1366 x58 Xeon Enthusiast overclocks club

If anything, by showing those GN benches, you showed that a OC'ed 2600K is far superior to a OC'ed Nehalem in both mins and avg and max, where it counts.
I think you proved horrorshow's point.

I'm actually surprised the 930 at 4ghz is that close to the stock 2600k which typically sits at ~3.6ghz all core turbo depending on the board used. A decent Westmere can hit 4.4-4.6ghz which would close the gap significantly but then a good 2600k can hit 4.8-5ghz.

It really depends on the game and quality settings you use, but on my second rig which has a x5670 @ 4.4ghz and GTX 970, using adjusted settings it can run at 144fps+ 1080p with no issue in Quake Champions and Overwatch. Pretty much always GPU limited.
 
I'm actually surprised the 930 at 4ghz is that close to the stock 2600k which typically sits at ~3.6ghz all core turbo depending on the board used. A decent Westmere can hit 4.4-4.6ghz which would close the gap significantly but then a good 2600k can hit 4.8-5ghz.

A 930 doing a 4GHz OC from a 2.8GHz base clock, is a feat of engineering at 45nm I would call remarkable, and what enables Bloomfield to stick around till this day.
Sandy is also a marvel for how much it has in the tank once OC'ed. I use to think Yorkfield (C2Q9550) was awesome till I saw what a 920 could do.
 
A 930 doing a 4GHz OC from a 2.8GHz base clock, is a feat of engineering at 45nm I would call remarkable, and what enables Bloomfield to stick around till this day.
Sandy is also a marvel for how much it has in the tank once OC'ed. I use to think Yorkfield (C2Q9550) was awesome till I saw what a 920 could do.

Penryn (which includes Yorkfield of course) was also usable for quite a long time, but it's usefulness in games for anything but the most low end GPU's has passed.

I think Nehalem is nearing the end of it's useful life for gaming, especially with AVX becoming more prevalent (WMR for example requires AVX). But it still pairs pretty well with mid to low end GPU's

This is especially true as Ryzen gets cheaper, one can now pick up a R5 1600 with a (lower-end) motherboard for around $120 at Micro Center.
 
That's all I was saying lol
Yep, I wouldn't think it would make much sense to pair anything more powerful than a 2070/1080/V64 with Westmere at 4.2-4.6ghz.

I know just from testing versus my 1680v2 the x5670 does sit around 15% lower FPS at the same frequency when CPU limited (maxing out one or two threads).

Many games still have the issue where a few threads will be maxed out, and the rest have a lighter load which is why better IPC and and higher clocks will almost always win for high FPS gaming. This is mostly a non issue if you're only looking for 60fps in all but the most extreme cases.
 
maxing out one or two threads

If one is facing such a game, disable HT and 4 cores and Oc the rest. the 2 remaining cores will clock at least 2 multipliers above your previous best clock. Xeons have 12MB L3 cache, that will make up for some IPC if you are dealing with just 2 threads in a game.
 
If one is facing such a game, disable HT and 4 cores and Oc the rest. the 2 remaining cores will clock at least 2 multipliers above your previous best clock. Xeons have 12MB L3 cache, that will make up for some IPC if you are dealing with just 2 threads in a game.

That's not to say the other threads are idle and disabling HT will cause more harm than good in many cases, but many games have logic that simply isn't well multithreaded and those few threads will be what brings the FPS down compared to a modern chip (this is especially true of many DX11 and older titles).
 
Hiya, folks.

I just wanted to chime in as an X58 driver.

I, in the good old years, OC'd Intel and AMD CPUs. Even an ATI 9500 Pro, or two.

As far as I know, there is no way to OC Dell Precision T3500 CPUs.
 
Picked up a Xeon X5670 to replace my aging i7-920 and I'm stuck with getting the most out of this new chip. It's been a minute since I've overclocked an X58 system.

Fractal Design Define R6
Asus P6X58D Premium
X5670 CPU
G.SKILL Ripjaws F3-12800CL9D-8GBRL 4GB x6 for a total of 24GB
Corsair H150i AIO

I'm stuck at a stable 4.2ghz (24x175) but my memory and uncore aren't at optimal speeds. I'd like to be able to hit 200 bclk to run my RAM at 1600 but I'm unsure where to start and safe voltages to use.

Current settings attached in images.

IBT stable 10x at max memory but only getting about 44 gflops




Screen Shot 2019-03-11 at 7.38.51 PM.png Screen Shot 2019-03-11 at 7.38.18 PM.png
 
Hiya, folks.

I just wanted to chime in as an X58 driver.

I, in the good old years, OC'd Intel and AMD CPUs. Even an ATI 9500 Pro, or two.

As far as I know, there is no way to OC Dell Precision T3500 CPUs.

