[Music] welcome to the kaby event earlier this year at comput test 2024 we launch a revolutionary solution that can realize local AI training and F tuning on your desk kab AI top open new possibility for, of AI application developer and demonstrate total integration of Highway and Subway today in just four months we have more to share we are going to introduce a strong line up of new product new features and the new technology let enable capabilities and flexibility we have made it more convenient for people to fting their models and the gab airp solution we are also excited to unve the next generation of multiple platforms they are denied to push more performance of latest processors and the dt5 memory by AI driven development and engineering now here's Alis to tell us more about gigabyte ai [Music] typ ai top is the all around solution designed to train your own AI on your desk it features a variety of Innovative and groundbreaking Technologies across hardware and software that can be easily adapted by beginners and professionals now with the AI top utility 2.0 we expand its capabilities to provide a complete workflow of AI training from using Rex to check which model to be selected for your project to creating data sets for fine-tuning and in app validation after fine-tuning your own AI model the AI top utility 2.0 is designed to take local AI funing to the next level first let's start with model selection since every model consisted of different patterns of parameters and pre-rain with different data sets it used to cost a lot of time validating results after each trending session now with the rack check you can easily use your ra data as RS to check which model works best as expected without actually fine tun in it this new capability significantly saves your time for preparation next let's talk about data sets when training llms using a data set without proper feature engineering or cleaning can lead to lower quality inference due to noise and irent data on the other hand a data set with well structured features significantly enhance the model's ability to understand and generate accurate responses to address this our software includes a feature that enables user to efficiently detect the language and knowledge field of later data set access the condition of their PC hardware and leverage llms to automatically generate a well organized Q&A data set this curated and structured data set can be directly used for fine tuning resulting in improved inference quality and more accurate model outputs in the AI top utility 2.0 we feature an
inapp data set creating tool that allows you to use a llm to prepare the data sets for you you only need to choose the raw data and select the llm you want to use and within a few clicks the data set preparation is now easily in progress we also bring a huge expansion to the capability of f the AI top utility 2.0 now supports large multi modile models or lmms models like llama 3.2 11b vision and pixstar 12b Excel by processing diverse inputs such as text images and audio enabling more intutive itive interactions and handling complex tasks like visual question answering and image captioning in the coming year multimodel models will become more efficient accessible and widely applied transforming Industries like healthare entertainment and customer service with more context aware AI Solutions you can now train or F you an lmm with data sets of text images and video making your AI model more capable than ever in generating images or even short videos instead of PL texts lastly we introduce the Ina validation feature in the AI top utility 2.0 you can now deploy the AI model you just trained or finetuned for inferencing and see if the outcome is optimal with inapp validation the complete workflows of AI training is now fulfilled as AI models grow in size and complexity including the integration of lmm the demand for Hardware resources increases significantly to mean this need we introduce the AI Top Class leing feature allowing you to connect multiple AI top machines via internet or stable boards this enables faster F tuning and quicker results providing enhanced computational power and efficiency in a multi 2pc setup trending speed is increased by up to 1.6 times significantly reducing overall training time while memory capacity is expanded to support larger model size for more efficient trending this is how clustering works the workload of trending and llm is split across two machines allowing them to share computational and memory resources each PC handles a portion of the model's parameters and data effectively increasing the overall capacity to train larger models this setup helps overcome the limitation of a single machine GPU memory enabling the system to support model that require more memory than what a single node could provide by Distributing the model and data across multiple nodes multi node training allows for scaling to larger LMS that wouldn't fit in the memory of just one PC the AI top clustering supports two connection methods a 10 GB ethernet standable 4 and the most Cutting Edge standable 5 10 GB ethernet provides longdistance cluster Computing when two AI top systems are away from each other which is very friendly to let electrical system however when Electric Supply isn't a concern we could place two system close utilizing Intel standable 5 networking to speed up the fting process and finally we are bringing the AI top utility 2.0 to the largest group of PC users upon the release of the AI top utility 2.0 it comes with the support of
