Test 1 -- Loading before unknown load address: Loading before unknown load address: 20 cycles Loading before known load address: 12 cycles Loading before unknown load address: No!
Test 2 -- Loading before unknown store address: Loading before unknown store address: 20 cycles Loading before known store address: 12 cycles Loading before unknown store address: No!
Test 3 -- Storeing before unknown load address: Storeing before unknown load address: 25 cycles Storeing before known load address: 14 cycles Storeing before unknown load address: No!
Test 4 -- Storeing before unknown store address: Storeing before unknown store address: 25 cycles Storeing before known store address: 14 cycles Storeing before unknown store address: No! Finished.
w_ty (w_ty@w_ty.net) on 10/13/06 wrote: --------------------------- > >Test 1 -- Loading before unknown load address: >Loading before unknown load address: 11 cycles >Loading before known load address: 11 cycles >Loading before unknown load address: Yes! > >Test 2 -- Loading before unknown store address: >Loading before unknown store address: 11 cycles >Loading before known store address: 11 cycles >Loading before unknown store address: Yes! > >Test 3 -- Storeing before unknown load address: >Storeing before unknown load address: 11 cycles >Storeing before known load address: 11 cycles >Storeing before unknown load address: Yes! > >Test 4 -- Storeing before unknown store address: >Storeing before unknown store address: 11 cycles >Storeing before known store address: 12 cycles >Storeing before unknown store address: Yes! >Finished.
Introduction to IntelR Core? Duo Processor Architecture http://www.intel.com/technology/itj/2006/volume10issue02/art01_Intro_to_Core_Duo/p03_improved_cores.htm >Intel Core Duo also improved the latency of some long latency integer operations > such as Integer Divide (IDIV). Although these instructions are not very frequent, > because of their extremely long latencies, their accumulative affect on integer > benchmark scores have shown to be very significant. The basic Divide algorithm > has remained unchanged; however, Intel Core Duo Divide logic exploits opportunities > for "early exit." The Divide logic calculates in advance the number of iterations > that are required to accomplish the operation. This is indeed data dependent; however, > it is often significantly smaller relative to the maximal number of iterations. > Once the required number of iterations is accomplished the divider wraps up > the results. This does not impact the maximal Integer Divide latency; however, > on average it is much faster.
*** POWER6 - ~2x frequency (4~5GHz) vs POWER5 (~2.3GHz) - instruction pipeline depth == POWER5 - Minimize power, Scale performance with frequency
- 13 stages FXU pipe (PWR5 =15 stages) - 64K-I , 64K-D-Cache (PWR5 = 32K-I, 64K-D) - FXU x2, FPU x2, 1 Branch (== PWR5) - Decimal Unit, VMX Unit, Recovery Unit - Added ~50 instruction to POWER ISA -- Add, Multiply, Divide, Conversions(<-> Int, BCD &DFP), insert Exponent(scaling), Quantize (single unit), Reround(round to less precision once) - 2 way SMT (7 instruction dispatch from 2 threads, max. 5 instruction per 1 thread)
- Cache -- 4MB Private L2 per core, total 8MB (PWR5 = 1.875MB shared ) -- 32MB Shared L3 Cache per chip , 128-bit/dir, 80GB/s (PWR5 = 36MB)
- Fabric -- 3x Intra-Node SMP bus for 8-way Node (total 80GB/s) -- 2x Inter-Node SMP bus for up to 8 Nodes (tota 50GB/s) -- 2x MemController total 75GB/s, GX+ bus 20GB/s
- New prefetching capabilities -- Coherent Multi-Cacheline Data Prefetch Operations -- Prefetching on stores
昨日は>>15の改変画像を上げただけで寝てしまったが今日もTukwilaに粘着します(w >>59の画像ではMontecitoのコアが1MBのSRAMの4倍程度なのに対して、Tukwilaでは6MBのSRAMと同等以上とコアが大きくなってるように見える。 http://ieeexplore.ieee.org/search/wrapper.jsp?arnumber=995709 >OOO tends to be effective in prefetching-for L1 misses; whereas SP is primarily good at covering L2 and L3 misses. (SPとOoOが実装されればL3が不要になる。) ってことでやっぱりTukwilaには動的スケジューリングが実装されてる気がする。
>Intel announced that the first of 15 45nm processor designs is scheduled to be completed by the end of the year, >with Intel's next-generation 45nm process technology on track for production in the second half of 2007. 45nm製品の投入次期はかなり流動的な気がする 出そうと思えば07Q3にも出せるだろうけど、稼動している工場が現状一つしかないので >>63でも書いたけど07Q3にYorkfield-XE(D1Dのみ) → 07Q4にHarpertown&Stoakley(Fab32稼動)こんな感じになると思う また、08H2にはFab28が稼動するのでその辺でNehalemが投入される予感
>The company introduced eight Dual-Core Intel Xeon 7100 series processors for multi-processor (MP) servers >offering up to twice the performance and up to four times the performance-per-watt of previous designs. >Dell, Fujitsu-Siemens and IBM announced new world performance records using the new Intel Xeon MP processors. TulsaできっちりOpteronを抜いてIntelはプラットフォームベンダとしての責任を果たしたと、そういうお話
>the currents going through the processor voltage regulator circuitry >(this way we could check the processor power consumption without the processor voltage regulator efficiency)
For example, when we resorted to BurnK6 tool that we used to use to heat up AMD K6 processors back in the days, we managed to get much higher power consumption numbers for Core 2 Duo E6300: 55W for the CPU and 229W for the platform. In other words, Intel Core based processors hit the maximum power consumption and heat dissipation in absolutely different type of tasks than their competitors and predecessors.
