<s>
AMD	B-Operating_System
Instinct	I-Operating_System
is	O
AMD	O
's	O
brand	O
of	O
professional	O
GPUs	B-Architecture
.	O
</s>
<s>
It	O
replaced	O
AMD	O
's	O
FirePro	B-Application
S	O
brand	O
in	O
2016	O
.	O
</s>
<s>
Compared	O
to	O
the	O
Radeon	B-Device
brand	O
of	O
mainstream	O
consumer/gamer	O
products	O
,	O
the	O
Instinct	O
product	O
line	O
is	O
intended	O
to	O
accelerate	O
deep	O
learning	O
,	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
,	O
and	O
high-performance	O
computing/GPGPU	O
applications	O
.	O
</s>
<s>
The	O
Radeon	B-Operating_System
Instinct	I-Operating_System
product	O
line	O
directly	O
competes	O
with	O
Nvidia	O
's	O
Ampere	B-General_Concept
and	O
Intel	B-General_Concept
Xeon	I-General_Concept
Phi	I-General_Concept
and	O
incoming	O
Intel	B-General_Concept
Xe	I-General_Concept
lines	O
of	O
machine	O
learning	O
and	O
GPGPU	B-Architecture
cards	O
.	O
</s>
<s>
Before	O
MI100	O
introduction	O
in	O
November	O
2020	O
,	O
the	O
Instinct	O
family	O
was	O
known	O
as	O
AMD	B-Device
Radeon	I-Device
Instinct	O
,	O
AMD	O
dropped	O
the	O
Radeon	B-Device
brand	O
from	O
its	O
name	O
.	O
</s>
<s>
Supercomputers	B-Architecture
based	O
on	O
(	O
AMD	O
CPUs	O
and	O
)	O
AMD	B-Operating_System
Instinct	I-Operating_System
GPUs	B-Architecture
now	O
take	O
the	O
lead	O
on	O
the	O
Green500	B-Application
supercomputer	B-Architecture
list	O
with	O
over	O
50%	O
lead	O
over	O
any	O
other	O
,	O
and	O
top	O
the	O
first	O
4	O
spots	O
,	O
including	O
the	O
second	O
,	O
which	O
is	O
the	O
current	O
fastest	O
in	O
the	O
world	O
on	O
the	O
TOP500	B-Operating_System
list	O
,	O
Frontier	B-Device
.	O
</s>
<s>
The	O
three	O
initial	O
Radeon	B-Operating_System
Instinct	I-Operating_System
products	O
were	O
announced	O
on	O
December	O
12	O
,	O
2016	O
,	O
and	O
released	O
on	O
June	O
20	O
,	O
2017	O
,	O
with	O
each	O
based	O
on	O
a	O
different	O
architecture	O
.	O
</s>
<s>
The	O
MI6	O
is	O
a	O
passively	O
cooled	O
,	O
Polaris	O
10	O
based	O
card	O
with	O
16	O
GB	O
of	O
GDDR5	O
memory	O
and	O
with	O
a	O
<	O
150	O
W	O
TDP	B-General_Concept
.	O
</s>
<s>
At	O
5.7	O
TFLOPS	O
(	O
FP16	O
and	O
FP32	O
)	O
,	O
the	O
MI6	O
is	O
expected	O
to	O
be	O
used	O
primarily	O
for	O
inference	O
,	O
rather	O
than	O
neural	B-Architecture
network	I-Architecture
training	O
.	O
</s>
<s>
The	O
MI8	O
is	O
a	O
Fiji	B-Device
based	O
card	O
,	O
analogous	O
to	O
the	O
R9	B-Device
Nano	I-Device
,	O
and	O
expected	O
to	O
have	O
a	O
<	O
175W	O
TDP	B-General_Concept
.	O
</s>
<s>
The	O
MI25	O
is	O
a	O
Vega	B-Device
based	O
card	O
,	O
utilizing	O
HBM2	O
memory	O
.	O
</s>
<s>
The	O
MI25	O
is	O
rated	O
at	O
<3	O
00W	O
TDP	B-General_Concept
with	O
passive	O
cooling	O
.	O
</s>
<s>
Following	O
software	O
is	O
,	O
as	O
of	O
2022	O
,	O
regrouped	O
under	O
the	O
Radeon	B-Operating_System
Open	I-Operating_System
Compute	I-Operating_System
meta-project	O
.	O
</s>
<s>
The	O
MI6	O
,	O
MI8	O
,	O
and	O
MI25	O
products	O
all	O
support	O
AMD	O
's	O
MxGPU	O
virtualization	B-General_Concept
technology	O
,	O
enabling	O
sharing	O
of	O
GPU	B-Architecture
resources	O
across	O
multiple	O
users	O
.	O
</s>
<s>
MIOpen	O
is	O
AMD	O
's	O
deep	O
learning	O
library	O
to	O
enable	O
GPU	B-Architecture
acceleration	O
of	O
deep	O
learning	O
.	O
</s>
<s>
Much	O
of	O
this	O
extends	O
the	O
GPUOpen	B-Application
's	O
Boltzmann	O
Initiative	O
software	O
.	O
</s>
<s>
This	O
is	O
intended	O
to	O
compete	O
with	O
the	O
deep	O
learning	O
portions	O
of	O
Nvidia	O
's	O
CUDA	B-Architecture
library	O
.	O
</s>
<s>
It	O
supports	O
the	O
deep	O
learning	O
frameworks	O
:	O
Theano	B-Algorithm
,	O
Caffe	B-Algorithm
,	O
TensorFlow	B-Language
,	O
MXNet	B-Algorithm
,	O
Microsoft	B-Algorithm
Cognitive	I-Algorithm
Toolkit	I-Algorithm
,	O
Torch	B-Algorithm
,	O
and	O
Chainer	B-Algorithm
.	O
</s>
<s>
Programming	O
is	O
supported	O
in	O
OpenCL	B-Application
and	O
Python	B-Language
,	O
in	O
addition	O
to	O
supporting	O
the	O
compilation	O
of	O
CUDA	B-Architecture
through	O
AMD	O
's	O
Heterogeneous-compute	O
Interface	O
for	O
Portability	O
and	O
Heterogeneous	O
Compute	O
Compiler	O
.	O
</s>