Windows subsystem for Linux or WSL the AI top utility 2.0 will be able to run on your Windows machine once the WSL is installed and the setup kit will be included in the package this can be a quicker and easier way to have a glimpse of local AI training or fine-tuning before you switch to Linux for full capacities and performance last the AI top utility 2.0 with Rex checking data set creating tool new support for lmm and in app inferencing for validation multin clustering and support for WSL the war flows of local AI training and fine tuning have never been so friendly before to maximize this software potential we've developed equally powerful Hardware our new AI top Hardware lineup is designed to perfectly complement the local AI training capabilities let's start with a crucial component in AI training SSD traditionally the workloads of data set processing during AI training or F tuning for majoring on vrms now with the memory uploading technology brought by the aop utility the massive amount of data can now be uploaded to the ssds meaning that you can upscale the capacity of model size with much lower cost simply adding up the ssds AI fine tuning gives tremendous data offloading between VM DDR and ssds meaning that an intensive ride and rebeh behavior will occur on the N flash this will wear out the SSD live in a very short period of use that's why we are introducing the AI top UD SSD 1 Tab and 500 GB build based on gigabyte Ultra durable technology to address this issue the AI top UD SSD delivers up to 6,500 megabytes per second for sequential read and up to 5,700 megabytes per second for sequential ride ensuring Ultra fast data processing in terms of of terabytes written or tbw the AI top UD SSD 1 tab features up to 10 19,500 tbw and if you are looking for more advanced solution for your memory hungry machine here's the AI top cxl card to meet your needs cxl Compu express link is a breakthrough technology that expands the size of system memory by leveraging the PCI 5.0 interface cxl can significantly improve performance in memory Bond applications with the laptop CSL car you can add up to 512 GB of ddfi Ting capacity to the system memory space when fine-tuning the Lama 2 33b model with full Precision utilizing the AI top C XL could delivers us up to, 22 tokens per second Which is higher than double the number of only using ssds to offload the data it is also built with an active cooling system to maintain the stability under heavy AI workloads the AI top cxl card is the one device that takes the capacity of viable model size to the next level let's now talk about the PSU and its power efficiency introducing the ap600 TM pg5 AI top it is a power supply unit that helps you save more power supports up to four PCI G 5 graphics cards and is tested on the strict simulation of heavy AI workloads it is also certified as at plus titanium which is by far the highest level that ensure the best efficiency and there is even more power meter it's a new feature in the AI top utility that appears only when pair with the ap600 TM PGI AI top you can see power consumption by kilowatt hour which is an accumulated number that begins to be calculated every time the PSU turns on this helps you to better keep track of the expense on electricity and you can also see current power consumption so you can better manage the system workloads for optimized efficiency this is also shown on the power Supply units LCD monitor as total output it's power saving units that helps you optimize and monitor the power efficiency speaking of optimizing the entire system let's move on to another critical component the chassis introducing Oro C 500g AI top and gigabyte c71g AI top they both support up to eatx motherboards and up to four two slot graphics cards the c71g AI top supports to 4 360 radiators up to 2 power supply units for higher power demanding solution more Drive Bas for your large AI model storage demand more air flows for your system to stay cool during heavy AI training workloads there are simply no better places than the air toop ches to build the AI training machine tailored to your needs and speaking of your needs that's where the AI top setup keep comes in introducing AI top 100 and AI top 500 setup kits the AI top setup kit contains a mod board a power supply unit a chis and and ssds as the bare bone for your AI training machine the AI top 100 starting from 1699 is the ideal startup solution to build a personal AI training machine with up to two compute cards capable of handling full Precision fighting of ab llm and larger 100 plus billion llm with Laura with the supportability of more computer cards the air top 500 starting from $24.99 is perfect for small business education research as the all around solution to finding a larger 40 plus billion llm model with full precision and is capable of handling 45 billion LM with Laura the AI toop setup kit will be available in Q4 2024 back at Compu TX 2024 we launch two AI top motherboards the TRX 50 a top and the w790 AI top for the first time ever we are bringing the innovative solutions