>Nehalem is 45nm based and we are going to see some exciting changes such as > IMC (Integrated Memory Controller) and CSI. Socket B will be LGA1366 and Socket H > will be LGA715. The reason why Socket B has so many more contact pads >is pretty obvious. It is due to the IMC on the Nehalem CPU. Since there is a Socket H version as >well, we can expect there will be another Nehalem based CPUs without IMC.
Intel is on track to complete its first 45nm design in the first quarter of 2007, and the company aims to launch related products in the second half of next year, Intel confirmed at the ongoing Intel Developer Forum (IDF) conference in Taipei. The company also revealed it expects Robson cache technology to become pervasive in notebooks in 2-3 years.
プロセッサコアの規模に対してパフォーマンスは平方根程度でしか向上しない(= Pollackの法則) ので、Largeコアでは、1命令を実行するのに必要なEnergy(EPI = Energy Per Instruction)が 必然的に大きくなり、Smallコアでは逆にEPIが小さくなる。 Energyのコストの大きいLargeコアを逐次処理用に少数、Energyコストの小さい コアを並列処理用に多数もつ構成の非対称マルチコアは、発熱と消費電力の 観点からみて理にかなっているわけである。
>The larger and faster cores will execute single-threaded programs and the serial > phases of multithreaded programs for high EPI, whereas the smaller and slower cores > will execute the parallel phases for lower EPI.
Fab 28 in Kiryat Gat involves an investment of over $3.5 billion by Intel, in addition to a grant of $525 million from the government of Israel. It will work at the 45-nanometer manufacturing node on 300-mm diameter wafers. Production is expected to start in the first half of 2008.
IBM rolls in Clover with Dell http://www.theregister.co.uk/2006/11/10/ibm_clovertown/ >Most Blades are currently equipped with a single hard disk drive which holds an image of the operating system and necessary swap files and the like. >This will be replaced by a 4GB Flash memory, which is considered sufficient capacity for the job and now costs the same as the disk drive. Robsonはブレードサーバーにも活かされる模様。
>The T7700, the T7500, the T7300 and the T7100 will have IDA, which stands for Intel Dynamic Acceleration.
>Intel claims that when single threaded apps or multi-threaded apps with extended serial code are executed, >IDA gives additional performance benefits.
Xeon E5335の追加とそれに伴う価格改定やXeon 3210は私的に新鮮なネタ。 Harpertown(45nm Quad-Core?)対応でWS向けと噂されるSeaburgが128GBまでメモリサポートを拡張するという話はガイシュツだが、 矢張り密度が上がるようだ。(ついでにAMBが4.8GHzで動作、つまりDDR2-800をサポートする見通し。) 大量に生産されるらしいので値下がりにも期待。 http://www.hpcwire.com/hpc/1097999.html >During this time, the available density of FB-DIMMs from Qimonda will grow from today's offering, of 1GB, 2GB and 4GB, to include 8GB and 16GB modules.
Intel Kicks ATI to the Door and Invites in SIS http://www.dailytech.com/article.aspx?newsid=4964 >On the other hand, there is one exception that’s expected in Q2’07; the new D201GLY Little Valley. >Intel Desktop Board D201GLY Little Valley is currently expected to be powered by a SIS 662 chipset and feature a mini ITX form factor.
SiS Announces 662 Chipset http://www.dailytech.com/article.aspx?newsid=3168 >Intel also just recently introduced a chipset to compete with SiS 662. >The Intel 946GZ was silently released on July 3, 2006.
その後CSI開発の主導権はPoulsonをやっていたハドソン(少数)とフォートコリンズに戻されたとみられる。 当然Poulsonは凍結 or キャンセルになったわけで、ブチキレたハドソン(少数)の要職―といっても彼は政治家だが―が辞職。 ハドソン(少数)の残骸はおそらくフォートコリンズに吸収された。まあリストラされたという話は聞かない。
Bloomfield Is Quad Core & Octo Threads http://www.vr-zone.com/index.php?i=4322 >The next generation Intel processor based on the Nehalem architecture is clearly exciting as > VR-Zone has learned. Successor to quad core Yorkfield which forms part of the 45nm Penryn architecture, > Bloomfield will come along and sit right on top of the 45nm Nehalem desktop processors in mid 2008. >Bloomfield will have 4 cores and is capable of 8 threads like the old Hyper-Threading technology but >only more advanced. Bloomfield will contain an integrated memory controller that requires a new >socket refresh called Socket B with 1366 contact pads.
Nehalem世代のデスクトップはBloomfield。 BloomfieldはSocket B 1366pinのパッケージで、mid-2008ごろに登場。 Native 4 coreで、Hyper-Threadingにより8 threads処理できる。
// VR-Zoneだけになんで最近Penryn&Nehalemネタがよくでてくるのかはわからんけど、 // 従来の情報と矛盾する点は特にないので信用してもよいかも。 // 以前、Gel氏の発言では、Nehalem世代で最大4 wayのHTという話があったけど、 // DesktopのBloomfieldでは、2 way HTで、Server用ののGainestown&Becktonでは4 way HT // という差別化はありそうである。
(※3) http://www.theinquirer.net/?article=28298 (2005年12月12日) >CSI was started with a "mandate" to beat HT at everything, speed, bandwidth, per pin bandwidth, >power usage, and everything else under the sun, not under Sun.
