Recent content by HumanSmoke

I'm continually amazed how many people get the development of this pivotal product in AMD's history wrong. Keller worked on the original K8 project. This was shelved and the actual K8 was designed by a team led by AMD's chief architect Fred Weber. What did Keller actually achieve at AMD? He led...

My point was that you cant make gross assumptions based on bias and a very limited number of data points. For example I've run watercooled systems for over fifteen years.I do so not because they are cost effective (they aren't), or because of prestige, or because they require less maintenance...

Sorry, but that is largely erroneous. The principle source of (mis)"information" regarding Bulldozer prior to it's launch was a high profile user going by the moniker "JF_AMD" on various high profile forums such as Anandtech, and OCN. "JF_AMD" was/is John Fruehe - who was as the the time Senior...

Titan X + EKWB wterblock is still cheaper than a Radeon Pro Duo Why would TPU be ranting about not getting a review sample when they do have a review sample? :SMFH: https://www.techpowerup.com/reviews/NVIDIA/Titan_X_Pascal/

No, it is actually quite easy to prove IP theft on a IC logic block basis. It was easy 35 years ago when NEC was found to have copied Intel's 8086/8088 and marketed them as the µPD 8086/8088, and it is even easier now where private companies routinely shave heatspreaders off chips and produce...

Since the Titan X has three times the HWBot benchmark submissions of all the Fiji-based cards combined, you could then conclude by your data that people buying AMD's top end cards are three times as likely to spend their computer time playing simple run-and-gun games and indulging in run-and-cum...

You clearly don't understand how the pure-play foundry model works. Analyzing and copying the IP logic blocks of a customers IC's would destroy Samsung's foundry business - they wouldn't retain or gain a single customer, and by the time the court system finished pounding them into dust for...

He already tried, but the neighbours started complaining whenever he got close to 1650MHz :laugh:

Show proof or quit with the guerrilla marketing. Neither TSMC nor Samsung (nor any other foundry) release wafer costs - especially those weighted for yield. What is known is that Samsung's 14nmLPC (the successor to LPP) and TSMC's 16nmFFC both offer similar production cost reductions due to...

In the greater picture, both Nvidia and AMD play pretty clean in the corporate world. Nvidia have had a hand slap, but there's nothing to say AMD couldn't be hot water either over misrepresentation of Llano yields. Raja Koduri's assertion of Polaris being "months ahead" of Pascal some months...

Free entry for a one of a kind signature edition graphics card that eats 99.999%+ of existing VGA's for breakfast?......Yeah, OK, you talked me into it - now that was a tough 0.2 seconds! Your standard response has been duly noted, Please collect you prescription from the receptionist on the...

Precisely. Once Nvidia decided to not wait around for 10nm and commit Volta to 16nmFFC production there probably wasn't any good reason not to go ahead with the HPC orders since it ties in with IBM's POWER9 schedule.

I suppose the card could be the mobile M295X, but if you are highlighting a new architecture it would seem strange that AMD would use a perf/watt optimized mobile part for its baseline - that just raises the bar significantly higher to achieve "2.5x perf/watt"

The discussion has nothing to do with the Polaris cards . The discussion point arose out of casecutter querying what 28nm GPU was used as the baseline for 1.0x perf. As you well know since YOU weighed in with some nonsensical observations. Now you go with the classic lowbrow troll move of trying...

Remember what I said about bad logic? The 280X launched in October 2013....now check the timeline on the AMD slide that positions the 28nm GPU being compared. It shows late 2014 Even if you had used the last HD 7000 rebrand, the R9 280, it launched at the beginning of 2014 and still doesn't...