Author Topic: New Huananzhi board a raging success  (Read 3583 times)

Antariy

  • Member
  • ****
  • Posts: 564
Re: New Huananzhi board a raging success
« Reply #15 on: December 20, 2021, 11:51:24 AM »
If CPU have a big L3 cache, DDRAM speed just don't matter so much ?


Partially yes.

But also remember - x99 XEONS and 'cunsumer' CPUs (like i7seven's) (and x299 CPUs/platform, too, either) are 4-channel memory mode capable. Usual (non-server, and non-HEDT (so-called)) 'consumer' grade CPUs, like, let's say, even naming 'top ones', like 9900k or 10900k or 11900k, are using TWO channel memory mode (controller). Thus, 2133 MHz RAM 'speed' 4th gen Xeon is capable of the SAME throughput like 4266 MHz RAM on consumer (and overclockable, and overclockable M/B) CPU.

3647 Xeon's are 6-channel memory mode/controller. So... 'even' 2133 MHz DDR4 will perform like ORDINARY (user-land/consumer) memory bandwidth/throughput like, let's say, some (practically inexistent in sense of every day use, BTW) 6399 MHz DDR4. With 2400 MHz DDR4 memory you will actually having like 7200 MHz DDr4 on consumer board. This presumable will for many (several) years from now even to beat and/or to compete (few years later) with DDR5 speeds, lol, which is just 'fake' 4 channel, by the way.

TimoVJL

  • Member
  • *****
  • Posts: 1223
May the source be with you

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #17 on: December 20, 2021, 12:14:01 PM »
Interestingly enough, memory speed with Xeon CPUs is controlled by the CPU and in the case of Xeons, they don't need any more than 2400. You could get 2666 in the past but the last I bought was 3200 but the Xeon still runs the memory at 2133. As Alex mentioned, gen 5 and 6 i7s run 4 channel memory as do the R5 26xxv3 series Xeons but with the bigger Xeons, 40 PCIe lanes. I think from memory that my old 5820k i7 had 28 PCIe lanes.

Something I was pleased with is the rebuild of my old box with the 14 core Xeon, I thought it may have been a bit laggy and perhaps clunky but its a nice enough machine to use and does all of the normal low core count tasks well.

The later Xeon Silver and Gold versions are coming onto the second hand market but getting a decent desktop board is another matter, very expensive for both a board and CPU so I am in no hurry to try and raise that much money.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy:

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #18 on: December 20, 2021, 12:14:56 PM »
Timo,

Oops! That page can’t be found.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy:

Antariy

  • Member
  • ****
  • Posts: 564
Re: New Huananzhi board a raging success
« Reply #19 on: December 20, 2021, 12:29:34 PM »
Quad-channel RAM vs. dual-channel RAM: The shocking truth about their performance

Sorry, even don't going to bother to read it :) From it's NAMING i already see it's 'a shocking news' (pulp fiction lol) type of reading, where the COVERS are removed off the UNIVERSES' MISTIRIES. :D :biggrin:

For first attitude even thought not to continue this post to state something more, lol :biggrin: :tongue:

There are MANY 'reviews' on the web, and on youtube (which is also part of web  lol, though), which will prove something or other thing. The only thing is there - when you get the detail.

If they were comparing there the speed of RAM for GAMES - then - the CRITICAL for games are RAM delays/latencies. Not bandwidth (in a lot lesser extent). So, FAST enough RAM with LOW latencies will do the job for most of games (or about all of them). Even 2600 like RAM on TWO channel memory mode INTEL CPUs (AMDs of later eras, like Ryzen, have HT linkage to RAM speed, what is critical as impacts the inter-blocks communication speed, that's why AMDs cpus are claimed by people as 'fast RAM demanding', but we will not touch this topic here as too fat to discuss since even the above statement about 4 channel memory mode can lead to further discuss) will do OK if has low latencies (around 50ns is the thing which you must try to achieve at least) - FOR GAMES.

But, also remember, than for 4-channel RAM you CAN also setup RAM latencies very low (especially on BRAND board, especially with NON-ECC/NON-REG RAM).... so, factically, having, STILL, TWO TIMES as much throughput as two channel CPUs and M/Bs, lol, but also absolutely GAMABLE gameplay in like fastest (and most costly) DDR4 (or DDR3 is possible on X99, though....) memory kits, lol.

But - for work - like VIDEO/3D processing, designing, ETC ETC ETC ETC - high memory bandwith is a must, of course.
But, still, of course you ALWAYS still can to craft a 'test' which will 'prove' that 2-channel memory mode is not worst than 4-channel memory mode.

But, still, again, for some reason, think of it, BIG GUYS with BIG BUCKS chosen for server's and data processing nodes LOW FREQUENCY but HIGH-CHANNEL-COUNT memory 'models' :) For some reason, must be.

Still, not going to argue here about 'speedness' or 'channelness' of RAM, lol, if above said is not yet enough lol then it's not enough and will be kept that way, lol.

