelevant.
- Jan 20, 2010
- 20,086
- 11,079
- 79,540
This is not true. Voltage scales with frequency and power scales with voltage. You can see this in how all of the laptop and phone chips run at lower frequencies. The best way to optimize perf/W is to maximize IPC, so you can keep clock speeds as low as possible and still have a fast chip.
TRENDING THREADS
-
-
Question Asrock B450 steel legend motherboard not loading OS through pendrive
- Started by surya_sapui
- Replies: 2
-
-
-
Discussion What's your favourite video game you've been playing?- Started by amdfangirl
- Replies: 4K
Facebook
Twitter
Instagram