Sigh. Line by line.
Intoxicus
Ok, I'll humor you. Remember what I said about how mixing fact and fallacy can convince of falsehoods? That video is an excellent example. He says things that are true sometimes, but comes to flawed am biased conclusions.
-For starters that A/C rig is not going to be reliable when it comes to managing variables. "Probably scales similar to a waterblock" is simply not scientific.
Spoken like someone who's never put AC on a PC and benched it. I can AC my room to 60F and my hybrid from a cool start (20C idle) will hang with a waterblock in a temp normal room. His AC box setup is close enough and he's monitoring core temps so they are comparable, in the 40C's.
-Using a shunt mod invalidates everything immediately. Shunt modding is very niche and very uncommon. It goes against the idea of controlling for variables. A no power limit bios is what would be needed to get good results. Also is the 2080Ti shunt modded? He never clearly states so.
Yes, his 2080 Ti is shunt modded (VERY obviously). Shunt modding and higher power bios are very common in gpu overclocking.
He is controlling temp and voltage, while measuring max clock and card power draw. Good info to get.
-He does not fully explain his methodology. Gamers Nexus in contrast has videos completely dedicated to explaining their methodology and also updates to their methodology.
Now you're just hand waving and appealing to authority. He fully explained what he was doing, he just didn't spoon feed it to you.
-He has no control for variables at all. This makes this video and results inherently unscientific. A control is needed for proper methodology and properly accurate results
This is a move to dismiss the test as "not scientific". He's performing a perfectly reasonable comparison test.
-He doesn't understand electrical engineering and ignores current(amps) as a factor/variable completely. Ampere needs higher current at the same voltage resulting in higher wattage.(Watts==volts*amps) If Watts increases at the same volts, then there is more current(amps). A higher current can deplete the lifespan of your card Which makes statements like "a shunt mod solves everything" foolish at best.
Your tenuous grasp of electronics, how gpu's work, plus more hand waving.
P = V x I.
Card is supplied with 12VDC. You can easily calculate board current from any GPU power reading. People don't usually discuss current, because cards run on 12VDC and cards report power. Here is how a shunt mod works:
https://youtu.be/IAqFQBgRE1M These cards run against the power limiter all the time. More power is good if you have the cooling for it and IF the chip scales with power under high load, which is what he is trying to determine. 500w for a reference 2080 Ti with good cooling is pretty reasonable.
Totally neglected here by you and FC is that the GA102 chip seems to pull considerably less power than the TU102 at similar board power levels. This needs to be investigated further. In the example here, the TU102 is pulling more power and more current, at a higher voltage:
https://forums.evga.com/G...-m3085826.aspx#3085860 Now I'm running on my 2080 Ti Hybrid standard Heaven at locked 1.025v (as the 3080 above) at 2100mhz, 55C, edge of stability, seeing 325w board power, 244.5w chip power, so notably more current on a lighter load than the 3080 on Port Royal.
Now in Port Royal I'm hanging around 338w board (XC bios nominal max), chip 252w, locked at 1.025v and 1980 MHZ. So the TU102 definitely pulls more power and current at the same voltage. 55C and stable.
-Port Royale is a heavier and more power demanding benchmark. The cards are not
eveb tapping their full potential without using RTX.
While this is factually true, most people don't care about ray tracing. FC doesn't care about ray tracing. In the 15+ months I've owned my 2080 Ti, I've used the RTX cores for all of 10 minutes, running Port Royal. TSE is a perfectly good 4K DX12 test and fully power loads a card.
-He misunderstands the difference between AMD & Intel CPUs when it comes to multicore vs single core performance. Intel's architecture is better at single core, while AMD's chiplet design is superior at multicore performance. These are architectural differences between the CPUs. Because gaming still doesn't fully utilize multicore processing that well single core high clocks do better, especially at lower resolutions. At 1080p the CPU is doing more work than it would at higher resolutions, resulting in a CPU bottleneck on all GPUs. This gives Intel an advantage because of the higher single core performance.
The 100% load vs not 100% load statements totally miss the reality of the difference.
He's making an analogy, not a technical comparison and honestly I didn't pay much attention to it. AMD has a ridiculous architecture tying a bunch of cheap chips together. That's why they are cheap and good at multicore loads.
-8nm has lower power draw with more cores/transistors. Smaller nodes are more efficient while the increase in amount of cores offsets that efficiency to result in a higher TDP. He's mixing fact and misunderstanding heavily on this point. Statements like "this is the same as 12nm" are inherently fallacious.
The GA102 consumes less power at the chip than the TU102. He is not aware of that yet, but will be soon. Regardless, they are very similar chips but clearly not the same.
-He overvalues clock speed on GPUs. In GPU tasks more cores can outweigh higher clocks. He speak like higher clocks is always better. The reality is that it's truly case by case and not something to generalize like that. Clockspeeds on the GPU won't help at 1080p so much because of CPU bottlenecking at that resolution.
Clock speed is very important in any processor. OBVIOUSLY. Especially for people interested in overCLOCKing. He makes point that in a light load, higher voltage, higher clock, less power limited scenario, the 2080 Ti can excel. The 1080 and 1440p benches back that up.
And no, not every CPU bottlenecks at 1080p....that's a reddit AMD proponent opinion. Of course, Ampere has more transistors per real core and higher IPC, but there are some real unanswered questions about chip vs. board power draw and where the power is going.
-All GPUs are CPU bottlenecked @ 1080p. Shunt mods don't fix that. Making a GPU go faster does not solve a CPU bottleneck.
This is just a ridiculous statement to make. Most people are at 1080P and most people are GPU bottlenecked most of the time.
There's a reason why people who know what they are doing buy Intel, fast ram, good cooling and OC. Many games can run into the hundreds of fps. But they want more GPU too.
All he's shown is it takes more current to run more cores/transistors at the same voltages. Which is kind of obvious to anyone that understands this stuff on a fundamental level.
As I said before, the GA102 is reporting less power draw at the chip. I have yet to see any comparisons of chip power vs. clock vs. voltage, which would be insightful, nor a real transistor count of the cut down GA102 in the 3080.
There is no "deep dive" here. It's surface level and he doesn't even get fundamentals correct.
In contrast Gamers Nexus is the most scientific, objective, and reliable reviewer I know of. He has a robust and well defined methodology. Uses controls for variables, has repeatable results, actually knows the fundamentals, and admits when he doesn't know or is not sure.
Another appeal to authority. Tech Jesus is very good, but doesn't have a monopoly on testing computers and far from a professional test engineer. This is one guy doing a test, in his basement, that nobody else has done.
post edited by kevinc313 - 2020/09/23 18:19:59