W-series, with software overclocks just fine on T3500/T5500.
 
Picked up a Xeon X5670 to replace my aging i7-920 and I'm stuck with getting the most out of this new chip. It's been a minute since I've overclocked an X58 system.

Fractal Design Define R6
Asus P6X58D Premium
X5670 CPU
G.SKILL Ripjaws F3-12800CL9D-8GBRL 4GB x6 for a total of 24GB
Corsair H150i AIO

I'm stuck at a stable 4.2ghz (24x175) but my memory and uncore aren't at optimal speeds. I'd like to be able to hit 200 bclk to run my RAM at 1600 but I'm unsure where to start and safe voltages to use.

Current settings attached in images.

IBT stable 10x at max memory but only getting about 44 gflops

How far can you push the uncore at your current 1.25v VTT? You should be OK up to 1.35v VTT but I typically try to stay under 1.3v.
 
W-series, with software overclocks just fine on T3500/T5500.
The T5500 board is Intel 5520 server level chipset and requires dual link CPU, even in single operation. None of which are unlocked. So no overclocking with T5500. Best we can do is X5687 quad core, or X5690 hex. Good thing is they do accept RDIMMs. Which are cheap right now. Popped a T5500 board in a T3500 box and populated it with 8gbx6=48gb factory refurb Samsung modules for under $66.
W3680 and W3690 only as they have unlocked multipliers. I run a 3680 at 4.0 on a T3500
Same. Humming right along at 4.0ghz. Some W3680 will do 4.3ghz in these T3500 but I did not get lucky. The W3680 is actually a 32nm Westmere. That and fact they are unlocked, were not widely known until fairly recently. No doubt why their value has risen so much while others have fallen.

[edit]
The W3570 and W3580 are also unlocked by the way. They have 1333mhz memory controller but are 45nm CPU and 8mb cache. Currently under $15 so may be a decent option for those on a budget.
 
Last edited:
setfsb isnt working on those dells ?
Good question. My experience is limited solely to the T3/5500. Have not tried setfsb personally. Someone please correct me if wrong. It is my understanding that setfsb does not like the PLL chip on these Dells. Slight changes results in crashes when it throws the SATA bus out of whack. So we are limited to unlocked CPUs on these X58 based T3500s.
 
Last edited:
The T5500 board is Intel 5520 server level chipset and requires dual link CPU, even in single operation. None of which are unlocked. So no overclocking with T5500. Best we can do is X5687 quad core, or X5690 hex. Good thing is they do accept RDIMMs. Which are cheap right now. Popped a T5500 board in a T3500 box and populated it with 8gbx6=48gb factory refurb Samsung modules for under $66.Same. Humming right along at 4.0ghz. Some W3680 will do 4.2ghz in these T3500 but I did not get lucky. The W3680 is actually a 32nm Westmere. That and fact they are unlocked, were not widely known until fairly recently. No doubt why their value has risen so much while others have fallen.

[edit]
The W3570 and W3580 are also unlocked by the way. They have 1333mhz memory controller but are 45nm CPU and 8mb cache. Currently under $15 so may be a decent option for those on a budget.

Clockgen is available on the T5500 motherboard, but I was unaware of the performance series Xeon limitation which puts a damper on any software overclocking indeed -- as mine came with an E5645, which I figured was the workstation series (but it's apparently just rackmount X series) that I chucked and replaced with an X5675 (because I've got a butt ton of them from trying to find a golden sample...) My T5500 doesn't even have the extra riser in it -- as it's just transcoding up to three sessions on Plex so it doesn't need anything else. I was actually considering replacing it with a W3690 at one point, but the metrics reported from that system have never indicated I needed more CPU power. Glad to know now that it won't post... Thank you.

I'm also sad to see T3500 costs so much these days on eBay, I think I got all of mine for $45-$60 shipped... They're nice little machines to set up in recording studios as long as you enable the PCI-E external throughput optimized mode (otherwise you'll get audio issues, on a Firewire or USB3 card), or as staging/compiler boxes.
 
Penryn (which includes Yorkfield of course) was also usable for quite a long time, but it's usefulness in games for anything but the most low end GPU's has passed.

I think Nehalem is nearing the end of it's useful life for gaming, especially with AVX becoming more prevalent (WMR for example requires AVX). But it still pairs pretty well with mid to low end GPU's

This is especially true as Ryzen gets cheaper, one can now pick up a R5 1600 with a (lower-end) motherboard for around $120 at Micro Center.

What is this MVR and AVX everyone keeps talking about?

If one is facing such a game, disable HT and 4 cores and Oc the rest. the 2 remaining cores will clock at least 2 multipliers above your previous best clock. Xeons have 12MB L3 cache, that will make up for some IPC if you are dealing with just 2 threads in a game.
That's not to say the other threads are idle and disabling HT will cause more harm than good in many cases, but many games have logic that simply isn't well multithreaded and those few threads will be what brings the FPS down compared to a modern chip (this is especially true of many DX11 and older titles).

