What controls Turbo Core in Xeons?

etherealfocus

Senior member
Jun 2, 2009
488
13
81
Is it just Tmax (ie chip stays at max turbo as long as CPU isn't too hot) or is it also measuring power draw or other variables?

Asking because I'm pondering building a workstation with one of those cheap v2/v3/v4 Xeons on ebay. Would be great if I can stay near max turbo just by slapping on a good hsf, but it's less appealing if they're gonna drop to base ~2.0ghz clocks every time I throw a sustained workload at them.

Goal with this machine is smoother multitasking. The laptop in my sig can do most of what I need but it doesn't stay responsive while recalculating big Excel sheets, batch processing images, etc. Would be nice to have a bunch of cores so it doesn't choke on that sort of lightly threaded stuff.

Also wanna do some experimenting with ESXi which apparently is pretty core hungry.
 

RichUK

Lifer
Feb 14, 2005
10,320
672
126
You can overide the majority of the limiting turbo parameters in the BIOS on most reasonable to decent motherboards.
 
  • Like
Reactions: Drazick

etherealfocus

Senior member
Jun 2, 2009
488
13
81
So in essence I could run it at max turbo 24/7 with a couple bios tweaks? I'd still want it to clock down when not in use of course; just be able to maintain max clocks under sustained load.

Any suggestion on a good hsf for maintaining that turbo? I'm looking at a Noctua i4 right now. Would prefer to avoid the hassle of water cooling.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
At least for >= E5 v3 Xeons it is the number of utilized cores. There is no way to override that behavior, i.e "Multicore Enhancement" doesn't make any difference.
 
  • Like
Reactions: kritisharma

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
So in essence I could run it at max turbo 24/7 with a couple bios tweaks? I'd still want it to clock down when not in use of course; just be able to maintain max clocks under sustained load.

Any suggestion on a good hsf for maintaining that turbo? I'm looking at a Noctua i4 right now. Would prefer to avoid the hassle of water cooling.
I have a Z97 board and it will run my 1231-V3 at 3.8ghz on all cores with no problem.
 

Bouowmx

Golden Member
Nov 13, 2016
1,138
550
146
I have a Z97 board and it will run my 1231-V3 at 3.8ghz on all cores with no problem.

How you do that? Running at maximum Turbo frequency regardless of number of active cores is a nice touch. I guess the Z chipset?

For reference: Intel Xeon E3-1231 v3 Turbo frequencies for 1 to 4 active cores (GHz): 3.8 3.8 3.7 3.6
 

coercitiv

Diamond Member
Jan 24, 2014
6,137
11,616
136
I'd still want it to clock down when not in use of course; just be able to maintain max clocks under sustained load.
The biggest gains in cutting power come from enabling deep C states. In fact, after enabling C states clocking down the chip won't make any discernible difference. (any option will do the trick though, clocking down achieves a similar effect)

Any suggestion on a good hsf for maintaining that turbo? I'm looking at a Noctua i4 right now. Would prefer to avoid the hassle of water cooling.
You can use the TDP guide - http://noctua.at/en/tdp-guide - for reference, even if it does not contain the cooler you are looking at. The NH-U9S looks and likely performs quite similar, and Noctua recommends max 140W TDP for this model. Considering you are not going to overclock, seems to me like the Noctua i4 or any other cooler similar in size & quality will be quite adequate for the job even at decent sound levels. (low noise config)
 
  • Like
Reactions: Burpo and Drazick

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
How you do that? Running at maximum Turbo frequency regardless of number of active cores is a nice touch. I guess the Z chipset?

For reference: Intel Xeon E3-1231 v3 Turbo frequencies for 1 to 4 active cores (GHz): 3.8 3.8 3.7 3.6
Yes, it's because the Z chipset allows overclocking. You can set the multiplier to 38. It runs all the cores at 3.8 when running IBT. Doesn't get too hot, either.
You can also run a 4790S chip at 4.0 this way with a Z board.
There's a thread around here somewhere about doing this.

