News ASRock show-off Micro-STX system with MXM & TB3 @CES 2017

Hi All

ASRock are happy to display our new model - Deskmini RX/GTX @CES 2017.
You can get more picture in :https://www.cool3c.com/article/116315/CES-2017-華擎展出具可擴充-MXM-獨立顯示卡的-micro-STX-主機板原型主機


This is prototype of Deskmini RX/GTX
Considering there are so many SFF enthusiasts still need more powerful 3D graphics performance, why don't we extend Mini-STX to larger board: based on 5x5 screw holes but implement MXM, more video outputs, m.2 connectors, even Thunderbolt 3!

And that is world's firsts' MICRO-STX system:
  • Support up to Intel® Core i7 7700K CPU
  • Support Intel box cooler
  • Intel Z270 (B250) Chipset
  • Support MXM Type B, up to 120W
    • RX = AMD RX480/470/460
    • GTX = NVIDIA GTX1060
  • Intel® Thunderbolt 3 with Type-C
  • 3 x M.2 2260/2280 slots
  • 1 x M.2 2230 Wi-Fi slot
  • AURA RGB LED connector!
  • Support Intel® Optane™ Technology
  • 2.x Liter!
  • 19V/220W adapter
ASRock planned sell barebones (include motherboard, MXM card, MXM cooler, chassis and adapter) and open mind corporate with system builder to develop new mini PC in micro STX motherboard. It should be launched on 2017 Q2.
 
Last edited:

Kmpkt

Innovation through Miniaturization
KMPKT
Feb 1, 2016
3,382
5,935
I don't think that Nvidia is quite as Tick/Tock bound as say Intel. If I'm not mistaken Maxwell to Pascal was a die shrink (28nm to 16nm) combined with a number of significant architectural improvements from one generation to the next. This is why we saw such a massive leap in performance combined with a massive reduction in power consumption relative to that performance.
 
  • Like
Reactions: Phuncz

EdZ

Virtual Realist
May 11, 2015
1,578
2,107
If we go by GV100, Volta is a significant departure in microarchitecture from Pascal as well as shifting to a smaller process.
Anandtech said:
Before we kick things off, one thing to make clear here - and this is something that I'll get into much greater detail when NVIDIA releases enough material for a proper deep dive - is that Volta is a brand new architecture for NVIDIA in almost every sense of the word. While the internal organization is the same much of the time, it's not Pascal at 12nm with new cores (Tensor Cores). Rather it's a significantly different architecture in terms of thread execution, thread scheduling, core layout, memory controllers, ISA, and more. And these are just the things NVIDIA is willing to talk about, never mind the ample secrets they still keep.
As we know from GP100 not all the new features trickle down to the consumer chips in order to keep die areas smaller and costs down, so likely we will not see the new Tensor Cores for one thing. FP cores are another unknown: high FP64 rates are not really necessary for consumer workloads, so a direct transplant of the FP cores from GV100 is unlikely. Those cores subdivide to provide single-precision and half-precision workloads (i.e. 1 FP64 unit splits to two FP32 units or 4 FP16 units), so a consumer die would either need a newly developed unit that is natively FP32 and splits to double FP16 (or mangle parts of the GV100 FP64 core to shrink it a bit while losing the FP64 capability, less efficient than a newly designed core but cheaper to develop), or possibly just re-use the FP32 units from Pascal if an increased FP16 rate is not necessary.
AMD went the route of using FP32 units that split to dual FP16 units with a separate smattering of dedicated FP64 units for Vega, but they don't have a comparable 'big die' with sub-dividable FP64 units in the first place.
 
  • Like
Reactions: Phuncz

SFF Scrub

Average Stuffer
Aug 7, 2017
56
16
Shit. I accidentally changed the conversation to GPU. Just to finish up the previous conversation, Coffee Lake in majority is also a die shrink.

