NVIDIA GPU Microarchitecture and Optimization (Paperback)
Langue : anglais
Edité par Independently Published, 2026
- Livre broché
- Neuf
Vendeur : Grand Eagle Retail, Bensenville, IL, Etats-UnisGrand Eagle Retail
Vendeur AbeBooks depuis 12 octobre 2005
Etat: Neuf
EUR 28,45
Quantité disponible : 1 disponible
Ajouter au panierA propos de cet article
Paperback. Have you ever pushed a CUDA kernel into production, watched the profiler, and quietly wondered why it's only using a fraction of the GPU you paid for?You're not alone. Most engineers learn just enough parallel programming to get something working, then hit a wall the moment they need it to actually perform. The gap between "it runs" and "it's fast" isn't a syntax problem. It's a hardware understanding problem. And that's exactly what this book is built to close.This isn't another surface-level tutorial that shows you how to launch a kernel and calls it a day. This is a complete, ground-up education in how modern GPU hardware actually thinks, so that every optimization decision you make afterward is grounded in real understanding instead of guesswork and copy-pasted tricks from forum posts.So let me ask you a few honest questions: Do you actually know what a warp is doing when your threads diverge, or are you just hoping the compiler handles it? Can you look at a kernel's memory access pattern and predict, before you even run it, whether it'll hit the bandwidth ceiling? Do you know the difference between a kernel that's compute bound and one that's memory bound, and why that distinction should completely change how you optimize it? If any of those questions gave you pause, this book was written for you.Inside, you'll build a real mental model of the hardware itself: streaming multiprocessors, the memory hierarchy from registers down to global memory, and the scheduling logic that determines whether your GPU is actually busy or just spinning its wheels. From there, the book walks you through the full programming model step by step, thread and block organization, warp-level execution, and the exact indexing patterns that separate a kernel that merely works from one that's genuinely efficient.You'll spend real time inside the memory system, because that's where most performance is won or lost. Coalesced access, shared memory tiling, bank conflicts, register pressure, and the caching behavior that quietly shapes every kernel you write. You'll learn to read a profiler's output the way a mechanic reads an engine, spotting exactly where cycles are being wasted and why.Then the book moves into territory a lot of guides skip entirely: concurrency and streams, multi-GPU scaling, tensor core programming, mixed precision training, and quantization for inference. If you're working anywhere near machine learning infrastructure, scientific computing, or large-scale data processing, these chapters alone are worth the price of admission.And because knowing a technique isn't the same as knowing when to use it, the book closes with hands-on capstone projects, a custom convolution kernel, a scientific simulation, a full inference pipeline, and a multi-GPU scaling exercise, each one built the same way real optimization work actually happens: start naive, profile honestly, fix what the data tells you to fix, and measure again.Here's what you won't find in this book: recycled documentation, filler chapters, or vague advice to "just try different block sizes." Every technique is explained with the reasoning behind it, so that when the next generation of hardware ships, you're not starting over. You'll already know how to think about it.Who is this actually for? Software engineers moving into performance-critical roles. Machine learning practitioners who want to understand what's happening beneath their framework of choice. Computer science students who want an edge going into systems or infrastructure roles. And working GPU programmers who know the basics but have never had the hardware explained to them properly, the way this book explains it.If you're ready to stop guessing and start actually understanding the hardware you're programming for, this is where that st Shipping may be from multiple locations in the US or from the UK, depending on stock availability.…
N° de réf. du vendeur 9798175045759
- Titre
- NVIDIA GPU Microarchitecture and Optimization (Paperback)
- Auteur
- Phyllis H. Frampton
- Éditeur
- Independently Published
- Année de publication
- 2026
- État de l'article
- new
- Reliure
- Paperback
- Langue
- anglais
- ISBN à 13 chiffres
- 9798175045759
- Série
- Livre 14 sur 17: The Modern Engineer's Library
Have you ever pushed a CUDA kernel into production, watched the profiler, and quietly wondered why it's only using a fraction of the GPU you paid for?
You're not alone. Most engineers learn just enough parallel programming to get something working, then hit a wall the moment they need it to actually perform. The gap between "it runs" and "it's fast" isn't a syntax problem. It's a hardware understanding problem. And that's exactly what this book is built to close.
This isn't another surface-level tutorial that shows you how to launch a kernel and calls it a day. This is a complete, ground-up education in how modern GPU hardware actually thinks, so that every optimization decision you make afterward is grounded in real understanding instead of guesswork and copy-pasted tricks from forum posts.
So let me ask you a few honest questions:
Do you actually know what a warp is doing when your threads diverge, or are you just hoping the compiler handles it? Can you look at a kernel's memory access pattern and predict, before you even run it, whether it'll hit the bandwidth ceiling? Do you know the difference between a kernel that's compute bound and one that's memory bound, and why that distinction should completely change how you optimize it? If any of those questions gave you pause, this book was written for you.
Inside, you'll build a real mental model of the hardware itself: streaming multiprocessors, the memory hierarchy from registers down to global memory, and the scheduling logic that determines whether your GPU is actually busy or just spinning its wheels. From there, the book walks you through the full programming model step by step, thread and block organization, warp-level execution, and the exact indexing patterns that separate a kernel that merely works from one that's genuinely efficient.
You'll spend real time inside the memory system, because that's where most performance is won or lost. Coalesced access, shared memory tiling, bank conflicts, register pressure, and the caching behavior that quietly shapes every kernel you write. You'll learn to read a profiler's output the way a mechanic reads an engine, spotting exactly where cycles are being wasted and why.