Here is one of my 1231-V3 posts:
https://forums.anandtech.com/threads/overclocking-on-a-i7-4790s.2447253/#post-37692772
 

lopri

Elite Member
Jul 27, 2002
13,209
594
126
You kind of have to look at how each chip behaves under load in the documentation. Check the links PG provided above. The max turbo clock is achieved only when one or two core are under active load. There are hard ceilings as to how high the turbo freq is going to be per CPU load, no matter what the temperature and voltage are. But of course they can have negative impact on max freq.
 

etherealfocus

Senior member
Jun 2, 2009
488
13
81
Unfortunately the Xeon E3 chips are still floating above $250, which kinda kills the value of buying old stuff. Having a big pile of cores is also really appealing... I get impatient faster than is probably reasonable sometimes. :)

I think 8 cores is probably plenty, but if I can nab a 14-core of course that'd be sweet... unfortunately it looks like the word got out on those and cheapest ones with retail stepping I'm seeing on ebay are over $500... boo, I'll keep digging I guess.
 

PG

Diamond Member
Oct 25, 1999
3,426
44
91
You need to research and check the forums I posted above, especially the thread on ES Xeons.
The great deals are E5 V3 or V4. Asrock or MSI X99 boards work best with these ES chips. Avoid Asus and Gigabyte. Their bioses are flaky with older stepping ES cpus.
I picked up a 10 core E5 v4 with hyperthreading for $165 recently, as in a few weeks ago. Oh, and it's ES, but the same stepping as retail so it should actually work in any board with a bios update. Runs great in my open box Asrock X99 from Microcenter. Good deals like this are possible if you research and then check ebay a lot, plus watch the forum like I said. Sometimes they catch and post good deals others won't notice.
 
  • Like
Reactions: etherealfocus

Dufus

Senior member
Sep 20, 2010
675
119
101
Is it just Tmax (ie chip stays at max turbo as long as CPU isn't too hot) or is it also measuring power draw or other variables?

Yes, power and current have an effect too. Ultimately controlled by Intel, manufacturers then us.

At least for >= E5 v3 Xeons it is the number of utilized cores. There is no way to override that behavior, i.e "Multicore Enhancement" doesn't make any difference.

Just guessing but I would have thought that would be possible with at least Haswell chips. I did pick up a used retail E5-2683V3 (just the one) but have to admit it's a bit of a learning curve from the usual i7. Would be nice to OC it by 50% (a bit slow with single threaded apps) but looks like that's not going to happen.
 

etherealfocus

Senior member
Jun 2, 2009
488
13
81
Thanks for all the info guys. Couple questions:

1. There's a lot of talk of overclocking here. I was under the impression that Xeons aren't unlocked; all you can do is make them run at max turbo all the time... and possibly force max turbo on all cores simultaneously. Correct?

2. How's something like this look? http://www.ebay.com/itm/Intel-Xeon-...443543?hash=item51ec1979d7:g:U0YAAOSwOyJX6IQs

That wouldn't be my first choice due to low per-core clocks, but would certainly get the job done. What do you guys make of it claiming to be stepping 0? And of this note in this listing? "Special Notice:This processor is currently only supported by ASUS, Supermicro,Intel C612 Chipset motherboard.X99 Motherboard will require a Bios update to support it (Newest Bios will be released in the short coming future)"

Avoid?
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Just guessing but I would have thought that would be possible with at least Haswell chips. I did pick up a used retail E5-2683V3 (just the one) but have to admit it's a bit of a learning curve from the usual i7. Would be nice to OC it by 50% (a bit slow with single threaded apps) but looks like that's not going to happen.

E5-2683V3 will run at:

1 cores = 3.0GHz
2 cores = 3.0GHz
3 cores = 2.8GHz
4 cores = 2.7GHz
5 cores = 2.6GHz
6 cores = 2.5GHz
7 - 14 cores = 2.5GHz

There is nothing you can do about that. The only way to OC is to use the BCLK (which isn't usefull).
 
