VIA ZhaoXin KX6000 benchmark

dark zero

Platinum Member
Jun 2, 2015
2,516
8
91
#1
Ok... This is weird but I found on Geekbench 3 this benchmark from VIA on Ubuntu...

http://browser.geekbench.com/geekbench3/8658463

Single Core score is 1670, on par of the olde Intel Core 2 Duo at 45nm. Is not that impressive unless we consider that they were on Atom levels....

And Multi Core is on 10K which is on Core i5 Skylake levels and that is pretty much impressive since they were pretty much on 5K on the previous generation...

http://browser.geekbench.com/geekbench3/8658365

Seeing the difference of performance between the previous and the current one, VIA did just made a Conroe Jump?

I didn't saw that jump since those days. And yeah, is an octocore with weaker cores than the Skylake i5.

Also, seems that the process used is on 16nm.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,439
196
126
#2
Seeing the difference of performance between the previous and the current one, VIA did just made a Conroe Jump?
No. The KX6000 only brings in the newer 16nm S3(Vk(Vulakn)/S(DX12)) GPU and integrates the upgraded ZX200+ chip.
 

NTMBK

Diamond Member
Nov 14, 2011
8,365
358
126
#4
Ok... This is weird but I found on Geekbench 3 this benchmark from VIA on Ubuntu...

http://browser.geekbench.com/geekbench3/8658463

Single Core score is 1670, on par of the olde Intel Core 2 Duo at 45nm. Is not that impressive unless we consider that they were on Atom levels....

And Multi Core is on 10K which is on Core i5 Skylake levels and that is pretty much impressive since they were pretty much on 5K on the previous generation...

http://browser.geekbench.com/geekbench3/8658365

Seeing the difference of performance between the previous and the current one, VIA did just made a Conroe Jump?

I didn't saw that jump since those days. And yeah, is an octocore with weaker cores than the Skylake i5.

Also, seems that the process used is on 16nm.
What has power consumption done? The reason Conroe was so impressive was because it significantly dropped power use at the same time.
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,439
196
126
#5
But, why the performance difference between that and the KX5000 series is so big?
Same architecture, better clocks from 16nm. There might also be an upgraded fabric in the SoC to support the integrated Southbridge(ZX200+).

Uncore enhancements + Frequency from FinFETs.
 

amd6502

Senior member
Apr 21, 2017
600
117
86
#6
Same architecture, better clocks from 16nm. There might also be an upgraded fabric in the SoC to support the integrated Southbridge(ZX200+).

Uncore enhancements + Frequency from FinFETs.

I'm impressed!

The integer multicore is similar to that of a FX-6300 (in fact it slightly beats it.) http://browser.geekbench.com/geekbench3/compare/5971809?baseline=8658463

Since 3.5ghz hex core should be similar to 3 ghz octacore (slightly in favor of octacore 8*3=24>21=6*3.5) we are talking multicore-IPC almost to that of Piledriver.
 

Gideon

Senior member
Nov 27, 2007
485
134
136
#7
we are talking multicore-IPC almost to that of Piledriver.
That is an impressive gain for VIA! But one must also remember that piledriver has a multicore-IPC of about The Phenom II, which is an almost 10 year old CPU.

Now let's compare:
Thuban vs ZhaoXin

ST: 1985 1670
MT: 10045 10069

Numbers look quite impressive considering AMDs clock advantage (especially multi-core), but one must remember that the VIA chip is 8-core vs a 6-core Phenom.

On Closer inspection though. There are a couple of things skewing results towards VIA:

1. It has dedicated hardware blocks for AES and SHA1 which boost it's scores a ton compared to phenom (same is true vs Bulldozer):


2. It probably uses DDR4 and has loads of more bandwidth (all the memory test scores are almost 2x in it's favor):


Yet despite those things, it get's beaten in most of the other tests (and quite badly in Floating Point).
Good progress, but there is still a long way to go. Hopefully they manage to get to Sandy-Bridge levels of IPC on the next run.
 

amd6502

Senior member
Apr 21, 2017
600
117
86
#8
That is an impressive gain for VIA! But one must also remember that piledriver has a multicore-IPC of about The Phenom II, which is an almost 10 year old CPU.

Now let's compare:
Thuban vs ZhaoXin

ST: 1985 1670
MT: 10045 10069

Numbers look quite impressive considering AMDs clock advantage (especially multi-core), but one must remember that the VIA chip is 8-core vs a 6-core Phenom.

On Closer inspection though. There are a couple of things skewing results towards VIA:

1. It has dedicated hardware blocks for AES and SHA1 which boost it's scores a ton compared to phenom (same is true vs Bulldozer):


2. It probably uses DDR4 and has loads of more bandwidth (all the memory test scores are almost 2x in it's favor):


Yet despite those things, it get's beaten in most of the other tests (and quite badly in Floating Point).
Good progress, but there is still a long way to go. Hopefully they manage to get to Sandy-Bridge levels of IPC on the next run.
So I'm talking per core multicore IPC, which (for integer and mixed code) is is basically per core IPC times the piledriver penalty for shared decoders and cache. I think Thuban's k10 is slightly higher IPC than Piledriver. So all of these IPC's are very comparable. (The k10 just lacks modern encryption instructions so it depends how you weight that in. )

The KX is doing all this on just a 65W tdp which includes the iGPU wattage. So it's much more power efficient than an FX-63xx which will take about 95W-125W and perform similarly, and slightly more efficient than an FX-8300 clocked at 2.9ghz. (On my 8300, linux 'sensors' command reads values somewhere around 68 Watts when I have all cores loaded.)

The FPU is weak because they apparently must have smaller FPUs per core, probably similar amount of FPU units as in dozer octacore/quadmodule.

It should have some appeal in China for those that want a homegrown CPU. Firefox or chrome can use all those threads, but at the cost of memory usage. On my 8300 I have just 3 threads enabled for now, because that's what gives more than enough browsing performance. So even their quadcore 3ghz should work really fine as a basic home PC. The octacore might be more for home and office servers or geek toys. Not too bad of a lineup.

I really doubt they'll have 7nm ready for 2019 like scheduled; see table https://en.wikipedia.org/wiki/Zhaoxin given what's going on with everyone except for TSMC. I wonder if it's practical to move their next gen design to 22FDX for a small 4 to 6 core APU so they can ultimately go to 12FDX which seems like a better bet for mass produced budget products.
 
Last edited:

Asterox

Senior member
May 15, 2012
244
118
116
#9
Ok... This is weird but I found on Geekbench 3 this benchmark from VIA on Ubuntu...

http://browser.geekbench.com/geekbench3/8658463

Single Core score is 1670, on par of the olde Intel Core 2 Duo at 45nm. Is not that impressive unless we consider that they were on Atom levels....

And Multi Core is on 10K which is on Core i5 Skylake levels and that is pretty much impressive since they were pretty much on 5K on the previous generation...

http://browser.geekbench.com/geekbench3/8658365

Seeing the difference of performance between the previous and the current one, VIA did just made a Conroe Jump?

I didn't saw that jump since those days. And yeah, is an octocore with weaker cores than the Skylake i5.

Also, seems that the process used is on 16nm.
Well except Geekbench, KX 6000 CPU performanse expectations or estimates it is old news.

Fritz Chess Benchmark is one and half year old information.Manufacturing process for KX-6000/16nm, this is also old news or announced almost a year ago.

http://www.kejixun.com/article/170317/298541.shtml

https://semiaccurate.com/forums/showthread.php?8505-Zhaoxin-KX-5000-amp-KX-6000



KX-6000 is just first step or bigger jump in CPU performance.New KX-7000 series(7nm) will be much more impresive no doubt."VIA+Zhaoxin money is not a problem", with KX-7000 they targeting at least Intel Sandy Bridge IPC.
 

dark zero

Platinum Member
Jun 2, 2015
2,516
8
91
#10
I would forgive VIA this time then due the massive improvements. BTW, seems that is rated 37 watts the chip like the previous generation.
 

amd6502

Senior member
Apr 21, 2017
600
117
86
#11
I would forgive VIA this time then due the massive improvements. BTW, seems that is rated 37 watts the chip like the previous generation.
Maybe the quadcore? My googoling points to 65W.
 
Oct 14, 2003
6,256
357
126
#13
It's more like Goldmont(not Goldmont Plus) in Geekbench 3. It's running at 3GHz versus N4200's 2.5GHz so about 10% slower? Individual results are quite comparable between the two.

http://browser.geekbench.com/geekbench3/compare/8603222?baseline=8658463

What has power consumption done? The reason Conroe was so impressive was because it significantly dropped power use at the same time.
Agree. Also because Conroe brought a big gain at the highest tier, which is much harder to do.
 
Last edited:

amd6502

Senior member
Apr 21, 2017
600
117
86
#14
As far I know the Quad Core was at 18.5W.... At 16 nm I hardly see it at that wattage.
Yes you're right wattage for C46XX is 18W, and that is pretty crazy efficient for 28nm and they ran at ~2 ghz. However, the 40nm chipset and and iGPU would add another ~15W (100s chipset, see http://en.zhaoxin.com/SupportDown.aspx which is the only source of wattage info I can find). This was later reduced to just 6W for the external IO portion of the soc, in the followup generation 200s/KX-5000 series, according to https://en.wikichip.org/wiki/zhaoxin/microarchitectures/wudaokou due moving most (all but IO) of the chipset function onto the CPU.

C46XX was very area efficient too, with just 300M transistors for the quadcore (though remember this doesn't include soc or igpu). The KX-5000 octacore soc cpu+igpu transistor count grew to 2100M transistors (so this puts quadcore count >1050M transistors for the soc+igpu).

37 watts for the 16nm octacore APU sounds pretty plausible, and that's mighty impressive for efficiency; even if it were 45W, that's great for 8x 3ghz. I think it would appeal even in the western world, for home servers and geekboxes.
 
Oct 14, 2003
6,256
357
126
#15

NostaSeronx

Platinum Member
Sep 18, 2011
2,439
196
126
#16
ZHAOXIN KaiXian KX-U5580 => CentaurHauls Family 7 Model 11 Stepping 1
KX-6840 => CentaurHauls Family 7 Model 11 Stepping 0

The D4600 is ZX-C.
ZXD D4600 => CentaurHauls Family 6 Model 31 Stepping 14
 
Oct 27, 2006
19,794
302
126
#17
This fascinates me. I'd love to grab a Mobo/CPU at some point down the line and play with it a bit. How long has it been since there was a 3rd CPU mfg that made something not completely awful for desktop type loads?

Any idea what pricing and motherboard availability is like? Drivers? Features like USB-C and nVME?
 

NostaSeronx

Platinum Member
Sep 18, 2011
2,439
196
126
#18


ASK THE COMMUNITY

TRENDING THREADS