It seems that you're using an outdated browser. Some things may not work as they should (or don't work at all).
We suggest you upgrade newer and better browser like: Chrome, Firefox, Internet Explorer or Opera

×
avatar
Shadowstalker16: Did you do all four benchmarks with SMT on? Also, what is 8/16T/C?
avatar
clarry: 8C/16T means 8 cores with 16 threads, i.e. SMT enabled. 8C/8T is 8 cores = 8 threads, which is what you get with SMT disabled.
Maybe I'm mistaken but from what I know, the number of threads listed is per core, and not per processor, which means that you get 8 cores * 8 threads per core, not one thread per each core.

As for performance and number of threads/SMT enabled/disabled, it depends on the threads themselves and how much yealding they do voluntarily. If the threads aren't bound by I/O and use no system services (which is to say they're 100% CPU bound) then 1 thread per core is optimal, but this isn't generally a use case. If the threads do I/O, synchronization and the rest of things which usually happen when software runs on a PC then having more threads available is a good thing*.

* but not too many threads; after a point you get into performance degradation because of scheduling overhead and all of the context switches, but to reach that point you generally need lots of threads, and run into it on multiple socket high end servers.
Post edited March 06, 2017 by AndrewC
avatar
clarry: 8C/16T means 8 cores with 16 threads, i.e. SMT enabled. 8C/8T is 8 cores = 8 threads, which is what you get with SMT disabled.
avatar
AndrewC: Maybe I'm mistaken but from what I know, the number of threads listed is per core, and not per processor, which means that you get 8 cores * 8 threads per core, not one thread per each core.
No, from the context (discussion about possible SMT regression on Windows 10) it is obviously hardware core/thread count, which is 8/16 and 8/8 with and without SMT, respectively. The results are comparing performance with SMT enabled and disabled, on two versions of the OS.

You don't see more than 2 threads per core in any consumer CPU.
Post edited March 06, 2017 by clarry
avatar
fishbaits: It's been shown that something in win10 is hampering Ryzen processors, so guess folk'll still have to wait a few weeks to see if micro$oft release the drivers.

Windows 10 - 1080 Ultra DX11:

8C/16T - 49.39fps (Min), 72.36fps (Avg)
8C/8T - 57.16fps (Min), 72.46fps (Avg)

Windows 7 - 1080 Ultra DX11:

8C/16T - 62.33fps (Min), 78.18fps (Avg)
8C/8T - 62.00fps (Min), 73.22fps (Avg)
Windowses are getting heavier and heavier with every new version (95->98->Me->XP->7->etc). System requirements are more demanding with every new version. Every new "feature", "bells and whistels", etc, consumes resources.

I remember hardware websites had articles on how to turn off unneeded Windows services through "services.msc".
When you turn unneeded services off, you free up system resources (RAM, some CPU cycles maybe) and your games run faster.

Maybe that's the case with Windows 10.
avatar
fishbaits: It's been shown that something in win10 is hampering Ryzen processors, so guess folk'll still have to wait a few weeks to see if micro$oft release the drivers.

Windows 10 - 1080 Ultra DX11:

8C/16T - 49.39fps (Min), 72.36fps (Avg)
8C/8T - 57.16fps (Min), 72.46fps (Avg)

Windows 7 - 1080 Ultra DX11:

8C/16T - 62.33fps (Min), 78.18fps (Avg)
8C/8T - 62.00fps (Min), 73.22fps (Avg)
avatar
vsr: Windowses are getting heavier and heavier with every new version (95->98->Me->XP->7->etc). System requirements are more demanding with every new version. Every new "feature", "bells and whistels", etc, consumes resources.

I remember hardware websites had articles on how to turn off unneeded Windows services through "services.msc".
When you turn unneeded services off, you free up system resources (RAM, some CPU cycles maybe) and your games run faster.

Maybe that's the case with Windows 10.
It's true, win10 has a ridiculous amount of crap pre-installed that folk don't need in an OS, like apps, that gem game wtf it's called & so on.
Downside with removing most things &/or turning them off, is that in the T&C/EULA/wtf micro$oft states it holds the right to re-install things & to turn things back on whenever they want, which defeats the point.


As for 8/16, that is 8 cores, 16 threads.
Grab any program like HWInfo or CPU-Z & it'll show you them.
Just like my processor, CPU-Z shows 4 cores, 8 threads.
Attachments:
cpuz.jpg (18 Kb)
avatar
fishbaits: It's true, win10 has a ridiculous amount of crap pre-installed that folk don't need in an OS, like apps, that gem game wtf it's called & so on.
Downside with removing most things &/or turning them off, is that in the T&C/EULA/wtf micro$oft states it holds the right to re-install things & to turn things back on whenever they want, which defeats the point.
Microsoft remotely can turn everything back on in Win10? This is ridiculous.
I heard that they spy on users, but they have no right to decide what service i need and what i don't. This OS is not free, jeez.
avatar
fishbaits: It's true, win10 has a ridiculous amount of crap pre-installed that folk don't need in an OS, like apps, that gem game wtf it's called & so on.
Downside with removing most things &/or turning them off, is that in the T&C/EULA/wtf micro$oft states it holds the right to re-install things & to turn things back on whenever they want, which defeats the point.
avatar
vsr: Microsoft remotely can turn everything back on in Win10? This is ridiculous.
I heard that they spy on users, but they have no right to decide what service i need and what i don't. This OS is not free, jeez.
Not free OS any more (although there's ways of getting it free still).
Yerp. Reddit went mental when it first happened.
Just like at times it'll uninstall things you have (happened to quite a few gog'gers & their games too) & not tell you it's done it for "security reasons" etc.
I have never been interested in buying AMD chips. And I won't in the future, either. So said, I hope Ryzen will be highly successful and force Intel to cut their fucking unbelievably high prices for Core i7 CPUs :-P
avatar
KingofGnG: I have never been interested in buying AMD chips. And I won't in the future, either. So said, I hope Ryzen will be highly successful and force Intel to cut their fucking unbelievably high prices for Core i7 CPUs :-P
I think the devil will ice skate to work long before intel will even contemplate lowering their prices ;p
Big, big letdown from AMD... Apparently the ECC will not be supported by Ryzen.

According to geizhals.at, Asus has pulled ECC support from all motherboards. Only Gigabyte, Biostar and Asrock support it.

However, I went over each motherboard to each vendor - and there is no ECC menu (ASRock) or even mention of ECC (Gigabyte, Biostar).

Then I encountered this: "... supports ECC, but will run it in non-ECC mode". Well, as someone using ECC I can state only one thing - ASRock is crap...:

There is no such thing as "Support ECC, but only in non-ECC mode". There are only two ECC sorts for DDR3/4, registered and unbuffered. They are not electrically compatible with each other! The "unbuffered ECC" is just a regular memory with an extra chip - its physically compatible with regular memory. Registered ECC is totally different and compatible only with itself.

So .. this claim is worthless. Running electrically compatible module without ECC function is called "no ECC support".

So what do we have: ASUS has pulled ECC support from everywhere, but the rest decided to market on clueless customers.


Looks like will be sitting on Xeon for long time for my gaming rig. What a shame!
avatar
Lin545: Big, big letdown from AMD... Apparently the ECC will not be supported by Ryzen.
AMD CEO claims ECC is supported by Ryzen. But I don't understand what's the deal with all mobo vendors pulling back... what a mess.

https://www.reddit.com/r/Amd/comments/5x4hxu/we_are_amd_creators_of_athlon_radeon_and_other/def58sv/
avatar
Lin545: ...
So what do we have: ASUS has pulled ECC support from everywhere, but the rest decided to market on clueless customers.
...
Out of curiosity, may I ask what is so important about ECC support? As far as I unterstand it, the error-correction is mainly important for much higher reliable system, such as servers, scientific workstations, navigation computers or maybe financial calculations. The trade-off for this reliability is higher costs and maybe even less speed.
I just do not really understand how I would benefit from ECC in an gaming rig or standard home computer.

What makes me more worried currently is that RAM seems to be an issue all again with the new AMD Ryzen. Single sided, double sided RAM and DIMM banks that can be used in which combination.
Looking at the certified memory list for the Gigabyte board I have chosen, makes me worrying that it would not be able to run at more than 2133 MHz instead of say 2666 MHz, let alone 3200 MHz which the board seems to support at max.

Nevertheless I think I will go with the setup posted the other day. Problem is the mainboard is out of stock at the moment and the other boards available do not really appeal to me. Also the CPU cooling is still a big question mark for me. :-\
Post edited March 06, 2017 by Quasebarth
avatar
Lin545: ...
So what do we have: ASUS has pulled ECC support from everywhere, but the rest decided to market on clueless customers.
...
avatar
Quasebarth: Out of curiosity, may I ask what is so important about ECC support? As far as I unterstand it, the error-correction is mainly important for much higher reliable system, such as servers, scientific workstations, navigation computers or maybe financial calculations. The trade-off for this reliability is higher costs and maybe even less speed.
I just do not really understand how I would benefit from ECC in an gaming rig or standard home computer.
Completely silent, unpredictable, and unchecked data corruption should be a concern for anyone with data that you can't just download off the net on demand. It could also result in unstability that is very difficult to diagnose. Memory is pretty much the *only* subsystem where we're still just assuming things work and not doing error correction. Pretty much everywhere else, a variety of error detection and correction schemes are in use. Because random bit flips *are* bad. There is no reason ECC shouldn't become mainstream, and for some reason it's dragged out for too long. The price shouldn't be a big issue. But right now it is made out to be, completely artificially, because there is not enough interest for ECC in the mainstream market.
avatar
Quasebarth: Out of curiosity, may I ask what is so important about ECC support? As far as I unterstand it, the error-correction is mainly important for much higher reliable system, such as servers, scientific workstations, navigation computers or maybe financial calculations. The trade-off for this reliability is higher costs and maybe even less speed.
I just do not really understand how I would benefit from ECC in an gaming rig or standard home computer.
No, no forget all this. Here is my practical experience, no theories and stuff:

ECC is simply ability to correct data corruption (and code=data) up to certain limit.
It can detect and correct 1,2,4 bits from 8 bits (byte). Without ECC the data will corrupt and travel further unnoticed.

Bit flips happen and will happen - different chances, different sources: from cosmic radiation... to faulty hardware or shielding. Sometimes it just can happen - for example on DDR4 the speeds become so fast, the error rate is too high, too probable, so they install CRC checks on write, to punch the probability withing "acceptable". Yet those CRC checks do not correct errors, but re-send the data - cheap solution to ECC at cost of loosing some performance (again, weighting of whats acceptable for every customer layer).

ECC is already installed in CPU in caches. But its still an "option" for RAM. The cost of ECC - one extra chip on memory bank, instead of 8 -> 9. instead of 16->18. Its cents. Of course market may overproduce/underproduce, but raw cost is this. Very affordable, but market behaves differently.

The speed - ECC brings no speed penalty. None. There is "registered" ECC ram type, now that has latency penalty, but not due to ECC, but due to "register" - a buffer that allows substantially higher bank capacities. But this latency is again not due to ECC logic per se. Unbuffered ECC does not have this.

Now about probability and damage extend.

- The probability for desktop machine is at least once per year, from one byte - to few bytes.
- And about the damage - it depends where it happens... great factor.

If it happens in the data, you may have damaged JPG,.. some useless in browser cache.. or some very important, which you never backed up. Your operating system might crash. Or it may just display some weird error and continue on. Its truely random.

What ends up is this: do you accept playing russian roulette over your whole address range in your machine, over your whole data once a year for, perhaps, a bit slightly higher cost of your memory banks? If you need additional, global, effects, see "Bitsquatting".

The motherboard manufacturers think - compared to your data is worthless, few cents in your wallet are more important. If you are other opinion - go for ECC. if you are not - use usually cheaper non-ECC. I have a file and media server that runs with 8GB of ECC DDR2 and $20 AMD CPU, and it works just fine,.. but Intel thinks that you should pay +$300-1000 premium for ECC. At least until hardware gets pulled from companies and is purchaseable used for small fraction of this price. Its intel nature. Until now AMD offered ECC virtually at no additional cost. That was a BIG difference, now gone.
Post edited March 06, 2017 by Lin545
avatar
Lin545: Now about probability and damage extend.

- The probability for desktop machine is at least once per year, from one byte - to few bytes.
To add to that, it is possible to have subtly faulty RAM that has a much higher frequency of random flips. Enough to be very likely to give you extremely hard to diagnose problems sooner or later, yet infrequent enough to make it easy to narrow down on the RAM as being the problem.

ECC corrections should get logged and that could save you from some significant headache.
Post edited March 06, 2017 by clarry
avatar
clarry: AMD CEO claims ECC is supported by Ryzen. But I don't understand what's the deal with all mobo vendors pulling back... what a mess.

https://www.reddit.com/r/Amd/comments/5x4hxu/we_are_amd_creators_of_athlon_radeon_and_other/def58sv/
Well, in a clash between Motherboard Handbook(s) - and CEO, I choose motherboard handbook(s). ;-)
avatar
clarry: To add to that, it is possible to have subtly faulty RAM that has a much higher frequency of random flips. Enough to be very likely to give you extremely hard to diagnose problems sooner or later, yet infrequent enough to make it easy to narrow down on the RAM as being the problem.

ECC corrections should get logged and that could save you from some significant headache.
Yes,yes.
Yes, those events are logged.
Problem is - workstation/server motherboard layouts/features are much weirder than desktop motherboards. Going to Supermicro just for ECC, I had to purchase PCI 3.3v sound card, saw-open the PCIe slot and there is not much PCIe extension slots.

Those pre-am4 ecc-supporting amd motherboards from asus were simply brilliant.
Post edited March 06, 2017 by Lin545