>>203のINQ記事だけが問題。 『CSIの速度から、Tukwilaのコアに入るところで2.4GHzに落とされる』というのが、 聞き間違いなどで変形して >Tukwila's speed might drop in the system interface section to 2.4GHz となったんじゃないかと勘くぐってるんだけど。 (4 coreで40GFlopsの推定演算能力なら、Tukwilaは2.5GHzである)
1枚目の写真はSPECfp_rate2000。おそらくIntelはSPECint_rate_base2000_で比較してくるので混乱しないように。 2枚目の写真はサーバーベンチでClovertwonがWoodcrestから60%の性能向上しか果たしていない(>>46参照)のに対して、K8Lは70%の性能向上を達成するというもの。 AMD leapfrogs Intel with quad-core server MPU http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=YBRWC3DDFNGDEQSNDLPCKH0CJUNN2JVN?articleID=196600668 >Jonathan Eunice, a principal IT advisor at Illuminata, expects that AMD will get a better performance boost with its quad-core processors than Intel did. >He says AMD might see as much as a 70% performance jump, depending on the application being run.
<サーバー系ベンチのスコア おおよその傾向> System x3950(XeonMP 71xx*16) > System x3950(XeonMP 71xx*8) > System x3650(Xeon 53xx*2) = System x3755(Opteron 8xxx*4) = System x3950(XeonMP 71xx*4) > Sun Fire X4600(Opteron 8xx*8) <可用性> System x3950 > System x3650 > System x3755 > Sun Fire X4600 売れるとしたらIBMのSystem x3950ぐらいか?
'You'd have to be nuts' to think AMD can match our tick – Intel http://www.theregister.co.uk/2006/12/14/intel_tick_tock/ >AMD, well behind on the 65nm front, looks to shrink the usual two-year lag between process generations by moving to 45nm within 18 months. >Intel executives, however, characterized the 18-month plan as a figment of AMD's imagination until the rival chip maker can prove otherwise. > >"They are so dreadfully behind," said Intel SVP Pat Gelsinger, adding that "you'd have to be nuts" to think AMD will come close to beating Intel to any manufacturing milestones. (中略) >More recently, market share figures show that Intel stopped the server chip hemorrhage – a fact not lost on the often thick analyst community. > >"I was very pleased to read that a number of analysts have downgraded AMD," Gelsinger said.
IBM And Intel Initiative Accelerates Virtualization On Multi-Processor Servers http://www.intel.com/pressroom/archive/releases/20061214comp.htm >Using vConsolidate to benchmark the IBM System x3950 server with four dual-core Intel Xeon 7100 processors shows the x3950 delivers >up to 46 percent more performance throughput than a competing system when running a mix of larger two- and four virtualized processor partitions. (中略) >"The success we’ve seen with our System x servers and the Intel Xeon 7100 processor series architecture gives us the confidence to invest in >a fourth generation of our Enterprise X-Architecture supporting quad-core Intel Xeon processors for multi-processor servers in the middle of 2007,” said Northington.
Quad-Core http://blogs.intel.com/it/2006/12/quadcore.html >Frankly, the entire team is very impressed at the scaling we are seeing relative to >the Woodcrest based servers (in the 1.5-1.7x range on an average ? more about our methodology later) >for a variety of typical Business computing workloads (OLTP, DSS, Reporting services and the like). >Over the last several weeks, it has become blindingly obvious to us that these processors pack >a tremendous amount of computing power in a very cost-effective “footprint”.
Bloomfield To Hit > 4GHz & Has 8MB Cache http://www.vr-zone.com/?i=4387 Over 4GHz L2-8MB*1 1S/4C/8T TDP130w HyperThreading(AsymmetricSimultaneousMultiThreading?)
>>327のWolfdale, is releasing in the end of this yearって話は、Prescottの時のようなFMBの変更でもない限り遅延はほぼありえないわけで、とりあえず信じてよさげ。 そうなると、俺は>>159のfab28が08H2→08H1ってネタはせいぜい1,2ヶ月早まった程度だと認識してたわけだけど少なくとも3,4ヶ月は早まったとみてよさそう。
流石にPenrynのES品はまだらしい。E6850は写真があるな。 http://www.vr-zone.com/?i=4507 >Will it support 45nm Penryn then? > >I can't determine that because I haven't gotten that CPU yet.
http://www.itjungle.com/breaking/bn012207-story01.html >Otellini referred to Sun as the strategic, mission critical Unix for Xeon processors, being very careful not to snub HP and its HP-UX for Itanium processors. >HP-UX does not run on Xeon chips, and this partnership with Sun could possibly encourage HP to rethink this strategy.
http://www.intel.com/technology/silicon/45nm_technology.htm?iid=homepage+42nm >For its 45nm technology, Intel is using a hafnium-based high-k material in the gate dielectric. >The high-k dielectric is created using atomic layer deposition (ALD) whereby a single layer of the high-k material molecule is deposited at a time.
http://www.eetimes.com/showArticle.jhtml;jsessionid=PPCOSGQHKXJS0QSNDLPCKH0CJUNN2JVN?articleID=196602084 IBMはultra-low-kの方向で進んでいる。 で、プレスリリース読んだ限りじゃ「high-kや歪siやSOIといった技術をすべてインテグレートした実弾を開発した」 とは言って無いからこのeetimesの記事と比較すると本当に株価対策専用って感じだ。 こんな醜聞記事まであるし。 http://www.theregister.co.uk/2007/01/28/intel_ibm_highk/ >Big Blue heard of Intel's plans to announce the High-K + metal gate transistors. >It managed to secure a pre-press version of Intel's news release, which was originally scheduled to hit the news wires on Jan. 29. >So, IBM quickly wrote up a news statement of its own and handed it out to reporters >under what's known as an "embargo" where reporters promise to release all of their news stories at the same time.