Then the book moves into territory a lot of guides skip entirely: concurrency and streams, multi-GPU scaling, tensor core programming, mixed precision training, and quantization for inference. If you're working anywhere near machine learning infrastructure, scientific computing, or large-scale data processing, these chapters alone are worth the price of admission.
And because knowing a technique isn't the same as knowing when to use it, the book closes with hands-on capstone projects, a custom convolution kernel, a scientific simulation, a full inference pipeline, and a multi-GPU scaling exercise, each one built the same way real optimization work actually happens: start naive, profile honestly, fix what the data tells you to fix, and measure again.
Here's what you won't find in this book: recycled documentation, filler chapters, or vague advice to "just try different block sizes." Every technique is explained with the reasoning behind it, so that when the next generation of hardware ships, you're not starting over. You'll already know how to think about it.
Who is this actually for? Software engineers moving into performance-critical roles. Machine learning practitioners who want to understand what's happening beneath their framework of choice. Computer science students who want an edge going into systems or infrastructure roles. And working GPU programmers who know the basics but have never had the hardware explained to them properly, the way this book explains it.
If you're ready to stop guessing and start actually understanding the hardware you're programming for, this is where that starts. Scroll up, and let's get into it.
« Synopsis » peut appartenir à une autre édition de cet ouvrage.
Grand Eagle Retail
Bensenville, IL, Etats-Unis
Vendeur AbeBooks depuis 12 octobre 2005
Frais d'expédition à l'intérieur de ce pays : Etats-Unis
| Article | 6 à 14 jours ouvrés | 6 à 16 jours ouvrés |
|---|---|---|
| Premier article | EUR 0,00 | EUR 0,00 |
Modes de paiement
Profil professionnel du vendeur
APOLLO ONLINE CORP.
605 Geddes Street
Wilmington, DE Etats-Unis 19805
Conditions de vente
We guarantee the condition of every book as it¿s described on the Abebooks web sites. If you¿ve changed
your mind about a book that you¿ve ordered, please use the Ask bookseller a question link to contact us
and we¿ll respond within 2 business days.
Books ship from California and Michigan.
Droit de rétractation
Si vous êtes un consommateur, vous pouvez exercer votre droit de rétractation sur le contrat conformément à ce qui suit. Le mot « consommateur » désigne toute personne physique agissant à des fins qui n'entrent pas dans le cadre de son activité commerciale, artisanale ou professionnelle.
Informations concernant le droit de rétractation
Droit statutaire de rétractation
Vous avez le droit d'exercer votre droit de rétractation sur ce contrat dans les 14 jours sans donner de raison.
Le délai de rétractation expirera au bout de 14 jours à compter du jour où vous-même, ou un tiers autre que le transporteur et désigné par vous, prendrez physiquement possession de la dernière marchandise, du dernier lot ou de la dernière pièce.
Pour exercer votre droit de rétractation, remplissez électroniquement et envoyez une déclaration claire sur notre site Web, sous « Vos achats » dans « Votre compte ». Nous vous communiquerons sans délai un accusé de réception de cette rétractation sur un support durable (par exemple, par e-mail).
Pour respecter le délai de rétractation, il vous suffit d'envoyer votre message concernant l'exercice de votre droit de rétractation avant l'expiration du délai de rétractation.
Effets de la rétractation
Si vous exercez votre droit de rétractation sur ce contrat, nous vous rembourserons tous les paiements que vous avez effectués, y compris les frais de livraison (à l'exception des frais supplémentaires résultant du choix d'un mode de livraison autre que le type de livraison standard le moins cher que nous proposons).
Nous pouvons déduire du remboursement la perte de valeur de toute marchandise livrée, si la perte est le résultat d'une manipulation inutile de votre part.
Nous effectuerons le remboursement dans les meilleurs délais, et au plus tard 14 jours après le jour où nous aurons été informés de votre décision d'exercer votre droit de rétractation sur ce contrat.
Nous effectuerons le remboursement en utilisant le même moyen de paiement que celui que vous avez utilisé pour la transaction initiale, sauf si vous en avez expressément convenu autrement ; en tout état de cause, aucuns frais ne vous seront facturés à la suite d'un tel remboursement.
Nous pouvons suspendre le remboursement jusqu'à ce que nous ayons reçu les marchandises ou que vous ayez fourni la preuve que vous avez renvoyé les marchandises, en fonction de la première éventualité.
Vous devez renvoyer les marchandises ou les remettre à Grand Eagle Retail, Bensenville, Illinois, U.S.A., sans retard injustifié et, en tout état de cause, au plus tard 14 jours à compter du jour où vous nous avez communiqué votre décision de rétractation du présent contrat. Le délai est respecté si vous renvoyez les marchandises avant l'expiration du délai de 14 jours. Vous devrez prendre en charge les frais directs du renvoi des marchandises. Vous n'êtes responsable que de toute diminution de valeur des marchandises résultant d'une manipulation autre que celle nécessaire pour établir la nature, les caractéristiques et le fonctionnement des marchandises.
Exceptions au droit de rétractation
Le droit de rétractation ne s'applique pas à ce qui suit :
- Distribution de journaux, de revues ou de magazines, à l'exception des contrats d'abonnement ; et
- Fourniture d'un contenu numérique qui n'est pas fourni sur un support matériel (par exemple, sur un CD ou un DVD) si vous avez accepté, lors de votre commande, que nous puissions commencer à le livrer et que vous ne puissiez pas exercer votre droit de rétractation une fois la livraison commencée.
Conditions d'expédition
Orders usually ship within 2 business days. All books within the US ship free of charge. Delivery is 4-14 business days anywhere in the United States.
Books ship from California and Michigan.
If your book order is heavy or oversized, we may contact you to let you know extra shipping is required.