P95v308b17.win64

0 views
Skip to first unread message

Adriana Gowen

unread,
Aug 4, 2024, 8:53:02 PM8/4/24
to anpytipan
Iwas warned more than once by acquaintances or friends to do this test and offer my name as the clean-up guy after the shitstorm and possibly get personally affected. Because even if it is often overlooked, each of us testers has our own name under the articles, Igor often only provides the platform. Anyone who works incorrectly in an article and does not admit their mistakes is potentially unacceptable in the entire industry for all eternity.

The heat generator is an Intel Core i9-13900KF on the LGA1700 platform and a Ryzen 9 7950X on AM5. The CPUs are each statically overclocked, including e-cores in the case of Intel, in order to generate a heat load that is as high and constant as possible. The exact settings follow on the next page in the test hardware section. Prime 95 (p95v308b17.win64) is used as the heat load with a fixed FFT size of 200K, in-place. This allows us to generate as much constant waste heat as possible and is also independent of the RAM, as the benchmark runs completely in the CPU cache.


To determine the temperatures, the individual core temperatures and the water temperature are recorded over the benchmark run of 15 minutes and the effective delta to the water temperature is calculated for each core for each data sample. This allows us to measure the water blocks pretty fairly even without a chiller and completely constant water temperature. It is important to understand that we really form the delta for each core, for each data set, and only at the end are metrics generated from the data sets, such as the average of all P-cores or the difference between the on average hottest and coldest P-core. The delta values are given in Kelvin, although these can of course be transferred 1:1 to degrees Celsius.


The block is then removed, the block and CPU are cleaned and the block is mounted for a second set of measurements. The procedure from the last paragraph is repeated so that we end up with 2 measurements for each of 2 pump configurations. At the same time, I can filter out one-off errors in the mounting of the block from the results, such as uneven tightening of the screws, because only the better of the two runs per pump configuration makes it into the diagrams later. In other words, I would have to be stupid twice in a row for a block to perform worse than it could.


The flow rate is also logged so that we can also measure how restrictive a block is for the water circuit. A constant flow rate via variable pump settings may seem fairer at first, but in practice a very restrictive block with a resulting low flow rate also performs very well, and vice versa. With static pump speed, however, the influence on the flow rate in the entire loop can also be derived.


By the way, we provide you with the CSV data of the individual test runs for reasons of transparency, so that there is really no doubt as to whether it is real test data or fantasy values: _K2ujN6f5csCPO5mlN4BUbVZeXpLmnn?usp=sharing


Since Intel only measures or logs the core temperatures in whole degrees, results to a tenth or hundredth of a Kelvin can only be reproduced to a limited extent. Although the water temperature is recorded accurately to a tenth of a degree Celsius and in the end we also form an average from the 419 individual data points, a certain measurement tolerance remains. According to my analyses, this amounts to 0.4 to 0.6 K. Incidentally, AM5 CPUs do log to within a tenth of a degree Celsius, so their results will be a bit more accurate.


Super Arbeit. Ich nutze auf AM5 den Wasserblock von NTech. Qualitativ und von der Montage ist das fr mich das beste Gesamtpaket fr den Preis. Von der Khlleistung und vom Flow gibt es auch nichts zu meckern.


Mir fehlt ja immer noch der Glaube daran, das man nur anhand des Losbrechmoments eines normalen Gewindes, die Kompressionsrate einer Schraubenfeder exakt einstellen kann. Bei meinen Versuchen hat das selbst mit einem feinen Trapezgewinde(Schraube und Mutter aus Edelstahl) nicht wirklich exakt funktioniert. So das am Ende doch wieder die klassische Methode mit dem Nachmessen per Schiebelehre an allen vier Schraubenfedern ntig war.


Kann natrlich auch sein, das es an meinem zu ungenauen Drehmomentschlssel lag. Daher meine Frage: ob du vorher mal getestet hast, wie genau sich die Anpresskrfte mit der DrehmomentMethode wiederholbar an den Schraubenfedern einstellen lassen. Denn das ist mMn schon sehr wichtig, damit der Khler gleichmig aufliegen kann.


Danke schn fr die investierte Zeit und das doch ernchternde Fazit. Fr mich zeigt sich einmal mehr, dass der Massenfertigungswahnsinn in Asien, die Qualitt der Produkte verschlechtert. Oder wie soll man sonst die Fertigungsmiesere mit zu kleinen Lchern und berstehenden Schraubenkpfen etc. nennen?

Auch wenn das eine oder andere Ergebnis in diesem Test, Soft- und/oder Hardwarebedingt, Tolleranzen hinterlt, kann man diese als valide ansehen.

Also noch einmal danke schn fr die zeitintensive Aufarbeitung und Neubewertung.


Also ja, du hast Recht. Wir montieren "blind" nach Vorgaben des Herstellers und vertrauen darauf, dass die Schrauben alle gleich sind bzw. das Drehmoment auch zum gewnschten Anpressdruck fhrt. So machen es ja auch die allermeisten Endanwender und somit ist es auch im Interesse des Wasserblock-Herstellers, dass die Drehmomente zum korrekten Ergebnis fhren.


Angezogen wird zudem ber Kreuz, schrittweise und in gleichmigen zgen, um "stick and slip" zu vermeiden. An einer einzelnen Schraube mit festem Anschlag habe ich auch mal die Wiederholgenauigkeit meines Amazon-China-Drehmoment-Schlssels getestet und bis auf 0,02 Nm schien das reproduzierbar zu sein. Aber mit 4 Schrauben kann sich der Block ja auch verkannten, evtl. sind nicht alle Schrauben gleich gefertigt oder die Wrmeleitpaste ist auf einer Seite der CPU dicker aufgetragen, sodass eine Schraube mal mehr Kraft transportieren muss. Alles ziemlich viel Blindflug :D


Groartiger Test, vielen Dank dafr!

Ich habe meine Khler von AquaComputer und Sockel 1155 auf AM4 umgerstet. Sollte es kein Umrstkit fr den nchsten Sockel, den ich verwenden werde, geben, kann man sich hier sehr gut orientieren. Die grte Verbesserung meiner Wak in den vergangenen Jahren war die Erweiterung des Kreislaufs auf einen externen Mora 420. Damit steigen die Wassertemperaturen trotz neuer GPU nicht mehr ber 34C. ;-)


Damals mit 2 oder 4 Kernen konnte man recht einfach unter Last vorsichtig die Verschraubungen lsen bzw fester drehen und dabei die Temperaturen beobachtet. Fingerspitzengefhl vorausgesetzt. Scheint inzwischen kompliziert.

Federn vertraue ich grundstzlich nicht. Wer schon einmal einen Zylinderkopf eingemessen hat (Ventilfedern) der kennt die Serienstreuung - im kleinen wie im groen halt.

Guter Artikel brigens @skullbringer ?


Wir verwenden Technologien wie Cookies, um Gerteinformationen zu speichern und/oder darauf zuzugreifen. Wir tun dies, um das Surferlebnis zu verbessern und um personalisierte Werbung anzuzeigen. Wenn Sie diesen Technologien zustimmen, knnen wir Daten wie das Surfverhalten oder eindeutige IDs auf dieser Website verarbeiten. Wenn Sie Ihre Zustimmung nicht erteilen oder zurckziehen, knnen bestimmte Funktionen beeintrchtigt werden.

3a8082e126
Reply all
Reply to author
Forward
0 new messages