Antariy

  • Member
  • ****
  • Posts: 564
Re: New Huananzhi board a raging success
« Reply #20 on: December 20, 2021, 12:37:25 PM »
I think from memory that my old 5820k i7 had 28 PCIe lanes.

Yes, you're right, Hutch. 6-cores had it so way. IIRC 4 core versions on x299 very also pretty limited by the memory size and PCIe lanes as well.


https://ark.intel.com/content/www/us/en/ark/products/82932/intel-core-i75820k-processor-15m-cache-up-to-3-60-ghz.html

the reading 'Max # of PCI Express Lanes'.

Antariy

  • Member
  • ****
  • Posts: 564
Re: New Huananzhi board a raging success
« Reply #21 on: December 20, 2021, 12:45:37 PM »
As Alex mentioned, gen 5 and 6 i7s run 4 channel memory as do the R5 26xxv3 series Xeons

Only on x99 (socket 2011), by the way, i meant it that way! i7 6700 for example for 1151 socket - consumer CPU - is still two channel memory mode.

And, talking about model-naming - and generations - well, Intel is misleading with that - for at least 2011 socket. i7 5xxx are still 4th GEN CPU (Haswell core - the desktop/consumer analog is core off 1150 socket time/era/age). i7sevens off 6xxx are '5th gen' - Broadwell core (still had variation released (pretty limitedly by support and marketing) on 1150 socket, as well, too).

I.e., for example, the architecture of 6800k (on 2011 socket) and 6700k (on 1151 socket) ARE NOT THE SAME. While first one was BROADWELL (5th GEN), the latter/second one named was SKYLAKE (real 6th GEN). Core 'microarchitecture'.
Though - broadwell was not much, about NO at all, more performant by IPC (instruction per cycle count) than HASWELL. But it was lower techprocess (14nm agains 22nm), but overall Haswell had bigger extent of possibility of overclocking.

Gunther

  • Member
  • *****
  • Posts: 4115
  • Forgive your enemies, but never forget their names
Re: New Huananzhi board a raging success
« Reply #22 on: December 20, 2021, 03:59:25 PM »
Alex,

good and fundamental contributions, as always.  :thumbsup: Hopefully we will see you here more often now.
You have to know the facts before you can distort them.

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #23 on: December 20, 2021, 04:10:26 PM »
 Except for the availability of the socket 2011-3 Xeons, I probably would have gone the way that Gunther has with the socket 2066 CPUs but I do have the option of a 6950X 10 core extreme edition which can be clocked at 4 gig with no problems but I have to look at what I gain by doing so.

The 5820k clocked at 4 gig is no slouch and if I need big multi core performance, any of the Xeons will do the job and its not worth throwing more money at what are at best incremental gains, both the 12 core and 14 core Xeons plow through large data very well.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy:

Antariy

  • Member
  • ****
  • Posts: 564
Re: New Huananzhi board a raging success
« Reply #24 on: December 21, 2021, 10:45:45 AM »
 
Alex,

good and fundamental contributions, as always.  :thumbsup: Hopefully we will see you here more often now.


Hi again, Gunther, glad to see you too  :thumbsup:

Hutch says you're on LGA2066 now? Which CPU? Desktop/HEDT (x299) or Workstation/Xeon (C422) chipset/and/or CPU? I must to check all your latest messages to see the truth  :biggrin:


Except for the availability of the socket 2011-3 Xeons, I probably would have gone the way that Gunther has with the socket 2066 CPUs but I do have the option of a 6950X 10 core extreme edition which can be clocked at 4 gig with no problems but I have to look at what I gain by doing so.

The 5820k clocked at 4 gig is no slouch and if I need big multi core performance, any of the Xeons will do the job and its not worth throwing more money at what are at best incremental gains, both the 12 core and 14 core Xeons plow through large data very well.

Hutch, In fact, as the best bet/bang for your buck, yes, x99 is still unbeatable. In therms of cheapness per performance ration you get. Cheap used CPUs (but not M/Bs - if brand name and new, lol) - high core counts - VERY enough performance even for today (like i wrote in some of another threads, some months ago, reality shows that even latest before latest (lol) gen, like 11th gen, has not really reached more than 20%-30% of IPC increase over 4th gen - all other 'rumours' are pure marketing) - what's required more for hard number crunching/working?  :biggrin: The price/actuality(performance) ratio of the x99 CPUs are unbeatable to the moment, still. Though you can have cheap enough 8 core LGA2066 Xeons on aliexpress - but, hey, only 8 cores. Or you can get ES CPUs (desktop/HEDT - not Xeons) - for cheap - but with bugs you not know where will appear in your real life application.