  • Like
Reactions: Drazick

Dufus

Senior member
Sep 20, 2010
675
119
101
E5-2683V3 will run at:

1 cores = 3.0GHz
2 cores = 3.0GHz
3 cores = 2.8GHz
4 cores = 2.7GHz
5 cores = 2.6GHz
6 cores = 2.5GHz
7 - 14 cores = 2.5GHz

There is nothing you can do about that. The only way to OC is to use the BCLK (which isn't usefull).

E5-2683V3 running 30x all cores.
o6w74i.png


It is a retail CPU, perhaps the ES and QS samples that a lot of people seem to be using are a bit flunky, idk.

http://valid.x86.fr/9shu3g There's a CPU bench score there too.
 

LTC8K6

Lifer
Mar 10, 2004
28,520
1,575
126
My experience with Haswell chips has been that if your board allows overclocking, you can get a locked CPU to run all the cores at the turbo multiplier under load.
 
  • Like
Reactions: etherealfocus

Dufus

Senior member
Sep 20, 2010
675
119
101
@etherealfocus There are unlocked Xeons from what I've read but I'm a noob with Xeon so don't have enough experience to advise. All I can say is personally it would have been nice to see some higher clocks for single threaded apps, I also have a feeling this HSW Xeon I have could support TB3 but probably not much chance of that being implemented.

My experience with Haswell chips has been that if your board allows overclocking, you can get a locked CPU to run all the cores at the turbo multiplier under load.

One does need firmware support and this is perhaps why it's not seen as a possibility by many.

vdoj9h.png
 
  • Like
Reactions: etherealfocus

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
E5-2683V3 running 30x all cores.
o6w74i.png


It is a retail CPU, perhaps the ES and QS samples that a lot of people seem to be using are a bit flunky, idk.

http://valid.x86.fr/9shu3g There's a CPU bench score there too.

That score (21307) doesn't appear to be anything close what it should be when all cores are at 3000MHz+.
14/28T scores 18900 at 2.3GHz.
 
  • Like
Reactions: Drazick

Dufus

Senior member
Sep 20, 2010
675
119
101
14/28T scores 18900 at 2.3GHz.

Where did you get that from?

Benchmarks recorded from the validated CPU-Z E5-2683 v3 Top 15 List

Submitted by DESKTOP-P6DET7C - 2016-10-19 06:48:29 1297 / 17598 BCLK 105MHz 28 threads
Submitted by BH-201611111245 - 2016-12-09 16:06:16 1336 / 17612 BCLK 105.5MHz 28 threads
Submitted by Anonymous - 2016-11-27 12:59:10 1176 / 17323 BCLK 105MHz 28 threads
Submitted by Anonymous - 2016-12-11 21:15:08 1266 / 16668 BCLK 100MHz 28 threads
Submitted by me, ucode - 2017-01-10 13:29:20 1424 / 21307 BCLK 105MHz 28 threads

So looks like it ties in according to those, about 20% higher in multi-threaded (28T). Also Single thread score is a little higher probably because those other results are relying on no more than 2 cores being active while mine doesn't.

I guess it doesn't matter anyway.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Where did you get that from?

Benchmarks recorded from the validated CPU-Z E5-2683 v3 Top 15 List

Submitted by DESKTOP-P6DET7C - 2016-10-19 06:48:29 1297 / 17598 BCLK 105MHz 28 threads
Submitted by BH-201611111245 - 2016-12-09 16:06:16 1336 / 17612 BCLK 105.5MHz 28 threads
Submitted by Anonymous - 2016-11-27 12:59:10 1176 / 17323 BCLK 105MHz 28 threads
Submitted by Anonymous - 2016-12-11 21:15:08 1266 / 16668 BCLK 100MHz 28 threads
Submitted by me, ucode - 2017-01-10 13:29:20 1424 / 21307 BCLK 105MHz 28 threads

So looks like it ties in according to those, about 20% higher in multi-threaded (28T). Also Single thread score is a little higher probably because those other results are relying on no more than 2 cores being active while mine doesn't.

