How do you get from TGP to TBP?
Here it gets tricky again (and also a bit inaccurate) if you don’t want to measure anything on AMD cards, but only use free read-outs. Actually, the power consumption of an AMD graphics card cannot be measured in its entirety! Thus, many screenshots of relevant games with the Afterburner overlay on AMD cards only serve to create legends, because the values are not wrong, but so incomplete that they actually seem almost useless. You can’t just add a few watts on top by rule of thumb either, that will definitely go wrong as well. And now? Let’s take a look at the recently reviewed MSI RX 6950 XT Gaming X Trio from the launch article and first its BIOS default for the TGP:
Now we create a full load in Ultra-HD and remeasure the card for over 15 minutes. Since I want to avoid the mass of readings on the oscilloscope, I measure only the first and last two minutes at a time in 20ms intervals (True RMS). We can see very clearly that the values read out with GPU-Z coincide very well with the BIOS default for the TGP. The average measurement with my own shunt measuring station and PDAT from NVIDIA also coincide pretty accurately. The oscilloscope measurement also shows at the end that the TBP (i.e. the entire card) increased significantly after heating, while the TGP read out with GPU-Z remained about the same. The voltage converter losses and the fans now play a role here, even though the GPU itself remains almost unaffected.
Rule of thumb or guessing game?
It can be stated that both the shunt measurements and the results of the oscilloscope agree quite well, because there are always some tolerances, even between two measurements under the same conditions using the same method. If we now consider the measured TGP of 330 watts (GPU-Z) and the TBP (igorsLAB, shunt) of 431.4 watts, the actual power consumption of the MSI RX 6950 XT Gaming X Trio is almost a whopping 30 percent higher than the sensor value for the TGP! But is that always about a 30 percent surcharge now, or a third more?
Let’s test this against! The MSI RX 6750 XT Gaming X Trio shows an average of 225 watts TGP in GPU-Z as a heated card. With my shunt measurement setup, it’s just under 270 watts for the entire board. If we now convert the difference, we arrive at an increase of 20 percent, which is still a lot, but significantly less. So for the time being, it’s not comparable.
Therefore, I finally also test the MSI RX 6650 XT Gaming X. With a TGP of now only 145 watts, it is then 180 watts TBP for the whole card, so suddenly 24 percent more again! Thus, one cannot simply use a simple linear relation or curve relation here, which also simply falls off as a surcharge when the TGP falls. The four percent increase compared to the RX 6750 XT is mainly due to the significantly lower efficiency of the voltage converters. The value would be even higher if the fan and co. consumed as much as on the RX 6750 XT.
Therefore, I won’t go so far out on a limb to define an AMD rule of thumb that covers all cases exactly. This is not technically possible and would therefore also be abundantly dishonest. However, depending on the level of the TGP and thus also the board and component temperatures or the quality of the installed voltage transformers, you can assume an addition of between approx. 20 and 30 percent to the TGP to arrive at the actual final value. This is very rough and inaccurate. And it ends up being one of the reasons why I absolutely dislike software “measurements” on AMD graphics cards. In most cases, the tools leave users with more unanswered questions than they offer answers. At the very least, I hope today’s article is enough to dispel a myth or two. For the creation of legends you have to look for other topics.
And if you really want to measure it reliably and not just guess, just buy the PMD from Elmor, then you also get the EPS for the CPU.
Further links (still up to date):
657 Antworten
Kommentar
Lade neue Kommentare
Urgestein
1
Urgestein
Veteran
Mitglied
Veteran
Urgestein
Urgestein
Urgestein
1
Urgestein
1
Urgestein
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →