Jump to content
Main menu
Main menu
move to sidebar
hide
Navigation
Main page
Recent changes
Random page
freem
Search
Search
Appearance
Create account
Log in
Personal tools
Create account
Log in
Pages for logged out editors
learn more
Contributions
Talk
Editing
Openai/6907fc4f-3f34-8000-8487-2b4781f86078
(section)
Add languages
Page
Discussion
English
Read
Edit
Edit source
View history
Tools
Tools
move to sidebar
hide
Actions
Read
Edit
Edit source
View history
General
What links here
Related changes
Special pages
Page information
Appearance
move to sidebar
hide
Warning:
You are not logged in. Your IP address will be publicly visible if you make any edits. If you
log in
or
create an account
, your edits will be attributed to your username, along with other benefits.
Anti-spam check. Do
not
fill this in!
== gpt-5-thinking == === User: Radeon AI PRO R9700 と nvidia rtx 5090 を較べて === Radeon AI PRO R9700 と nvidia rtx 5090 を較べて === Assistant: !https://www\.amd\.com/content/dam/amd/en/images/products/graphics/3402302\-amd\-radeon\-ai\-pro\-r9700\.jpghttps://tse… === !https://www\.amd\.com/content/dam/amd/en/images/products/graphics/3402302\-amd\-radeon\-ai\-pro\-r9700\.jpg<ref>{{cite web|title=https://www\.amd\.com/content/dam/amd/en/images/products/graphics/3402302\-amd\-radeon\-ai\-pro\-r9700\.jpg|url=https://tse1.mm.bing.net/th/id/OIP.s_H3FCvPYczPRYwH4jNR8QHaEK?w=474&h=474&c=7&p=0|publisher=tse1.mm.bing.net|access-date=2025-11-13}}</ref> ざっくり言うと、 * 生のAI性能・エコシステムはRTX 5090が上(3,352 AI TOPS、FP4/FP6/FP8対応、CUDA/TensorRT-LLMが成熟)。NVIDIA Docs<ref>{{cite web|title=NVIDIA Docs|url=https://pc.watch.impress.co.jp/docs/topic/feature/1656874.html|publisher=NVIDIA Docs|access-date=2025-11-13}}</ref> * 消費電力と価格のバランスはRadeon AI PRO R9700が良い(32GB GDDR6で約$1,299クラス、TDP 300W)。GameGPU<ref>{{cite web|title=GameGPU|url=https://en.gamegpu.com/iron/The-first-Radeon-AI-Pro-R9700-benchmarks-are-twice-as-cheap-and-cooler-than-the-RTX-6000.|publisher=en.gamegpu.com|access-date=2025-11-13}}</ref> * 帯域は5090が3倍近く広い(5090: 512bit GDDR7で最大1.7〜1.8TB/s、R9700: 256bit GDDR6で640GB/s)。NVIDIA<ref>{{cite web|title=NVIDIA|url=https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/|publisher=nvidia.com|access-date=2025-11-13}}</ref> * 用途イメージがちょっと違う:R9700は「ローカルAI・LLMをワークステーションでたくさん回す・複数枚で積む」寄り、5090は「最強ゲーミング+生成AIを1枚で全部やる」寄りです。AMD<ref>{{cite web|title=AMD|url=https://www.amd.com/en/products/graphics/workstations/radeon-ai-pro/ai-9000-series/amd-radeon-ai-pro-r9700.html|publisher=amd.com|access-date=2025-11-13}}</ref> ===== 主なスペック比較 ===== | 項目 | AMD Radeon AI PRO R9700 | NVIDIA GeForce RTX 5090 | | -------- | -------------------------------------------------------- | --------------------------------------------- | | アーキテクチャ | RDNA 4, 64CU, 128 AIアクセラレータ | Blackwell GB202, 21,760 CUDA, 680 Tensor Core | | メモリ | 32GB GDDR6, 256bit, 20Gbps, 640GB/s, 64MB Infinity Cache | 32GB GDDR7, 