1) ノートPC上でWindows VistaとOffice 2003を実行しているデュアルコア モバイルプロセッサ 2) Windows Vista上でHDV 1080iを再生しているデュアルコア デスクトッププロセッサ@2.13GHz 3) Windows Vista上でRainbow Six Las Vegasを実行しているクァッドコア デスクトッププロセッサ@1.86GHz 4) Windows 2000 Advanced Server上でGlaze Workstation applicationを実行している2基のデュアルコア プロセッサ@2.16GHz 5) Windows Vista上でAdobe Premierを使ってビデオをエンコードしている2基のデュアルコア プロセッサ@2.16GHz
Penryn does not have HT http://www.theinquirer.net/default.aspx?article=37316 >Penryn does not have HT, nor will it ever. (中略) >More than enough engineers, Lenovo sales people in outer Mongolia and the usual rabble picketing my house all confirm that it is not there. 一行で書くと、PenrynとHTTに関連性が皆無なことをエンジニアは当然としてLenovoのセールスマンから野次馬までいろいろな人が確認したとのこと。
>>415 >It is also likely to include extended visualisation, release 3.0 of iAMT, >and will come with additional instructions to improve text processing and string search.
アナリストから見たDell CEO交代劇 http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=KXWWPVCZGWH52QSNDLSCKHA?articleID=197006371 There are other bad signs for AMD.''We see Michael Dell's return as CEO at Dell as a negative development with strong ties to Intel,'' he said. ''OEM relationships are straining to stay engaged with AMD at this time given its stale product line-up. We continue to think AMD needs to get new products out to regain a competitive stance.''
「Intelはすでに90nmプロセスで製造されたデュアルコアItanium2において24MBのSRAMをキャッシュとして利用しており、45nmプロセスでは更に多くのメモリをチップに置くことができる。」 Bohr(A 1.1GHz 12μA/Mb-Leakage SRAM Design in 65nm Ultra-Low-Power CMOS with Integrated Leakage Reduction for Mobile Applications @ISSCC2007 の共同執筆者)は上記の事実を強調して言いました。
http://www.theinquirer.net/default.aspx?article=37772 >He said that Intel will be able to produce better bins and bus rates during this year. > >He also said that Intel was likely to disclose its CSI plans at the Beijing April Developer Forum.
Tukwilaは45nmか? http://news.zdnet.co.uk/emergingtech/0,1000000183,39286049-1,00.htm?r=36 >The first realisation of that is Tukwila [quad-core Itanium] in late 2008, the next step in the product family, where we move to common system architecture elements, as well as full alignment on design tools and process.
結構イイ線いってたっぽい http://seekingalpha.com/article/24929 >Dirk R. Meyer > >Of course, it’s hard for me to speak to precisely what the pricing environment will be, but from a cost basis, >you can think about Barcelona being not that much different than the introductory dual-core Opterons.
http://www.vr-zone.com/?i=4721 >Boards development for its successor "Eaglelake" are going start soon and it will be pairing up with ICH10 Southbridge to support higher FSB Penryn processors. Yorkfieldの次に1ダイ版の4コアが計画されているのかな? Wolfdal/Yorkfieldと同じく2コアがFSB1600で4コアがFSB1333というパターンもありうるか。 何時の間にか話題に上らなくなったRidgefieldというコードネームもあるしデスクトップにCSIが降りてくるのは2009年?
http://www.channelregister.co.uk/2007/03/05/amd_channel_miss/ >"We did lose share in the channel to our competitor," Ruiz said, when commenting on AMD's most recent quarter. 〜 >And facing lower average product prices seems a more realistic explanation than slumping channel shipments for AMD's sales miss.
> Patterson: It's a shocking statement, but the era of faster sequential processors is over. > All hardware companies rely on parallelism for performance, > and there are no plans for fast sequential processors.
Morgan Stanley Technology ConferenceでのOtellini CEOのスピーチをネタに InfoWorldのTom Yagerさんが書いた記事の感想部分を、inquirerの中の人が 意訳したのが "seeking to kill AMD by forcing prices down"
NVIDIA Getting Ready For Shader Model 5.0 GPU http://www.vr-zone.com/?i=4768 a new NVIDIA GPU by late 2007 that supports double precision arithmetic so that could be G90.
http://www.theinquirer.net/default.aspx?article=38232 >Pat Gelsinger confirmed it 仕様は>>415-431辺りのやつで確定??? >In the end, it works out to be IMCs on Xeons and EE/XE, and nothing else. 初期のCSIはXeonとCore 3 XEのみに提供される。(=very high end, >>510) >IMCs come in two flavours, possibly more if the moon is in the seventh house and Intel does something odd with the EE/XEs. 2種類のIMCがデビューする。(MP向けとDP向け? あるいはDP+向けとUP向け?) the moon is in the seventh house(宝瓶宮? 1Q09?)にCore 3 XEに変り種のソリューションが追加されるかも知れない。(V8の後継か?) >Add in sockets aplenty, and you have Intel circa Q4/2008 and on. そのまんま。
http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=FASQ15F4WOPHYQSNDLPCKH0CJUNN2JVN?articleID=198100065 >Intel is also expected to introduce new VBI form factors as a way to give system builders more flexibility for designing ultra-mobile PCs. >The current VBI offerings are for 14.1-inch and 15.4-inch form factors. >Yet Intel is expected to unveil 12-inch, 13-inch and 17-inch form factors and announce more manufacturers supporting VBI, sources said.