Compared to Kaby Lake, Coffee lake has (predicted) slower Quad-core chips, but introduces Hexa-core (probably where their money went) because even with six cores the chip is pulling less power than the flagship Kaby lake Quad-core; Coffee flagship 95W / Kaby flagship 112W. The only other thing that I can spot right not with Coffee Lake is that its L3 cache is quite larger 8MB to 12MB.
 

BeerNsoup

Cable-Tie Ninja
Mar 12, 2017
205
149
Shit. I accidentally changed the conversation to GPU. Just to finish up the previous conversation, Coffee Lake in majority is also a die shrink.

Compared to Kaby Lake, Coffee lake has (predicted) slower Quad-core chips, but introduces Hexa-core (probably where their money went) because even with six cores the chip is pulling less power than the flagship Kaby lake Quad-core; Coffee flagship 95W / Kaby flagship 112W. The only other thing that I can spot right not with Coffee Lake is that its L3 cache is quite larger 8MB to 12MB.

I'm FAR from an expert so I'm prepared to be told I'm wrong... isn't l3 cache tied in chunks to each core? so 12mb for 6 cores, or 8mb for 4 cores, is effectively the same amount? I've seen it mentioned as "x"mb per core (maybe I'm misinterpreting it and various sites are just dividing the total l3 cache by the amount of cores).

Is it actually just a block to be used by all cores and if one core wants to use more it can use more than its "share" if other cores are being under/less utilized?

Guess what I'm asking is does the l3 increase actually make more available in single/few core performance, or is it just increased to have the same amount per core/available to each core as previous generations?
 
  • Like
Reactions: AleksandarK

EdZ

Virtual Realist
May 11, 2015
1,578
2,107
Guess what I'm asking is does the l3 increase actually make more available in single/few core performance, or is it just increased to have the same amount per core/available to each core as previous generations?
L3 is per-core, but cores can grab data from the L3 of other cores via the Ring Bus for Intel's small die chips. For Skylake-X they've moved to a mesh network with a slightly higher but more consistent inter-core latency, and Ryzen/Threadripper use a 'tree' hierarchy with various different Infinity Fabric busses (cores in the same CCX can access each other's cache quickly, cores between CCXes have a larger latency penalty due to the inter-CCX link, cores on different physical dies have an even larger latency penalty due to the link having to go through the substrate).
 

EdZ

Virtual Realist
May 11, 2015
1,578
2,107
Mystery solved on those two 'missing' USB pots: they;re on the side of the front panel!
Oddly, while there is a Z270 version listed as barebones & motherboard-only options, the variants that include a GPU are all use B250.
 
  • Like
Reactions: Phuncz and Weredawg

alinescoo

Average Stuffer
Feb 3, 2017
57
14
Indeed, it is a good price.

I am thinking that a case like Lazer3D + SFX PSU + MB + 1060ITX is a bit cheaper to get and that is quite a small case. Indeed, it's double+ in liters however the upgrade options are way numerous (you could easily switch from AMD to Intel), the cooling seems very capable... Also, new Coffee Lake is around the corner...
I'll have to give it a good think on what to get.
 

dotch

Caliper Novice
Mar 25, 2017
24
8
Awesome product!

At this price it probably even makes sense to get it and transfer the mainboard, GPU and power supply to a smaller case, right?
(Not a fan of the large height of this)
 

Weredawg

SFF Lingo Aficionado
Apr 5, 2017
119
162
Last edited:

rcodi

SFF Gamer
Aug 5, 2017
176
165
This thing is tiny! The only thing that concerns me is the MXM heatsink; whether or not it will fit other cards and if it has any headroom for higher TDP cards.
 

Kmpkt

Innovation through Miniaturization
KMPKT
Feb 1, 2016
3,382
5,935
This thing is tiny! The only thing that concerns me is the MXM heatsink; whether or not it will fit other cards and if it has any headroom for higher TDP cards.

I'll be able to answer that for you next week for certain, but I believe fan mounting points are common to the spec. I have an RX580 being delivered with a fan next week and am planning to use it on my 1070 which I currently do not have a fan for.
 
  • Like
Reactions: Curiosity and rcodi