512bit, 1.7〜1.8TB/s | | 公称演算 | FP32 47.8TFLOPS/FP16 95.7TFLOPS/INT4 1,531TOPS | FP32 104.8TFLOPS/AI TOPS 3,352 | | 消費電力 | 300Wクラス、2スロット・ブロワ | 575Wクラス、大型3スロット | | インターフェース | PCIe 5.0, DP 2.1a ×4 | PCIe 5.0, DP 2.1b ×3 + HDMI 2.1b | | 価格目安 | $1,250〜1,300程度で登場と報道・日本でも搭載機が50万円弱から GameGPU +1 | $1,999(Founders基準) NVIDIA +1 | PC Watch<ref>{{cite web|title=PC Watch|url=https://www.amd.com/en/products/graphics/workstations/radeon-ai-pro/ai-9000-series/amd-radeon-ai-pro-r9700.html|publisher=PC Watch|access-date=2025-11-13}}</ref> ===== AI・LLM用途で見ると ===== # ピークAIスループット - RTX 5090の3,352 AI TOPSはR9700のINT4 1,531 TOPSのざっくり2倍強です。これはBlackwell世代TensorコアがFP4や低精度を前提にしているからで、Stable Diffusion系やTensorRT-LLMの最新ビルドを素で走らせると5090が有利になります。PC Watch<ref>{{cite web|title=PC Watch|url=https://pc.watch.impress.co.jp/docs/topic/feature/1656874.html|publisher=PC Watch|access-date=2025-11-13}}</ref> - R9700もINT4 1,531 TOPSでこのクラスとしてはかなり強く、300Wでここまで出るのは効率が良いです。sapphiretech.com<ref>{{cite web|title=sapphiretech.com|url=https://www.sapphiretech.com/ja-jp/commercial/radeon-ai-pro-r9700|publisher=sapphiretech.com|access-date=2025-11-13}}</ref> # ソフトウェア栄養価 - 5090はCUDA 13世代でTensorRT-LLMや各種OSSがどんどん追随しているので「新しいモデルを持ってきたらとりあえず動く」側。WSL2でも事例が出ています。Reddit<ref>{{cite web|title=Reddit|url=https://docs.nvidia.com/deeplearning/tensorrt/latest/getting-started/support-matrix.html|publisher=docs.nvidia.com|access-date=2025-11-13}}</ref> - R9700はROCm 7.0.xで公式サポートが入り、vLLMのテストでもRTX 6000 Adaと良い勝負をしていて、2枚構成だと追い抜くケースもありますが、Linux前提がまだ強いです。Windowsは2025/9時点でPyTorch on AMDが「公開プレビュー」で、実用はできますがNVIDIAほどの“そのまま感”はありません。gpuopen.com<ref>{{cite web|title=gpuopen.com|url=https://www.phoronix.com/review/amd-radeon-ai-pro-r9700|publisher=phoronix.com|access-date=2025-11-13}}</ref> # メモリの質 - 両方32GBなので「量」では並びますが、帯域は5090が約1.8TB/sでR9700の約640GB/sの3倍近いです。RLHFやMoEでアクティベーションをたくさん回すと効いてきます。NVIDIA<ref>{{cite web|title=NVIDIA|url=https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/|publisher=nvidia.com|access-date=2025-11-13}}</ref> # 複数枚運用 - R9700は300W・ブロワ・2スロットなので、ワークステーションに4本指すような使い方をかなり現実的にできます(AMD自身もマルチGPU推し)。asrock.com<ref>{{cite web|title=asrock.com|url=https://www.asrock.com/Graphics-Card/AMD/Radeon%20AI%20PRO%20R9700%20Creator%2032GB/index.jp.asp|publisher=asrock.com|access-date=2025-11-13}}</ref> - RTX 5090を4本は電源も冷却もかなり大掛かりになります。1〜2枚で最高性能を引き出すカードと考えた方が自然です。The Verge<ref>{{cite web|title=The Verge|url=https://www.theverge.com/2025/1/6/24337396/nvidia-rtx-5080-5090-5070-ti-5070-price-release-date|publisher=The Verge|access-date=2025-11-13}}</ref> ===== クリエイティブ/ゲームもやるなら? ===== * RTX 5090はRTコア318TFLOPSで完全にハイエンドゲーミング兼用ですし、DLSS 4でのAIフレーム生成もこのカードの得意技です。PC Watch<ref>{{cite web|title=PC Watch|url=https://pc.watch.impress.co.jp/docs/topic/feature/1656874.html|publisher=PC Watch|access-date=2025-11-13}}</ref> * R9700は名前のとおり「AI PRO」なので、ゲームでの最適化やDLSSに相当する機能が欲しいなら5090の方が楽です。Tom's Hardware<ref>{{cite web|title=Tom's Hardware|url=https://www.tomshardware.com/pc-components/gpus/amd-launches-radeon-ai-pro-r9700-to-challenge-nvidias-ai-market-dominance|publisher=Tom's Hardware|access-date=2025-11-13}}</ref> ===== 結論を一文でまとめると ===== * 「1枚で最先端AIもゲームも、しかもWindowsで、とにかく最速が欲しい」なら → RTX 5090 * 「32GBを安く・静かに・何枚も積んで、LinuxでvLLMやROCmを回す」なら → Radeon AI PRO R9700 どちらに寄せるかはここで決まると思います。 === User: 日本での実売価格を調べて === 日本での実売価格を調べて === Assistant: いま(2025年11月3日・日本)だとこうなってます👇 === いま(2025年11月3日・日本)だとこうなってます👇 ===== 1. Radeon AI PRO R9700 ===== * 単体カードの店頭想定価格:246,800円前後が各社のアナウンスと価格.com掲載の最安ラインです。価格.com<ref>{{cite web|title=価格.com|url=https://search.kakaku.com/AMD%20Radeon%E3%83%BB%20AI%20PRO%20R9700/|publisher=search.kakaku.com|access-date=2025-11-13}}</ref> * 実売レンジ:だいたい 24.7万〜26万円台前半で出てくるはず(発売が10/31で出始めなので大きな値引きはまだ出ていない)。問屋想定どおりに並んでいて、在庫が切れると上振れする感じです。ASCII<ref>{{cite web|title=ASCII|url=https://ascii.jp/elem/000/004/338/4338887/|publisher=ascii.jp|access-date=2025-11-13}}</ref> * 参考:同GPUを2枚積んだツクモのBTOが84万9,800円なので、1枚あたり25万円弱で積んでる計算になり、上の数字とだいたい合います。〖TSUKUMO(ツクモ)〗プレスリリース -<ref>{{cite web|title=〖TSUKUMO(ツクモ)〗プレスリリース -|url=https://news.tsukumo.co.jp/20250910_mgpu/|publisher=〖TSUKUMO(ツクモ)〗プレスリリース -|access-date=2025-11-13}}</ref> →つまりR9700を「バルクに近い感覚で1枚だけ買う」なら25万円前後を見ておけばいいです。 ===== 2. GeForce RTX 5090 ===== * 国内公式の発表価格(発売時):393,800円〜。でもこれは“建前価格”で、今はほとんど見かけません。PC Watch<ref>{{cite web|title=PC Watch|url=https://pc.watch.impress.co.jp/docs/news/1652618.html|publisher=PC Watch|access-date=2025-11-13}}</ref> * 現在の最安実売:在庫があるところだと42万2,000円前後が最安、という報告が10月末〜11月頭で出ています。ギャズログ | GAZ:Log<ref>{{cite web|title=ギャズログ | GAZ:Log|url=https://gazlog.com/entry/geforce-rtx5090-price-rise-oct/|publisher=ギャズログ | GAZ:Log|access-date=2025-11-13}}</ref> * 価格.comで並んでいるボード:安い方で45万5,800円〜46万円台、水冷やOCの上位だと50万円台半ばまで上がります。価格.com<ref>{{cite web|title=価格.com|url=https://kakaku.com/pc/videocard/itemlist.aspx?pdf_Spec103=500|publisher=kakaku.com|access-date=2025-11-13}}</ref> * Amazon等での在庫薄モデル:53万〜54万円クラスまで跳ねているものもあり、ここは完全に品薄プレミアです。価格.com<ref>{{cite web|title=価格.com|url=https://search.kakaku.com/GeForce%20RTX%205090/|publisher=search.kakaku.com|access-date=2025-11-13}}</ref> →要するに、今日本で「普通に買おう」とすると 42万〜50万円台が現実的レンジで、たまたま在庫を抱えてる店に当たれば40万円台前半、人気のOCモデルだと50万円オーバー、という相場です。 ===== まとめると ===== * Radeon AI PRO R9700:おおむね 25万円前後 * RTX 5090:おおむね 42万〜50万円台(公称39万3,800円はほぼ見かけない) この差がそのまま「1枚あたりの始めやすさ」の差になってるので、ワークステーションに複数枚挿す前提ならR9700の方が日本では組みやすいです。
Summary:
Please note that all contributions to freem are considered to be released under the Creative Commons Attribution-ShareAlike 4.0 (see
Freem:Copyrights
for details). If you do not want your writing to be edited mercilessly and redistributed at will, then do not submit it here.
You are also promising us that you wrote this yourself, or copied it from a public domain or similar free resource.
Do not submit copyrighted work without permission!
Cancel
Editing help
(opens in new window)