Transaction Processing Council Launches TPC-E Benchmark http://www.itjungle.com/tfh/tfh031907-story07.html >"People are doing more complicated things in memory and CPUs than they did in the past, and the TPC-E test reflects this," says Molloy. >"Customers are keeping track of more data and they are making more correlations between data sets."
http://www.theregister.co.uk/2007/03/20/intel_labs_mario/page3.html >We spied one board code-named Coalbrook (or CoalCreek: it was hard to tell as Intel's staff urged us away from the confidential systems) and another called Springville that had built-in optical modules. >Both systems were identified as using Intel's upcoming CSI (common system interconnect) technology, which is the company's attempt to catch-up to AMD's Hypertransport/integrated memory controller technology. 私達はCoalbrook(CoalCreek?)というコードネームのボードとSpringvilleというコードネームの内臓光学モジュールを有するボードを嗅ぎ当てました。 どちらのシステムもCSIを利用することが確認されました。CSIはIntelがAMDのHypertransportと統合型メモリーコントローラーに対抗するために開発されている技術です。
<メモ> http://www.theregister.co.uk/2007/01/27/intel_silicon_modulator/ >Intel's researchers have turned out a silicon optical modulator that can encode data at 30 gigabits per second, making it the fastest such device on the planet. >That speed notches Intel closer to rivaling the 40 gigabits per second of non-silicon devices used today in the fiber optic world. >And it places Intel on the cusp of delivering fast, cheap networking hardware capable of rack-to-rack, server-to-server and chip-to-chip communications. (中略) >At the moment, a pricey non-silicon modulator can cost around $8,000. >Intel thinks the price for a similar product made with silicon needs to get to less than $10 for it to be a cost effective server part.
http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=198100278 Intel CEO Paul Otellini told partners Monday that the company now has five 45-nm processors running on six operating systems: a notebook dual-core processor, dual- and quad-core desktop processors, and dual- and quad-core DP server processors.
"The new UltraMobile CPU is in addition to the first five," he said. "The first UltraMobile devices we'll see in the latter part of the year."
最初のCSIは銅配線? >Realistically, I don't see a way that the first iterations of CSI could be optical. >This optics break through was very recent - and I'm pretty darn sure the first products using CSI had already taped out by then. >I don't see a reason why either. As I mentioned, copper's fine for the data rates that CSI will run at, both initially, and over the course of the following 4 years (I'm assuming a start at 6.4gbps and climbing to ~20gpbs after 4 years).
45nm?32mn?22nm?MCP? >The reality is that CSI won't be optical at 45nm, because Intel isn't using the right materials. >Everyone who works with optical stuff uses SOI. Bulk silicon just doesn't have the right properties AFAIK. > >Intel has said they might consider SOI or other materials at 32nm, but not 45nm. >Alternatively, I suppose they could do a MCP, where one chip is the optical interface to the outside world, which is connected to the MPU by a very very very high speed copper interface.
チラ裏 >Optimal wavelength for silicon is around 1330nm >When using a InP coupled to Si structure, about 80% of the light from the InP will go to the lasing chamber (requires SOI silicon)
http://www.theinquirer.net/default.aspx?article=38566 >In addition the most interesting bit is the advances on loads and stores. >The old way of dispatching speculative stores would stall when the data would cross cache lines. > >Penryn removes this limit. >You can cross cache lines without introducing a long wait, very handy for a lot of multimedia apps that don't use regular accesses to memory. PenrynではStore Forwardingが改良される。
>The core count itself goes up quite a bit, with Beckton being listed as having four, six and eight core with TDPs ranging from 90W to 130W. six - core?Nehalemにはsingle-dieの8coreが計画されている? BecktonのFBD Channel l4+1の「+1」は、6coreや8coreの際にメモリバンド幅を増強できるという意味か?
>Since this is not set to be pulled on die till Gesher, I can only take a semi-educated guess that they mean on package graphics, not on die.
http://www.realworldtech.com/page.cfm?NewsID=370&date=03-28-2007 >However, Penryn is not just a shrink to 45nm, there are several specific enhancements to improve performance. According to Intel, > their initial performance analysis shows that a 3.2GHz Penryn system using a 1.33GHz front-side bus is 20% faster on gaming workloads than the current high-end 2.93GHz Conroe with a 1.03GHz bus. >For more bandwidth and floating point intensive workloads, Intel claims that a 3+GHz quad core based on Penryn, with a 1.6GHz bus will see a 45% improvement relative to a 2.67GHz/1.33GHz Clovertown server system. >This is most likely measured using the SPECfp_rate 2006. 最初のシリコンが2GHz超で動作しただけに、既に3.2GHzで動いているのか、それともシミュレーションなのか興味深いところ。 後者の45%は単なる想定値っぽいな…
今回のanandtechの記事はRWTのものと比較して粗が多いけど、適当にメモ http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2955&p=1 >The quad core version of Penryn contains 820 million transistors (Kentsfield has 582 million) in two very small dies of 107mm2. >That makes the new design 25 percent smaller than Intel's current 65nm Quad core (143 mm2). Penrynのダイサイズは107平方mm。
>Finally, Intel mentioned that in the server space, the fastest quad core Penryn available (>3GHz) vs. a 2.67GHz quad core Xeon resulted in a greater than 45% increase in performance in "bandwidth and FP intensive applications". >It's incredibly vague (and oddly similar to AMD's claims of Barcelona vs. Xeon performance), but Pat mentioned that STREAM and certain benchmarks in SpecFP could be considered to be "bandwidth and FP intensive". 3GHz以上のPenrynはメモリインテンシブなアプリでClovertown 2.67GHzと比較して45%性能が向上。 Penrynのスペックが曖昧すぎる上にAMDのClovertownに対する主張と同じなので話半分に聞いておけ、と。
CSIはHypertransportより低レイテンシ http://www.theregister.co.uk/2007/03/28/intel_nehalem_deets/ >The chip maker is expected to introduce CSI in its Itanium family in 2008 as well and has told some customers that the technology shows "much lower latency" than AMD's Hypertransport.