of AI top to the latest generation of motherboards here is Garrett to tell about these new members of AI top hardware and the latest generation of [Music] motherboards yeah look my going crazy I got break me I'm too dop like I'm back in the 0s got a lot of play ain't coming down got toate I know my next win [Music] the the we are exacted to introduce you The Cutting Edge orus z890 series modable a true Game Changer in the world mbls first we have D5 bionic corser this is our Advanced AI enhance overclocking technology for dd5 memory leveraging AI models user can rapidly achieve Pak memory performan through a feature we call AI sned we also employ AI sistic tools to streamline the multiable development process including AI driven PCB design on the hardware side and the hyper tune bies on the phunware side through this Technologies the oros DN achieve a standing ddfi speed of 9500 Mt per second let's start with one of our key Innovation the oros AI snitch this Innovative feature leveraging the neural network models to analyze the Hardway configuration in real time directly within windows with just one single click you can maximize your memory performance for example it can take native 6400 cud memory and instantly create 8933 overclocking profile this means you can achieve top tier performance with minimal effort making highend over Target accessible to everyone at the core of all a s it's the true AI overclocking model we build this using two neural networks and a massive amount of real world data we took thousands of memory testing result and match them with the hardware stup to create a robust data set this data was used to train our AI teaching it to understand how different Hardware configuration affect memory performance we then F tune the AI by constantly comparing its prediction to a result from the expert manual tuning this process lead to the creation of AI snat engine the brand behind oros AI s impressive capability our AI power overclocking offers impressive result alongside traditional methods by using neural networks instead of R based method we are seeing speed boost of over 20% next AI driven PCP design PCP design is a critical for dd5 performance and a gigabyte z90 leverage AI models to optimize our Modo PCB we have Implement three key AI driven optimizations first there is AI Trace optimization it finds the best signal pass from CPU to memory High PCP are compress made of copper foil epoxy racing and the fiberglass wave we found that the fiberglass wave direction can affect memory signal traces surprisingly the shortest PA is not always the best for the signal quality to optimize this we need to align trouting with the wave direction our AI steps in here using AI signal is simulation we have find two hundreds of chest pads from CPU to memory the result Ultima overall signal quality we have essentially create the perfect data Highway between your CPU and the memory second we have ai layer optimization this is a fasinating is Select ideal material for each PCB layers our gigabyte z90 mod has up to 12 PCB layers each with a unique purpose we cannot use the same material for each layers each needs to be tailored for each speciic specific job whether this is power deity or signal transmission using Advanced signal simulation we have determined the optimum material combination for each layer we have essentially customized every layer for each specific task this approach create mall where every layers is perfectly optimized working together to deliver Superior overall signal performance when it comes to critical memory signal routing layer we have gone the exra mile we have incorp ated low L PCB a technology typically reserved for the highend AI server Hardware this material boost extremely low dissipation factor or DF for short it translate to minimal energy loss and the exceptional signal quality by using sub gr material in the consumer modle we have pushing the boundary of qu passort in the memory performance lastly lay is AI VR optimization this ensure stable signal transmission between layers V stands for vertical interconnect SS is how we move signal between PCB layers when things get crowded traditionally this V has a fixed size but our AI simulation reveal something interesting different signal need different size passway on a gab z90 we have customized V diameter for high speed low speed and power delivery chases the results are impressive this led to a 28% reduction in the signal reflection loss this a technology together to achieve timal signal quality in memory loting layer composition and the course layer communication simply put we are using AI to make our modelable design smarter and more powerful now let's shift the gears from Hardware to phunware we have integrated AI into our bios development tool which what we call hypertune bios as many of you know MRC or memory reference code is crucial for CPU memory signal strength traditionally MC tuning was manual process heavily relying on R&D expertise but with the gigabyte z90 we have revolutionized this we have employed AI to determine optimal MRC parameter combinations for individual multiple configurations then our R&D team find us this AI generator settings ensuring we are squee out every last bit of per this AI human collaboration achieve truly optimal MC settings for