I guess it doesn't matter anyway.

Tested myself on a 2699.
 
  • Like
Reactions: Drazick

Dufus

Senior member
Sep 20, 2010
675
119
101
14/28T scores 18900 at 2.3GHz.
Tested myself on a 2699.

Ah I see. Then I think maybe something went wrong with the way it was tested.

If we adjust your score for a normal E5-2699v3 with 36 threads and all turbo of 2.8GHz then we get :-
Score multiplied by clock difference multiplied by thread difference : 18900 * (2.8 / 2.3) * (36 / 28) = 29582

If we look at a validated score for the E5-2699v3 we see at 2.8GHz 36 Threads a score of only 24431. It doesn't tie in with your 2699. If you do a full run of all threads @ 2.8GHz what do you get?

If we adjust that 24431 for 3.15GHz with 28 threads then we get :-
Score multiplied by clock difference multiplied by thread difference : 24431 * (3.15 / 2.8) * (28 / 36) = 21377

Now that's very close to my score.

Not sure why the difference with you own score perhaps you selected 28 threads from the CPU-Z bench window and got thread imbalance. That is it wasn't run on 14 cores with 2 threads on each core but 18 cores with 10 cores using 2 threads and 8 cores using one thread. Only guessing though.

BTW that's a monster CPU you have, wouldn't mind one myself but over my budget :(
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Ah I see. Then I think maybe something went wrong with the way it was tested.

If we adjust your score for a normal E5-2699v3 with 36 threads and all turbo of 2.8GHz then we get :-
Score multiplied by clock difference multiplied by thread difference : 18900 * (2.8 / 2.3) * (36 / 28) = 29582

If we look at a validated score for the E5-2699v3 we see at 2.8GHz 36 Threads a score of only 24431. It doesn't tie in with your 2699. If you do a full run of all threads @ 2.8GHz what do you get?

If we adjust that 24431 for 3.15GHz with 28 threads then we get :-
Score multiplied by clock difference multiplied by thread difference : 24431 * (3.15 / 2.8) * (28 / 36) = 21377

Now that's very close to my score.

Not sure why the difference with you own score perhaps you selected 28 threads from the CPU-Z bench window and got thread imbalance. That is it wasn't run on 14 cores with 2 threads on each core but 18 cores with 10 cores using 2 threads and 8 cores using one thread. Only guessing though.

BTW that's a monster CPU you have, wouldn't mind one myself but over my budget :(

This is interesting.
Indeed, setting the thread count from CPU-Z wasn't working. The real MT score with 14C/28T (set from bios) @ 2.3GHz is just 15660.
This is the first time I've seen any indication that MCE would be working on E5-26xx v3 Xeons. Either it is 2683 v3 or ASRock specific, but the MCE on ASUS X99-M WS/SE + 2699 v3 doesn't do anything. None of the configurations allow running it beyond the standard turbo bins.
If it would be certain that either the model in question (2683 v3) or the motherboard can do this, I would swap my system to such config immediately. I would literally kill for few hundred more MHz, since my primary workload doesn't scale beyond 16 threads.
 
  • Like
Reactions: Drazick

Dufus

Senior member
Sep 20, 2010
675
119
101
It does need intervention (BIOS Mod usually).

MCE is just a fancy marketing name. Basically Intel produced Haswell CPU's with an bug / errata which allows full turbo on all cores. This is fixed up with a microcode update. In essence apply full turbo before the microcode patch takes place (one of the microcode versions that fixes this particular errata) and you will probably find your 2699 running 3.6GHz across all cores providing the extra power draw can be handled.