VPがGesherは2009年と発言。前倒しか? http://www.digitimes.com/mobos/a20070330VL201.html >Stephen Smith, Intel’s Digital Enterprise Group director stated, motherboard makers have started the development of Penryn-based platforms and will launch at the same time with Penryn processors. >The 32nm processors codenamed: Westmere and Gesher will launch in 2009, Smith added.
Intel lists role models http://www.fabtech.org/content/view/2643/ >Preferred Quality Supplier (PQS) awards, which highlight those who are considered to be doing a good job for the Intel cause. >The list below shows this year's top dog suppliers, but it seems the list is about double that of the previous year! > >That means that more suppliers have reached or passed the points system award requirements than did last year. >I would assume that Intel is pushing suppliers harder than ever to perform, and that strategy seems to be working rather well!
Intel debuts two new Centrino logos! http://www.theinquirer.net/default.aspx?article=38679 >In 2008, you need a Penryn, Cantiga GM or PM and ICH9M to get the logo. >This will basically bring AMT to v4.0, a new spin of VT, and up the FSB from 800 to 1066.
Xeon chip prices to fall through the floor http://www.theinquirer.net/default.aspx?article=38687 >Intel is eager to introduce the Clovertown in Q3 2007 and it will be called by the enigmatic name the X5365 and be a 3GHz quad core. >Come July we will see reductions on Intel's DP line and the X5365 will cost over $1,100.
>また、サン・マイクロシステムズ社は Solaris オペレーティング・システムをインテル Dynamic Power テクノロジーに対応したインテル Xeon プロセッサー 5100 番台上で稼動させるデモを行いました。 >インテル Dynamic Power テクノロジーはメモリー・サブシステムの消費電力削減を実現する技術です。 "Dynamic Power Technology"って初耳(本家で検索してもヒットしない)。 詳報待ちだが仮にFB-DIMMの消費電力削減が実現できるのなら喜ばしいことである。
Intel tips more details on Itanium roadmap http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=HMHUPMYPFSQY2QSNDLSCKHA?articleID=199100362 >The new Itanium CPU is expected to sample this year and will also use a form of simultaneous multi-threading expected to support two threads per core. (中略) >Intel will follow up Tukwilla with Paulson, which may use between 6 and 10 cores in a CPU that has no announced shipping date yet. >It is expected to double performance of Tukwilla. > >Intel is likely to use 65 and 45 nm process technology respectively for the two Itanium generations, although it has not yet announced in what processes the chips will be made. TukwilaのMT技術はSoEMTではなくSMTなのか? Tukwilaが前世代の増コア版でPoulsonが新アーキというのが定説だったが逆の可能性も出てきた。 (>>477-481でPoulsonが意外に早いという話もある。) 6-10コアということでどちらが新アーキか現時点ではわからないがItaniumも新アーキはNehalemの様にスケーラビリティ重視の設計の公算が高い。
The Long Awaited Penryn Update http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2968&p=2 Finally, with regards to motherboard support, Intel isn’t making any guarantees about Penryn’s backwards compatibility. While Penryn will still use the LGA-775 socket that Prescott and Conroe have used, motherboard support will require more than just the presence of the socket. If the appropriate VRM spec is implemented, then Penryn will work on your LGA-775 motherboard, the problem is that motherboard manufacturers haven’t yet released information on which of their boards will support the Penryn VRM changes. If history repeats itself, you can expect very limited official support for Penryn in currently shipping motherboards and guaranteed support with boards based on Intel’s new 3 series chipsets (e.g. P35). We did see Penryn up and running on an Intel BadAxe2 board, but it had a hardware VRM modification done to it in order to properly support Penryn. Penryn may also be able to work on boards without a VRM mod, however at increased (potentially out-of-spec) voltage settings.
http://pc.watch.impress.co.jp/docs/2007/0420/kaigai353.htm 「Super Shuffle Engineは、レガシーSSE命令を含めてSSEタイプの命令の実行を助ける。 基本的に、SSE4で得られるパフォーマンスは、Super Shuffle Engineなしには達成できないだろう。 Super Shuffle Engineは、SSE4命令のパフォーマンスに直接貢献している。 だから、我々はSuper Shuffle EngineとSSE4命令の2つを同時に実装した。 一方、Radix-16とSSE4命令については、もっと独立したものだと考えている。 ソフトウェア側からの要請があって実装したが、SSE4命令との関係は、(Super Shuffle Engine)より独立したものだ」 --Stephen L. Smith (Vice President Director, Digital Enterprise Group Operations, Intel)
> based on the SPECcpu2006 benchmarks show AMD's Barcelona processor will have >"up to a 50 percent advantage in floating point performance > and 20 percent in integer performance" > over Intel's high-end quad-core chip running > "at the same frequency,"
かなり荒っぽく計算してみたがDynamic Power TechnologyはFB-DIMMの弱点の一つを大幅に改善してくれそう。 San ClementeはDDR2でThurleyに至ってはDDR3ということだがFB-DIMMは終わってしまうのだろうか? サーバーやデスクトップではそこまでシビアな電力効率は要求されないと思うんだけど。
Rumors about 45 nm Core 2 Duo production begin to fly http://www.tgdaily.com/content/view/31804/135/ Intelはこれ(↓)と同じことを(去年は7月14日だったのだが一日遅らせて)7月15日に画策しているらしい。。。 今年中に全セグメントに45nm製品を投入するという話もあるし量産前倒しのルーマーは大スクープだった?
HP Turns in a Solid Fiscal 2007 Second Quarter (>>665) http://www.itjungle.com/breaking/bn051607-story02.html >Even though Itanium-based Integrity server sales rose by 60 percent in the quarter, > sales were not sufficient to counterbalance declines in legacy AlphaServer and HP 9000 sales. >(The Integrity machines support HP-UX, Windows, Linux, OpenVMS, and NonStop platforms.)