You have top broadwell for 2011 huh? (6950X) Well, everyday (or even the more so 24/7) usage of overclockable CPUs with decent overclocking and decent LOAD is absolutely nonsense - they tend to consume TOO much (comparing to 'normal CPU' at the same frequency/load) even in absolute idle (as you have to raise voltage, depending on how much to overclock, of course, 4GHz for Haswell is 'nothing', and even for Broadwells it's far off limits so you have a big safety buffer maybe). I think that trying to replace the 14 or even 18 core Xeon with overclocked 6950X will make you to regret your choice finally - extreme heatage, power consumption - it's the only thing eXtreme CPUs do OUTSDANDING in overclocking condition at full loads - well - even in NON-overclocked conditions overclockable CPUs tend to 'eat' more energy/require more IDLE voltage, due to their leakage current is always higher than NORMAL/stable crystalls used for building Xeons and 'regular' desktop chips . BTW XEON chips - ARE TOP GRADE - i.e. in 'binning' stages they are more 'bug-free' and stable (cleanier/less defective chips), because they are intended to work for years 24/7 with maximum load and not to degrade noticeable (to not at all) - while eXtreme (overclockable CPUs) by all means are 'consumables'/'expendables'.


RED TEXT - EDITED - ADDED A TEXT AS FIRS WROTE NON-SENSIBLE/NOT TOTALLY IN A CONTEXT/OUT OF CONTEXT PHRASE
« Last Edit: December 21, 2021, 11:55:02 AM by Antariy »

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #25 on: December 21, 2021, 06:33:17 PM »
Hi Alex,

I have 4 X99 machines running, the 5820k i7 which I use for general purpose computing, 2 x 12 core E5-2690v3 boxes and I rebuilt my old box with a 14 core E5-2697v3 so I have all of what I need at the moment. There are faster and more powerful CPUs available but the X99 CPUs have the advantage of size and with a decent liquid cooler, they are not hard to keep within a safe temperature range.

The old 5820k i7 run the hottest at about 65c when thrashed on all 6 cores but its clocked at 4 gig. On the other end, the 14 core box runs at about 45c while being thrashed on all cores and it drops to 2.9 gig. The 12 cores are inbetween. I long ago learnt that keeping the temperature down made CPUs last just about forever and by not frying the board, they last longer as well.

Where I live in Sydney, we get hot summmers so decent cooling is necessary. 3 Boxes have double fan radiators, the old box has a triple fan radiator with fans in push / pull configuration.

The ultimate temperature control is between the chip and the CPU lid and you see some guys delidding a CPU to try a better junction but you risk trashing the CPU for little gain.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy:

daydreamer

  • Member
  • *****
  • Posts: 2185
  • my kind of REAL10 Blonde
Re: New Huananzhi board a raging success
« Reply #26 on: December 22, 2021, 12:37:52 AM »
So how much performance gain a very hot summers day do you gain in water-cooled x99 vs original cooling?
Might be easier measure cooling that isnt good enough to be able over clock vs overclocked performance possible thanks to water cooling

If I stop turbo and/or underclock laptop to prevent. It go full turbo all the time on legacy game that is designed to run as fast as possible, but only require less ghz, hope it helps lifespan
my none asm creations
http://masm32.com/board/index.php?topic=6937.msg74303#msg74303
I am an Invoker
"An Invoker is a mage who specializes in the manipulation of raw and elemental energies."

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #27 on: December 22, 2021, 01:51:13 AM »
Performance gain is relative to what a badly cooled CPU runs at. If it run hot enough to thermally throttle, then the cooling is not adequate.

Electronics last much longer if they are not being cooked, 90c and higher is dangerous and you are relying on thermal throttling by the CPU. The gaming guys start with a fast CPU at least on the first few cores, maximise the cooling (big radiators) then tweak the clock speed to survive at a viable temperature.

The i7 5820k I use for my general purpose box happily runs at 4k as I have enough cooling for it and it peaks under maximum load at about 65c which is OK. The Xeons run cooler but they are locked frequencies that cannot be overclocked.

A laptop is not really the right hardware to thrash as their size prevents any serious cooling. Very expensive ones have better cooling but they are no big enough to cool properly under heavy loads.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy:

mikeburr

  • Member
  • **
  • Posts: 184
Re: New Huananzhi board a raging success
« Reply #28 on: January 09, 2022, 01:56:45 AM »
id be very interested to know what improvement  in performance moving the OP Sys to an NMVE drive has
installed a  drive here but havent tried it yet
has anyone else seen significant improvements doing this ??
regards mike b
 

hutch--

  • Administrator
  • Member
  • ******
  • Posts: 9988
  • Mnemonic Driven API Grinder
    • The MASM32 SDK
Re: New Huananzhi board a raging success
« Reply #29 on: January 09, 2022, 02:12:56 AM »
I have been very pleased with each of the Samsung NVMe drives I have used. Except for my oldest box which I have just modernised, the other 3 have 500 gig Samsung gen 3 NVMe drives and they all benchmark at 3.2 write, 3.5 read. My oldest box has an Intel Gen 1 NVMe drive which after 6 years is still perfect. It has a gen 3 NVMe as well.

I do a tweak on NVMe drives, set the OS parttition at about 200 gig and use the rest for the next partition where I load stuff that must be fast loading.
hutch at movsd dot com
http://www.masm32.com    :biggrin:  :skrewy: