• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Zhaoxin's ZX-F/KX-7000/KH-40000

Page 14 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
Didn't find any Jasper Lake, but this N5000 score gets 1232.
Yeah I can find Goldmont/Goldmont Plus and in a few nooks and crannies, Lakefield (fyi Lakefield scores an ~1870 or so; unsure of the numbers due to the weird scheduler issues with Lakefield in some benches).
 

bononos

Diamond Member
Aug 21, 2011
3,732
70
91
Are Zhaoxin cpus commercially available in quantity? Are they sold/used by the Chinese govt? I remembered someone posting about Baikal cpus being used by the Russian govt (not sure).
 

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
So just some napkin math:

The n5000 with an all-core burst of 2.6 GHz and an all-core sustained speed of . . . actually I don't know at all, so there's no telling what clockspeed it runs in CBR23. Ahem. Right. Anyway, the N5000 at ???? clockspeed (that is most certainly lower than 2.6 GHz) scores 1232 in 4c/4t configuration as Goldmont Plus.

A 4c/4t Tremont with the same boost behavior would probably score ~1600 (+30%)
A 4c/4t Gracemont with the same boost behavior would probably score ~1920 (+20% ???)

I would love to have some Jasper Lake results for comparison (with sustained clocks in CBR23) but alas they're hard to find.

Anyway, the KX-U6580 with twice as many cores and a sustained clock of 2.5 GHz isn't looking that much better, before we even take into account power. A hypothetical 8c Goldmont Plus would already be faster (~2400), not to speak of Tremont or Gracemont. Intel's "small cores" seem better per core in this task.
 

Kosusko

Member
Nov 10, 2019
115
73
71
In other words.

In the desktop: Mini-ITX, Nettops etc.
Four small cores with a higher IPC but also with a higher frequency or eight small cores with a lower IPC but also a lower frequency are better.

P.S. here is Pentium Silver N6000 (Tremont) in Cinebench R23 (1 680 pts) and other benchmarks.






I don't like cross-review comparisons. But...

Cinebench R23
● Intel Pentium Silver N6000 1.1GHz ~ 3.3GHz (10nm, 4C/4T, 1.5MB L2 cache, 4MB L3 cache): 1 680 (75,44%)
● ZHAOXIN KaiXian KX-U6580 2.5GHz (16nm, 8C/8T, 8MB L2 cache): 2 227 (+32,56%)

Fritz Chess

● Intel Pentium Silver N6000 1.1GHz ~ 3.3GHz (10nm, 4C/4T, 1.5MB L2 cache, 4MB L3 cache): 3 304 (34,98%)
● ZHAOXIN KaiXian KX-U6580 2.5GHz (16nm, 8C/8T, 8MB L2 cache): 9 730 (+185,84%)

WinRAR

● Intel Pentium Silver N6000 1.1GHz ~ 3.3GHz (10nm, 4C/4T, 1.5MB L2 cache, 4MB L3 cache): 3 302 (57,88%)
● ZHAOXIN KaiXian KX-U6580 2.5GHz (16nm, 8C/8T, 8MB L2 cache): 5 187 (+72,78%)

Of course. I'm still waiting to compare the mainboard with Tremont QuadCore processor SoC...
 
Last edited:
  • Like
Reactions: lightmanek

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
Four small cores with a higher IPC but also with a higher frequency
There's no way an N6000 sustains 3.3 GHz in CBR23. That's the single-core boost. Good on you for finding an N6000 score though! Looks like my estimates were close.
 

Kosusko

Member
Nov 10, 2019
115
73
71
Anyway, the KX-U6580 with twice as many cores and a sustained clock of 2.5 GHz isn't looking that much better, before we even take into account power. A hypothetical 8c Goldmont Plus would already be faster (~2400), not to speak of Tremont or Gracemont. Intel's "small cores" seem better per core in this task.

There's no way an N6000 sustains 3.3 GHz in CBR23. That's the single-core boost. Good on you for finding an N6000 score though! Looks like my estimates were close.
I will test processors as they are and not as they should be. Not hypothetically.

