88
(当時間接編書 13号 18号
今日銀1日 日曜日
2資料及びコメントの提供方法上記1(2)の物
品に関する一般的な参考資料及び同(6)の要求要
件に関するコメント並びに本システムの構成や
性能等の見地から有用と思われる資料等の提供
を招請する。
(1)資料等の提供期限令和8年9月1日17時
00分 (郵送の場合は必着のこと。)
(2)提供先319-1195茨城県那珂郡東海村
白方2-4日本原子力研究開発機構原子
カ科学研究所システム計算科学センター高
性能計算技術利用推進室電話029-282-
6764
3説明書の交付本公表に基づき応募する供給
者に対して導入説明書を交付する。
(1)交付期間令和8年7月24日から令和8年
8月31日まで。
(2)交付場所上記2(2)に同じ。
4説明会の開催本公表に基づく導入説明会を
開催する。
(1)開催日時令和8年7月30日14時00分
(2)開催場所オンライン
5その他この導入計画の詳細は導入説明書に
よる。なお、本公表内容は予定であり、変更す
ることがあり得る。
6 Summary
(1)Classification of the products to be pro-
cured:14
(2) Nature and quantity of the products to be
rent : GPU computing system 1 Set
(3) Type of the procurement : Rent
(4) Basic requirements of the procurement:
A Outline of Basic Requirements
a The GPU computing system should
be commposed of a GPU processing
units, a job allocation subsystem, a
login management subsystem, a paral-
lel file system, a 10Gigabit Ethernet
switch, a web service system, and vari-
ous management servers, etc.
b The GPU computing system should
be one of most power-and space-
efficient systems.
CEach component of the GPU comput-
ing system should have high reliability.
Proper countermeasures should be tak-
en a gainst even a small component
trouble so that the entire system would
not crash and the influence would be
minimized
d The computing memory
bandwi dth, the interconnect-network
bandwidth, etc. of the GPU computing
system should have performance high.
e Fortran, C/C++, and Python co-
mpilers for effective parallel process-
ing and MPI and OpenMP library for
message passing should be provided on
the GPU computing system
f The GPU computing system should
have appropriate application develop-
ment environments (parallelizing co-
mpilers, parallelization tuning tools,
etc.) to efficiently use hundreds of thou-
sands of cores and experienced opera-
tion tools to efficiently and stably man-
age all of the cores.
g Support for smooth porting of the
user codes and data from the present
computer systems (HPE
SGI8600) to the new system should be
provided.
h The immediate support for a mainte-
nance and a trouble shooting should be
provided for the GPU computing sys-
tem. When JAEA judges that a system
trouble is serious, an urgent trouble
shooting team should be established to
settle the trouble quickly.
B Outline of Total Evaluation The suc-
cessful bid depends on total Evaluation of
the price, the performance, the functions,
etc. The introduction manual shows the
outline of the total evaluation.
(5) Time limit for the submission of the re-
quested materials:17:00,1 September,
2026.
(6)Contact point for the notice:Admiministra-
tion Section, Center for Computational Sci-
ence and e-Systems, Japan Atomic Energy
Agency, 2—4 Shirakata, Tokai-mura, Na-
ka-gun, Ibaraki 319-1195 Japan. Tel 029-
282-6764