저자: TechKnowledgeVideo진행자 · 외국인
요약 보기
AMD 불도저(Bulldozer) 아키텍처와 FX 시리즈는 흔히 실패작으로 평가되지만, 영상은 그 평가가 과장됐다고 주장한다. 출시 배경: 2009년 경 인텔 샌디브리지와 AMD 불도저가 차세대 아키텍처로 개발됐고, 출시 후 각각 인텔 Core 2세대와 AMD FX 시리즈가 됐다. 이전 AMD 제품은 인텔에 뒤처졌기에 불도저는 동등 이상의 성능을 낼 기회였다. 성능 평가: FX-8150 출시 당시 언론은 단일 코어 성능이 이전 세대보다 나쁘다며 실패로 규정했다. 게이밍 벤치마크에서 FX-8150은 이전 플래그십 Phenom II X6 1100T보다도 자주 뒤졌고, 워크스테이션 작업에서는 샌디브리지 i5/i7과 엇비슷하거나 일부만 앞섰다. 설계 문제: 불도저는 코어마다 독립 부동소수점 스케줄러를 두는 대신 두 코어(모듈)가 자원을 공유했다. 이는 전력 효율과 확장성을 노렸지만 약 15%의 성능 손실을 낳았다. 캐시 지연 시간이 길고, 잘못된 분기 예측 페널티가 샌디브리지보다 컸으며, 서버 시장을 겨냥한 설계였다는 점도 한계로 꼽혔다. 소프트웨어 문제: Windows 7이 FX 아키텍처에 최적화되지 않아 코어 파킹이 제대로 작동하지 않았다. UEFI와 OS 업데이트 이후 개선됐고 Windows 8 프리뷰에서는 성능과 전력 소모가 나아졌다. 당시 게임은 OpenGL이나 DirectX 10을 썼고 멀티스레드 활용도가 낮아 8코어를 제대로 쓰지 못했다. DirectX 12와 Vulkan이 나온 2016년 이후에야 다중 코어 확장이 잘 됐다. 마케팅 과대: AMD는 2011년 초 샘플이 Phenom II보다 최대 50% 빠르다고 주장했으나 실제 제품에서는 재현되지 않았다. 언론도 Gizmodo 등에서 기대감을 크게 표출했다. 전력·서버: 고부하 시 전력 소모가 커서 발열, 소음, 비싼 쿨링 비용이 문제였다. 리눅스 성능도 출시 초기에는 기대에 못 미쳤고 약 6개월 뒤 개선됐다. 그 사이 서버 커뮤니티가 이탈해 2016년 AMD 서버 점유율은 1% 미만으로 떨어졌다. 결론: 불도저의 진짜 문제는 제품 자체보다 출시 시점과 경쟁 환경이었다. 샌디브리지가 가격 대비 성능을 강하게 선점한 가운데 AMD는 이를 따라잡지 못했고, 낮은 마진으로 인해 연간 수억 달러 손실이 누적되며 인텔의 고가 정책과 서버 시장 진입을 허용했다. 이후 라이젠과 EPYC로 AMD가 점유율을 되찾고 있다. 투자 관련 발언: 대상 자산에 대한 1~3개월 가격 방향 예측은 없다.
다듬은 글
in 2009 the two largest desktop processor manufacturers Intel and AMD were in the process creating two next generation processor architectures code names Sandy bridge and bulldozer respectively when released these products became the Intel Core ey Second Generation and AMD FX series for the past few years amd's previous micro architectures delivered poorer performance in Intel's lineup and therefore the new bulldozer lineup was seen as an opportunity to achieve parity or exceed Intel's efforts Intel's Sandy Bridge launch was strong with the product offering excellent performance for the money when the flagship bulldozer product launch the FX 8150 processor the media and Technology press were unimpressed citing it as a failure and highlighting how the single core performance of the processor was worse than that of the previous generations the other processes in the family the FX 6100 and FX 4100 series were also reviewed as unimpressive this hurt the company in the long run microprocess development is a multi-year process and as a result subsequent architectures were also based upon this design this resulted in the company losing market share for years to come hitting a low of just 177% of all PC sales in late 2016 according to pasmar Additionally the company suffered large losses for the next few years posting a massive loss of 4 $97 million in 2016 since the launch the bulldozer architecture and its derived successes have earned the reputation of one of the biggest failures in the Computing landscape in recent memory however while the overall processing power was underwhelming there was nothing intrinsically wrong with the product the pricee of the lineup was low enough that it was worth the purchase if the user was doing multi-threaded work codes regularly it is therefore puzzling as to why the processor lineup is so notorious for being a failure to answer this question we need to look at benchmarks and in this case taken from a nanch in gaming benchmarks the FX 8150 performs poorly against the competition and rarely pulls ahead of amd's previous Flagship the Phenom 2 X6 1100t Black Edition in civiliz ation 5 World of Warcraft and Starcraft 2 for example the performance difference is pretty noticeable in workstation benchmarks however the FX 8150 performs better but still not that great it pulls ahead or equal with Sandy bridge i5s and i7s in some tasks but loses in others the performance deficit was due to the design of the bulldozer architecture being radically different than that the thuan architecture used in the phenon 2x6 the previous generation processor instead of having each processor made up of individual cores each with their own floating Point scheduler each bulldozer core called a module was actually just two individual cores that shared the same resources including the floating Point scheduler this radical approach was supposed to deliver better power efficiency and scalability but it took a performance hit of roughly 15% compared to traditional design the processor also had a very limited amount of cash which had very high latency which was one of the reasons which let down gaming performance another key issue was that bulldozer paid a heavy penalty for incorrect Branch prediction in comparison to sandybridge which paid a lighter penalty finally the architecture itself was designed more for the server Market than it was for the the desktop which limited its potential in certain tasks it's also important to note that it was hampered by limitations at the time because the architecture was different from typical desktop processor Windows 7 had not been optimized for the FX processor line which hurt