each unique how setup pushing performance to a new height by combining all our D5 bionic cor Technologies software hardware and the phunware we have hit a new dd5 Milestone we have achieved outstanding 9500 Mt per second on the folding z90 mod bols we use the z90 ORS Master pair with v latest C memory this configuration now only hit 9500 per second but also pass m test without errors next up kaby z90 some more innovation this is where things get interesting for the new corra CPU Intel shifted the main power input to the top of the CPU how do we te this we have Implement a both wide heat pipe design on the Gaby z80 this balance temperatures across heat syn on both sides we have also tricked the PW funware to reallocate higher currents from the CPU left side preventing EXA he build up above it for you each over cers the z90 oros master and oros pro I come with bonus a dedicated active DDR fan when you are pushing ddr5 Beyond 900000 Mt per second temperature Spike our DDR wind bre can SL memory module TS by up to 10° maintaining stability at the L insane clock speed next are the highly ACC easy DIY friendly features first up Wi-Fi easy plug forgot the hle of twisting SMA connectors now you can snap in your Wi-Fi antenna in just one second every Gab zn9 B come with the PCI easy ledge Plus for storage we have got M2 easy ledge plus and the M2 easy ledge click lastly we have a sensor panel link it's a internal HDMI or type c connector right on the modable perfect adding a small display inside your transparent case to Showcase your RIS desk let's move on to Rapid connectivity every gab z94 come with The Cutting Edge Wi-Fi 7 we are talking about the fatest Wi-Fi speed available period L GCC software includes a new Wi-Fi Compass feature this tool displays real time signal gain value between your antenna and the router asides from Hardware our bios also have some standard features this time remember gigabyte per drive from 700 series we have involved into AI per drive it now includes Intel default settings and generate constant preset profiles based on your system actual summer performance this means Ultima setting tailor to your specific setup we have also added a recommend me profile function it use a unique loading tool right in the Bios to intelligent evaluate and suggest the best preset profile for your system it's like having the expert looking over your shoulder you can now switch between different bio thing you can choose between ORS oros Sprite and our new gr scale mode speaking of a great scale is not just about looks this clean distraction free interface is a perfect for you when you need to focus on serious bios tuning it's all about enhance your tring experience next durability durability is ensured by our Ultra durable technology take our PCI UD slot X it's not your average PCI slot with integrated single pieces design it can handle up to 58 kog of load that's crucial for Ro's massive gpus you are seeing this days now let's look at the ddfi slot on a gigabyte ZM 90 we have reinforced this with stainless steel we have designed this to extend over 5,000 insertion and the removals for you Inus who love to swap and test different memory kits this is a game changer now that we are covered the Innovations across z90 lineup let's zoom in on our top tier models the z90 oros a top is the flagship model of gigabyte z90 series it combines all the highend features of oros and air top series including support for dual graph cast dual 10g L and the Dual sandable fire it also supports the air typ utility which can be used for MTI car A5 tuning the D90 oros Master AI tuab also features highend capabilities of oros and air tuab series including support for dual graphic C dual 10g lamp and dual S 4 with 40 GB per second speed he also supports dd5 speed up to 9500 and includes all the easy DIY friendly features addition additionally is supposed the air typ utility which can be used for multi car air F tuning to R up our gab z99 we have craft a range that spend from Cutting Edge Flagship model to start this mainstream options whatever you need we have got to covero the gigaby z90 is our most advanced modable yet with D5 bionic Cora we have P AI at this C dd5 speed up to 9500 mt/ second the latest Wi-Fi 7 with directional Conners S 5 hitting 80 gb per second on our top models but it's not just about low power we have made it easier than ever to build and maintain your rig our easy DIY friendly features ensure the smoothest building experience and with ultra durable components your system is buil last we are entering a new AI era and the gigabyte z90 is your gateway all these amazing models are just halfway our journey through the latest generation of gigabyte marable next let's hear from angrew about our new MD maros [Music] now let's Shi gears and talk about our x870 series motherboards for amd's new ryzen 5 9000 Series CPUs MD's latest processors are Making Waves delivering a stagger in 26% performance boost over the 7,000 series but great CPUs and great model boards and that's where our x870 series comes in many of the groundbreaking features us just heard about in the za9 series we've brought them into the MD platform too let me round you through the highlights first up oros asnat with an AI engine inside that can boost the system with ease we've also carried over our popular easy di friendly features pcie m.2 easy latch