Waiting for motherboard with Intel Pentium Silver N6005 2.0GHz ~ 3.3GHz (10nm, Tremont 4C/4T, 1.5MB L2 cache, 4MB L3 cache) as well as the motherboard with Intel Pentium Silver J5040 2.0GHz ~ 3.2GHz (14nm, Goldmont Plus 4C/4T , 4MB L2 cache).

The only thing that will be a bit hypothetical resp. conclusion of the review for the future = Alder Lake platform also with eight small Gracemont cores... but of course only as a combination of bigLITTLE ie. with Golden core big cores.

I guess not only that will be Rise of the small cores
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
I will test processors as they are and not as they should be. Not hypothetically.
So you don't want to discuss cores anymore? That was the whole point of my analysis: VIA/Zhaoxin's "small" cores don't perform so well on a per-core basis vs. Intel's, even from a few generations ago. But hey if you want to deal in "actual" products:


As you can see, 8c Tremont does exist, with a base frequency of 2.2 GHz. They're used in comm gear since Intel doesn't sell 8c Tremont to the consumer market. Maybe you should ask some critical questions about WHY Intel doesn't sell those products on the consumer market and why VIA/Zhaoxin does (or has to; they can't sell anything else).
 

Kosusko

Member
Nov 10, 2019
115
73
71
So you don't want to discuss cores anymore?
Excuse me?
However, there will be a review.
But not about hypothetical products resp. paper dragon.


That was the whole point of my analysis: VIA/Zhaoxin's "small" cores don't perform so well on a per-core basis vs. Intel's, even from a few generations ago. But hey if you want to deal in "actual" products:


As you can see, 8c Tremont does exist, with a base frequency of 2.2 GHz. They're used in comm gear since Intel doesn't sell 8c Tremont to the consumer market. Maybe you should ask some critical questions about WHY Intel doesn't sell those products on the consumer market and why VIA/Zhaoxin does (or has to; they can't sell anything else).

Where can I buy an Intel Atom Processor P5921B with Tremont eight cores (mainboard embedded / industrial / server) for vertical market? Isn't it just a paper dragon since Q1'20 ?
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
Excuse me?
Comparing the performance of Intel's "small cores" - Goldmont Plus, Tremont - vs. the cores in the KX-U6580 (or any other CPU they sell on a "per core" basis, since you refuse to benchmark a KX-U6580 against something in the same power envelope.

As to where you can buy a P5921B, you can get one from a telecom hardware/services provider. Or you could act line a sane human being and just bench against a 4C TigerLake-U or something. Either CPU will wipe the floor with a KX-U6580, which is why you respond with rhetorical devices and crow about a KX-U6580 outperforming a CPU with a TDP of 6W.
 

Kosusko

Member
Nov 10, 2019
115
73
71
I have been working with small cores for about 20 years. Mostly in vertical hardware systems (embedded, industrial, digital signage, etc) as well as horizontal hardware systems (consumer). So I beg you ...

When the arguments run out, you use BIG cores and their processors to compare with small cores and their processors. Like several times.

P.S. Please find us mainboard for a real comparison with Intel Atom P-series processors (10nm Tremont eight cores).
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
When the arguments run out, you use BIG cores
Let me spell it out for you:

INTEL DOESN'T SELL 8c TREMONT ON THE CONSUMER MARKET BECAUSE THEY DON'T HAVE TO.

The market for anything more than a 4c Atom in the consumer space is small enough and sufficiently cash-poor that it doesn't make any sense for Intel to deploy more Atom cores. The only reason why VIA/Zhaoxin hawks 8c CPUs like the KX-U6580 with "small" cores is that they don't have a better product.

Inevitably, the end-user doesn't care how "big" or "small" are the cores. The only factors that matter are price, performance, and power consumption. VIA/Zhaoxin's latest products do not perform well when you consider how much it costs to get your hands on one (especially in the States!) and how much power they draw/heat they produce. They are improving against their older products, but they have a long way to go. Intel could saturate VIA/Zhaoxin's target market with cheap 8c Tremont or Gracemont if they felt like it, but there isn't enough money in it for them to even bother.
 