NVIDIAが『45nmプロセスのx86チップをOregonで』開発中というルーマー記事 http://www.theinquirer.net/default.aspx?article=35216 >And now we hear that development is underway at Nvidia's just-announced >Portland, Oregon, Design Center, where chip folk are beaving away on 45 nanometre designs.
>携帯機器向けの省電力プロセッサなどで、毎回質問に立ってコア電圧とTDPをしつこく >聞いていたのがIntelのBob Jackson氏。「今は一体何やってるんです」と聞いたところ、 >案の定「初日にMark BohrがLow Power Mobile Processorの話してただろ。あれだよ」 >という返事。Low Power IA and Technology Groupに所属されているそうで、なるほど >Intelはこのマーケットを真剣に考えているんだなぁ、としみじみ思った一幕だった。
Intel discusses quad-core processor server market: Q&A with Boyd Davis, general manager of server platform group marketing of Intel http://www.digitimes.com/mobos/a20070608PD206.html >The company expects to mass produce the processor in the fourth quarter this year. Harpertownの量産出荷開始は4Q07。
表スレを見て、その後ちゃんと>>321を読んで(今更)気づいたこと。 Lucid Information Technologyという会社について調べてみると http://www.lucidlogix.com/About.html >parallel architecture that turns traditional off-the-shelve graphic cards, >graphic processor units (GPU) and graphic cores into powerful and scalable visualization solution, >transparently to the applications, enabling price-performance ratios never considered possible before. 「既存のグラフィックスコアを用いたパラレルアーキテクチャで高いコストパフォーマンスを達成する」と書いてあるではないか。
なーんだ。Larrabeeじゃないのかよ。しかもGMA X3500(X3100?)の使い回しかよ。ツマンネ。 等と思うなかれ。>>321に下記のような表現がある。 これを見るとハードウェアでロードバランシングを行うのだと予測できる。 >The chip acts like a hub connecting the main processor to the multiple graphics processors, serving as a traffic cop. >The chip helps divide the graphics processing work among several GPUs 当時この記事を読んだときには「ドライバが鬼門ではないか?」と思ったのだが、 >>727に「VRAMをメインメモリとシェアしなければ良い」という記事もある。 なかなか興味深い製品になりそうだ。
751 :名称未設定 :2007/06/01(金) 19:58:43 ID:nbv1FNiv0 Intel set to announce graphics partnership with Nvidia? http://www.tgdaily.com/content/view/32282/137/ なんとLarrabeeはIntelとNVIDIAの共同開発で、早ければ6/26のISC2007で詳細が公開されるらしい。
Itanic still rides the waves http://uk.theinquirer.net/?article=40372 > Talking about Tukwilla, besides the four cores and up to 6-channel CSI interconnect setup > with integrated memory double-device error correction and mirroring options, > it is confirmed that the cores themselves will be substantially reworked with supposedly much higher efficiency per clock, > both per thread and in simultaneous multithreading. >>643のeetimesに続いてinqもTukwilaはSMTだと言っている。 しかもTukwilaではコアも刷新され、クロック辺りの性能が高くなるとも。
>>831 10GbEとInfiniband両対応 http://www.intel.com/pressroom/archive/releases/20070627corp.htm?iid=pr1_releasepri_20070627r >Intel also unveiled an innovative new technology that will allow organizations to build bigger clusters with improved performance. >Intel Connects Cables enable Infiniband and 10 GbE customers to achieve data rates up to 20 Gbps and extend the reach between servers up to 100M.
Intel postpones Turbo Memory for desktop PCs http://www.digitimes.com/mobos/a20070628PD212.html >During Computex Taipei 2007, Intel released its plans to transfer Turbo Memory to the desktop and scheduled to begin shipping version 1.0 in the third quarter. >However, as the performance of Turbo Memory 1.0 has not been as expected, >the company has decided to cancel version 1.0 and will directly launch version 1.5 by the end of 2007 or first quarter of 2008, noted the sources. Intel Turbo Memory for desktops is on target >But that's version 1.0 of Robson and it appears there were always going to be technical difficulties implementing RAID support for desktops. >>741>>814
AMD announces Barcelona http://www.theinquirer.net/default.aspx?article=40680 >What you get is better than AMD's current offerings but not much to threaten Intel sales. 長い記事だがこの一行だけで事足りる。 この時期にプレスリリースだけが出るってことはやっぱり第2四半期も赤字だったのか?
Intel begins shipping entry-level P31 and G31 chipsets to motherboard makers http://www.digitimes.com/mobos/a20070705PD210.html >Intel also plans to ship its performance IGP chipset G35 and high-end chipset X38 by the end of July in order to catch the sales peak during the summer time, added the sources. > >Intel plans to increase the proportion of its 3 series chipsets to 50% of total chipsets shipments by the end of 2007 and 60% in first quarter of 2008. >In second quarter of 2008, the new Eaglelake chipsets will take the 3 series' market position, while 945 chipsets will take 865 chipsets' position, added the sources.
Intel phasing out Dempsey and Paxville MP server processors (>>410) http://www.digitimes.com/mobos/a20070706PD211.html >Intel plans to announce the phase out date for its Xeon 5000 series (Dempsey) 2-way server processors in July, >while its Xeon 7000 (Paxville MP) series 4-way server processors is scheduled to be phased out by August of this year, according to sources at server makers.