performance and increased power draw one of these features was core parking where the processor shuts down cores that aren't being used in order to save power this was not working correctly when the initial reviews launched it was only after several UEFI updates and operating system updates this feature was enabled when the FX processor line was tested on Windows 8 preview the performance improved and power draw was produced additionally when the reviews were released all games tested ran using either open gel or Microsoft's d3d 10 these Graphics apis not very low level and not very multithreaded which means that games using its interfaces would rarely be able to use all eight bulldozzer cores it is only since 2016 with the release and Adoption of Tourette 3d2 from Microsoft and Vulcan from the Kronos group The Gaming performance can scale well with multiple CES while the performance of bulldozer was perhaps slightly limited by this fact factors it is clear that the performance generally was underwhelming but to understand why people have such a poor memory and view of the borders architecture is important to look at other influences one of the reasons that the bulldozer architecture's performance seems so lackluster is because of how successful Intel Sandy Bridge architecture was launched before bulldozer the performance was a large improvement over Intel's previous architecture it also overclocked well and was extremely well priced it isn't so much that bulldozer is a failure more that it didn't represent the step forward in performance and value that Sandy Bridge achieved the architecture itself was also overhyped to the extreme for example in early 2011 A&D claimed that their bulldozer samples were up to 50% faster than the Phenom 2 processes of the time as seen in the benchmarks this level of performance was simply not present in the final version important to note is that the technology press were also very excited for the new architecture when the architecture was announced amd's new Core Design was seen as having huge potential for example technology website Gizmodo stated that the new Core Design could potentially offer quote phenomenal performance again this was not seen in the final product it is also worth mentioning power when a new architecture is released the idea is not only that performance improves but also efficiency however the bulldozer FX processors were very power hungry under load which means the processor would get hotter fans would run louder users would require more expensive calling and the processes themselves fail more frequently the competition Sandy Bridge was very power efficient at the time and therefore Not only was the performance of bulldozer objectively worse than Sandy bridge but it also Drew more power this again did not help AMD to convince the users that their new architecture was advanced in comparison to the competitions finally the bulldozer systems saw problems in the data center bulldozer was originally designed to be a server CPU That Was Then designed for the home user whereas Sandy Bridge was built from the home user first however the high power consumption of bulldozer meant that it wasn't suitable for large D servers due to the cost of calling not only that but upon launch the Linux performance was significantly worse than expected and it took approximately 6 months for the True denx Performance to be seen by that time many in the server Community dismissed the wozer based optron CPUs this resulted in AMD server market share dropping to less than 1% of the market by 2016 it is fair to say that amd's bulldozer architecture was not what either the consumer or what AMD was hoping for however I believe that it's stated is a failed product is overstated the biggest problem with bulldozzer is not the processor itself but the timing of its release and the Computing landscape at that point games were not optimized to take advantage of the eight cores and Intel's phenomenal Sandy Bridge represented a largely forward in performance per dollar the AMD simply could not keep up with regardless of its True Performance or amd's Public Image the real legacies of bulldozer is not the performance issues but how much it hurt AMD in the long run AMD could only price bulldozzer up to a certain point and therefore it meant profit margins were low and not much money could be made on the product this meant that Intel could price their high-end products however much they wanted and in the process posted record Revenue at a time where AMD was losing hundreds of millions of dollars a year additionally amd's loss of market share in the server industry also allowed Intel to get a strong foothold in that market luckily with the recently released ryzen and epic processors AMD looks set to take back market share in the home and server only time will tell how successful that becomes and hopefully that was a bit of Technology a
Comments 0