Kosusko

Member
Nov 10, 2019
115
73
71
There is no dispute in this. I still agree with that.
This forum is 99% consumer forum, so I will compare small cores from VIA / Centaur / Zhaoxin with small cores from Intel Tremont in the consumer (QuadCore with higher frequency and higher IPC) segment and especially realistically available in the vertical segment. No paper dragon !!!

Just as I will not consider the BIG Centaur CNS cores (CHA SoC) at present as better product because also is only paper dragon.
 
Last edited:

Kosusko

Member
Nov 10, 2019
115
73
71
I'm very excited with the Intel Pentium Silver N6000 with four Tremont cores.
It will be a very good duel.
I'm not worried about Alder Lake at all.

But let's not get ahead of ourselves.

The marathon of the rise of small cores is just beginning...

Stay tuned and don't go away.
 

Thunder 57

Golden Member
Aug 19, 2007
1,684
1,733
136
I'm very excited with the Intel Pentium Silver N6000 with four Tremont cores.
It will be a very good duel.
I'm not worried about Alder Lake at all.

But let's not get ahead of ourselves.

The marathon of the rise of small cores is just beginning...

Stay tuned and don't go away.

You can polish a turd, but it's still a turd. And what are you going on about regarding small cores? Nobody cares if a core is large or small. They care about performance, power, and price.
 
  • Like
Reactions: Markfw

Kosusko

Member
Nov 10, 2019
115
73
71
Rise of the small cores








You keep dropping links and images without commentary.
Your posting "rise of the small cores" means nothing as its some sort or mantra
that you have posted nine times in your past dozen or so posts.


esquared
Anandtech Forum Director
 
Last edited by a moderator:
  • Like
Reactions: lightmanek

Kosusko

Member
Nov 10, 2019
115
73
71
Little Atom E-Cores Gracemont - coming out this week.
This is the performance of today's small cores.

Tremont vs LuJiaZu
 

DrMrLordX

Lifer
Apr 27, 2000
18,204
7,098
136
N6000 does extremely well considering it's only a quad core. And those ST scores, oof. VIA/Zhaoxin has a lot of work left to do. Also convenient that you have not included any power numbers.
 
  • Like
Reactions: lightmanek

NostaSeronx

Diamond Member
Sep 18, 2011
3,326
860
136
Also convenient that you have not included any power numbers.
My invasive google searching:
N6000: 18W for 15 seconds, 6W TDP/sustained power.
U6580: 70W TDP, 61W sustained power.

So, power-wise N6000 is 1/10th the power of U6580.

Going towards the actual topic:
ZX-F/KX-7k appears to be a quad-core with a 15W TDP and KH40k is a dual-die dual-socket implementation of ZX-F(2x4c * 2socket). Both of which are TSMC 16nm.

Probably have to wait for ZX-G in 2023? on SMIC's real 7nm?

These aren't real ZX-F's but Centaur parts as development systems:
Hence, CHA001 & Codename CHA

Hence, given the current roadmap the die should still be CHA-esque but with a S3/Zhaoxin DX12 IGP:
zxfrealdie.png
zxfrealdie2.jpg

KX-7k
4c/2DDR4/16pcie out 44pcie enabled/DX12-igpu, 2 64-bit DDR4 with up to 4 DDR4 DIMMs.
KH40k
4c+4c/2+2DDR4/(32+32)64 out of (44+44)88pcie enabled/DX12+DX12 igpu * 2 socket, 8 64-bit DDR4 with up to 16 DDR4 DIMMs. GPU potentially isn't used for graphics, but compute in KH40k.
 
Last edited:

MadRat

Lifer
Oct 14, 1999
11,668
46
91
How is this CPU going to compete against modern CPUs for actual silicon. If they are spending to compete against far smaller processes that are off the shelf, then surely it is a money pit. No economy of scale. No hope for performance parity. Draining excessive electricity at a performance per Watt that is unjustifiable when there are so many efficient options in comparison. No wonder car makers cannot find silicon, when stuff like this soaks up global production capacity.
 

ASK THE COMMUNITY