plus and Wi-Fi easy plug they are all here for better PC installation experience and yes we've include that sensor panel link with an internal HDMI connector on board making it convenient to inst a small display case to Showcase personal style in the Bios you'll find out our multi them option with great skill for a refreshing user experience we've added AI fan control for non Windows users and Q flash auto scan to save your time finding the suitable bios files durability we've got you covered with our memory UD slot and PCI UD slot a protecting your cherish memory and graphic card without tearing the motherboard itself and of course a 100% coverage support for Wi-Fi 7 our Wi-Fi Compass part of the gigabyte control center helps you to find the optimal signal Direction in real time now that we've covered the key features let's dive into our x870 Echo and xa7 motherboard lineup we provide wealthy spec models for every single segment for our reflection model we have the X a70 Echo ORS extreme AI top with Superior specs and accessories for the top rank SE we have x870 echo or smas which is designed for heavy overclocking and intense loading for the middle high rank segment we have x870 Echo orus Pro ice and x870 Echo orus Pro ready for any performance challenge with style for the middle rank segment we have x870 Echo and x870 orus Elite Wii 7 which contain competitive spec that suits for any Combat X a70 Echo ERS extreme AI top is the flagship model of gigabyte x870 series motherboard as its name shows it not only is the best choice of extreme overclocking and performance but also capable of using gigabytes unique AI type utility this tool opens a new world of possibilities allowing you to build and find your local AI Solutions right on your desktop let's talk about power we're looking at 22 faces of 110 M digital power and absolute overkilled power spec support cd5 speed up to a blister in A600 mt/ second with an exclusive dgr Wing blade as accessory to ensure cooling solution able to do multiple graphic card extensions weapon with Wi-Fi 7 dual 10gbps land dual USB 4 type c 4 GB per second ensuring M circumstances of R speed transaction this is speed this is lightning this is x870 Echo orus extreme AI top combining all the exclusive software hardware and firmware Technologies we have achieved an astounding dgr 5 speed of A600 mt/ second on MD platform to achieve A600 Mt per second such high memory speed our revolutionary ddr5 technology plays an important role which includes shed and isolated memory routing impendance optimized topology and eight layer Ser grade PCB I would like to talk more about eight layer server grid signal low loss PCB with ultra low disp factor to 0.77 and 2 O cover PCB internal test results confirm the design can reduce signal Interruption improve signal stability during highspeed signal transmission not only that the PCB material recipe can also reduce electromagnetic interference improving the anti- interference ability of the circuit orus x870 Echo mod boards share 100% coverage support of dual usb4 type-c with DP old mode for those of you who are seeking CPU performance optimization and test transition efficiency gigabyte provides active OC tuner in vs this allows the system to detect and Shuffle from MD PBO for single threat focus and manual OC For multi- Thread Focus automatically moreover orus X a70 motherboard is built for ryzen x3d from a hardware and software perspective ryzen x3d CPU known for its high density of 3D catch and unparalleled gaming decoding ability requires intense durability of motherboard Hardware to fully unleash its performance we equip all x870 GB motherboards with high current digital power faces up to 22 faces per board so users no longer need to worry about power endurance anymore on software wise gigabyte x870 model War plus ryzen x3d CPU is ready for Extreme Gaming with gigabytes AI engine software ORS AI snatch to boost the memory speed we can see up to 4% more enhancement during real world gaming benchmarks if you seek to find best gaming partner for the ryzen x3d CPU gigabyte xa7 motherboard is the only choice now back to Eddie thank to Angel and to every presenter today it's been an incredible event full of announcement and excitement today we are further expanding the scope of AI top they can find you not only LM models but also L mm models it can support cluster to fer improve Computing performance all a top now help create data set and the validate your model after F tuning KOB AI top is a fantastic solution for local F tuning at the same time we continuously launch a series of new AI typ Highway including new mod b aut durable SSD c car and more aop can provide a variety of combination it can be used in different scenarios with reable prices we are proud that AI top is an important driving force for further deploy the AI applications as a leader in high mbls G has better improveed the compatibility stability and overclocking performance of mos upon the design the data transfer rate of DT 5 can easily exceed 9500 Mt second and we keep updating the record 2024 has been a wonderful and a meaningful year for gigaby we are proud to have you witness all the Innovation with us thank you for joining us have a great day
2024-10-12