Intel Pushs Nehalem Into H2 2008 http://www.vr-zone.com/?i=5113 > Intel has revealed some information on their Nehalem processor in the latest roadmap updates. (中略) > Nehalem-EP processor will arrive earlier than expected in H2 2008 最新のロードマップではNehalemの投入時期が若干前進している模様。 (>>536) > Tylerburg chipsets チップセットのコードネームはTylersburgではなくTylerburg? > EP stands for Efficient Performance, one of the new naming for the server platforms. Nehalem/Westmere世代のXeonDPやXeonMPはEPというクラスに属する。 既報の通りスケーラビリティー重視の設計なのでMCHやGMCHを統合する/しない等で色々派生していくと。
PTI said to have landed NAND flash TSOP orders from IM Flash http://www.digitimes.com/bits_chips/a20070712PD218.html PTIはIMFTからTSOPを受注しはじめた。 業界関係者の話では、PTIは7月から月当たり100〜200万の注文を受けているという。 また、現在SamsungとHynixはAppleへの供給を優先している。 その一方で市場にはNANDフラッシュに根強い需要があることから供給不足が起こることが懸念されるという。 IMFTはmicroSDカードの供給を一時停止しチップだけを供給する予定。 IMFTは粗利益率における優位性がこの方針を選択する主な理由であるとした。
Intel Capital Invests US$65M in Taiwan-Based Powertech Technology, Inc. Funds to be aimed at growth within memory component industry http://www.intel.com/capital/news/releases/070228.htm >PTI focuses on assembly and testing services on memory products. >PTI currently provides wafer testing and stacked-MCP, mold uBGA, wBGA, TFBGA, TSOP, micro SD card, assembly, testing and burn-in testing services for various memory devices.
>>872 Intel New Nehalem & Tukwila Platforms In H2 2008 http://www.vr-zone.com/?i=5115 Itanium2 → MC (チップセット Boxboro) Xeon MP → EX Xeon DP → EP (チップセット Tylersburg-36D), EN (チップセット Tylersburg-24D) Xeon UP → EP (チップセット Tylersburg-36S)
Intel prices its first 45nm processor at US$999, to hit the market in 4Q http://www.digitimes.com/mobos/a20070719PD203.html >In addition to the quad-core Core 2 Extreme processor, >there will be two performance level quad-core processors adopting a 45nm process launched soon after, >with both processors adopting a 1333 MHz FSB and 12MB L2 Cache.
Intel Prices "Penryn" Xeons http://www.dailytech.com/Intel+Prices+Penryn+Xeons/article8074.htm Q3のラインアップ(>>537)とQ4の新ラインアップの比較。 200〜300ドル台のXeonに力が入れられている感じ。 値段 クロック比較 FSB比較 MN $1,172 ↑166MHz ±0 X5460 vs. X5365 $851 ↑333MHz ±0 E5450 vs. X5355 $690 ↑166MHz ±0 E5440 vs. X5355 $455 ↑333MHz ±0 E5430 vs. E5345 $316 ↑500MHz ±0 E5420 vs. E5335 $256 ↑466MHz ↑266 E5410 vs. E5320 $209 ↑400MHz ±0 E5405 vs. E5310 $320 ↑466MHz ↑266 L5410 vs. L5320 注:E5405のスペックは推定で、これより高くなる可能性がある。 E5405 2.00GHz FSB1066MHz 6MB*2
PCI-SIG Announces PCI Express 3.0 Bit Rate For Products In 2010 And Beyond http://www.pcisig.com/news_room/08_08_07/ >These protocol extensions, which are being defined by the PCI-SIG to expand the usage model and adoption of PCI Express architecture in new and emerging applications, > are intended to optimize interconnect latency and platform resource efficiency by introducing new specifications for data reuse, > dynamic power management,and related optimizations in the I/O hierarchy. これ、Geneseoっぽいね。
NMI-window exitingの説明 http://www.intel.com/technology/itj/2006/v10i3/1-hardware/8-virtualization-future.htm ・NMI-window exiting. The interrupt-window exiting VM-execution control (described earlier) causes a VM exit when a guest is ready for maskable external interrupts, allowing a VMM to deliver such interrupts in a timely way. NMI-window exiting provides corresponding support for non-maskable interrupts (NMIs), which are blocked by other conditions than those that block maskable external interrupts.
Intel lines up nine 45nm CPUs for the desktop http://www.digitimes.com/mobos/a20070817PD207.html > With Intel aiming to eventually scale the 45nm range up to a maximum core frequency of 4.0GHz, > the sources estimate that at least four more CPUs will appear at a later time with frequencies higher than the initial 3.16GHz. Xeon X5460が3.16GHzで一息ついている感の有るIntelですが、 量産サンプルを触っているマザーボードメーカー関係者の予想だと、 3.83GHz位まではリリースされるのではないかということです。
Intel and Symantec Team Up for On-chip Security. http://www.xbitlabs.com/news/cpu/display/20070816110103.html Windows Server 2008ではHypervisorが統合され所謂準仮想化がWindowsでも出来るようになる。 現行の、OSより上のレイヤで動作する(= 単一のOSイメージでシステムを運用することを前提とした)セキュリティソフトは、 準仮想化や完全仮想化によるサーバー統合が一般的になると時代遅れな代物になるだろう。 SymantecはHypervisor上でLinuxやWindowsといったOSイメージと並行して(= 同じレイヤで)動作し、 直接ハードウェアと対話することでシステムのセキュリティを保持するソリューションをIntelと協力して開発している。
>>971 (まだちゃんと読んでない) "Initial CSI implementations in Intel’s 65nm and 45nm high performance CMOS processes target 4.8-6.4GT/s operation, thus providing 12-16GB/s of bandwidth in each direction and 24-32GB/s for each link."
Intel to deliver X38 chipsets in mid-September and probably X48 parts by year-end, say sources http://www.digitimes.com/mobos/a20070914PD207.html > The X48 eventually will come with the same chip design as the X38 but will support 1600MHz FSB and DDR3-1600 memory, the sources stated.