HW News: GDDR6, HBM3, Zen CCX Architecture, PCIe Gen4
HW News: GDDR6, HBM3, Zen CCX Architecture, PCIe Gen4
2016-08-26
gddr5 X only just came out and Samsung's
already trying to one-up micron with GG
dr6 obviously is a numeric follower of
gddr5 if that wasn't obvious and then of
course SK Hynix is trying to start
pushing HP m3 at least in the research
and development side before HP MT is
really on the mass-market so those are
two of the main items for this news
video but we've got a couple more coming
up before getting to that this coverage
is brought to you by iva power and their
new element gaming PC with the full
tempered glass side window and LED is
it's basically an s3 40 that's been
modified by iBUYPOWER
if you're curious about what the base
case is so gddr5 X is brand new that is
from micron it's only found right now in
the GTX 1080 and a Titan X Pascale cards
and it operates at 10 gigabits per
second on those devices but micron has
previously demonstrated that the at
least the earlier versions of the
product were able to hit somewhere in
the 13 ish 12 to 13 gigabit per second
speed obviously that's not on the cards
yet but memory tends to have a range
like that like with gddr5 the kind of
high-end that we're seeing today is
about 8 gigabits per second from the
best GPUs on the market with the best
memory and then it'll scale down to
something closer to 6 so there is a
decent range there and that's true for
the new GDD are 6 memory as well which
Samsung details as having a range of
about 14 years per second at the low end
to about 16 at the high end so there's
room for growth and it's still
outperforming g5x today with the 14
gigabits per second of minimum gddr5 has
been in production used since 2008 and
in its present fully matured form it is
operating at that eight gigabit per
second number in the RX 480 and the GTX
10 series GPU is at least most of them
Samsung's increased speed is only one of
the advertised features right now
there's also reduced voltage requirement
for the speed so we're at one point
three five volts for the 14 to 16
gigabit per second speed range coupled
with lower voltages than even g5x by
using LP 4x Samson indicates a power
reduction upwards of 20% with post LP
for memory Tecna
and is looking toward 2018 for
production of GDR six so that does
actually give g5x some breathing room as
of now as for HB m SK Hynix is already
looking toward the world of HB m 3 HB m
2 in its current form is only really
existent on the Tesla accelerated card
accelerated card the P 100 using the GP
100 GPU that was the first Pascal chip
shown so that's the only place that HP
m2 really is known to exist in a form
that's being produced right now
HB m 2 obviously didn't hit the game
inside of Pascal we're not quite sure
when the next AMD series of cards will
come out but we do know that they will
feature HB m 2 so other than Vega what
we're looking at for HB M 3 is a doubled
throughput so it's theoretically on a
4096 bit interface with upwards of 2
terabyte per second throughput and
that's 512 gigabytes per second per
stack of HB m3 and we'll talk about this
tech more in the semi distant future
because it is a ways off there's also
news this week from Tom's Hardware who
spoke with some of the PC is IG folks to
talk about power draw and the power
allowance in the PCIe interface for
version 4.0 this primarily detailed a
minimum spec of 300 watt power transfer
through the slot but could be upwards of
500 watts it's not finalized yet and
without even talking out the bandwidth
promises which by the way I moved to
nearly 2 gigabytes per second per lane
the increase of power budget will mean
that the industry could begin a shift
away from PCI Express power cables and
the power would obviously still come
from the power supply but it would be
delivered through the pins in the PCIe
slots and the cards rather than through
an extra cable this same set up is what
allowed cards like the sound 50 Ti and
other low power cards to pull all of
their power through the PCIe slot rather
than requiring any PCIe power cables
from the power supply at all so that's
the future we're looking at and that
would work on the high end cards on
market today so it's pretty interesting
one and the PCIe gen 4 spec should be
ratified by end of year 2016 of course
they're currently on version point seven
point nine is to come and
point-o should be before the end of the
year one quick note that we don't yet
know the rollout plans for consumer
products so that's just ratifying the
spec it doesn't mean it'll be available
in platforms just yet and the also
detailed more of its Xen architecture
something we talked about last week when
the company camped out near IDF Intel's
conference and hosted its own mini
conference of sorts to talk about Xen
the summit ridge chips have primarily
been on display thus far showing an 8
core 16 thread demo with and these
implementation of SMT but we haven't
heard much about other processors just
yet and he is ditching modules from the
bulldozer series in favor of new CPU
complexes or CC X's each of these CC X
units host for CPU cores and each CC X
runs 512 kilobytes of l2 cache per core
as seen in this block diagram with l3
sliced into four pieces of eight
megabytes total for a low order address
and relieve cache and these says that
each core can communicate with all the
cache on the CC X and promises the same
latency for all accesses definitely big
news and given this architecture it
looks like the smallest chips in terms
of core count will be four core units
because each CC X does have four cores
the only reason you would really see
something less than this is if they're
disabling cores either to fill a market
or because a unit failed validation
which has happened in the past but
Andy's been moving away from things less
than four cores for quite some time now
so we would expect his end to stick to
that trend at least for the market that
we're most interested in on the gaming
and sort of enthusiast user side of
things so that's the main news items for
this video check out the rest of the
channel for reviews things like that we
have an EVGA hybrid comparison coming up
versus the MSI Seahawk and of course our
hydrographic same card but that's coming
up next week or very shortly anyway as
always thanks for watching patreon link
personal video I'll see you all next
time
you
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.