Is that really how it works? For pre 2016 games, just disable cores and HT and OC to 4.6ghz?

Picked up a Xeon X5670 to replace my aging i7-920 and I'm stuck with getting the most out of this new chip. It's been a minute since I've overclocked an X58 system.

Fractal Design Define R6
Asus P6X58D Premium


View attachment 147529 View attachment 147530

Any bumps in gaming performance? What resolution do you game at?
 
Should not need to. It could flag some software that is tied to hardware combination for licensing though.
 
Was able to find the sweet spot for my setup.

200 bclk
20 multi
1.35v vcore
1.35 vtt
1600mhz RAM
3600 uncore

All other voltages at stock. I could probably try lowering the vcore and vtt next but why mess with a good thing.

Temps at idle are about 30-35c and full load while running prime95 peaks at 76c

Screen Shot 2019-03-14 at 1.06.37 PM.pngScreen Shot 2019-03-14 at 1.11.05 PM.png

Picked up a Xeon X5670 to replace my aging i7-920 and I'm stuck with getting the most out of this new chip. It's been a minute since I've overclocked an X58 system.

Fractal Design Define R6
Asus P6X58D Premium
X5670 CPU
G.SKILL Ripjaws F3-12800CL9D-8GBRL 4GB x6 for a total of 24GB
Corsair H150i AIO

I'm stuck at a stable 4.2ghz (24x175) but my memory and uncore aren't at optimal speeds. I'd like to be able to hit 200 bclk to run my RAM at 1600 but I'm unsure where to start and safe voltages to use.

Current settings attached in images.

IBT stable 10x at max memory but only getting about 44 gflops




View attachment 147529 View attachment 147530
 
Last edited:
lowering the vcore and vtt

You probably can run at a lower vcore.

But 1.35 VTT is about as low as you can go with 6 sticks and 24Gb RAM.

Lower multiplier, higher BCLK is another strategy. On one hand you squeeze your memory , on the other hand you put your cpu at a lower clock and more stress on the north bridge.

i go the other way:
put my memory as slow as its stable
enable speedstep
go as high multi as possible
increase bclk until max core speed
the slowly try to increase memory. there will be gaps on memory tuning, some straps and timings will not work, while a faster setting will.

and please, show us how your xeon + 1060 behaves at 1080p. Do you feel CPU bound at 1080p?
 
Lower multiplier, higher BCLK is another strategy.

I started the overclock going that route but couldn't get my memory at a decent speed. The 5670 allows use of the 24x turbo multiplier on all cores so 24x175 was also stable for a 4.2 overclock but memory was only running at 1403 and I couldn't get the QPI speed up higher than ~3000 without running into stability issues.

and please, show us how your xeon + 1060 behaves at 1080p. Do you feel CPU bound at 1080p?

Not running a 1060 or gaming. This is a Plex server that does video transcoding which is why I switched from an i7 920 to a hexacore xeon.
 
3 days stable. When I get some time I will try lowering the vcore a bit (or go back to offset voltage) and see if I can get it to run a bit cooler but it's been performing well!

Screen Shot 2019-03-17 at 7.32.24 PM.png
 
just received a 1080ti.

What games do you suggest for a test if a couple of Xeons x5675 can CPU-limit a 1080ti at 1080p?

My monitor can do 1080p 240hz. i am willing to try up to 3 games with the 970 before installing the 1080ti.

edit: my suggestions

far cry primal
skyrim enhanced edition

2 old games that i have installed.

"newer" games that i have:

Fallout 4
Witcher 3
Kingdom come deliverance
Street Fighter V
subnautica
no man's sky
 
Last edited:
Hmmmm....

Well, I know Destiny 2 will hit a bottleneck on Nehalem. (it's why I upgraded)

Assassins Creed: Odyssey would be a fun one because of all your extra cores.

Fallout 4 should bottleneck once you have shadow distancing cranked.

I'm certainly curious to see how it would handle the Division 2 benchmark ;)

Congrats on your purchase! (I just RMAed my 980 Ti and got a new, sealed 1080 back!)
 
upload_2019-3-18_22-38-35.png


For comparison, Core i7 5960X (Haswell-E) @ 4.3 GHz on all eight cores running DDR4 at almost twice my speed.

index.png


My very wild guess is that somewhere around 1070ti performance this games starts to become CPU limited. I am running 25x160=4GHz to keep a fair comparison with X5650 owners. Even running at 4.3Ghz the best i can hope for is 103 fps average at 1080p
 
all depends on resolution

5760x1080 1080ti I don't believe I have seen the CPU limit yet, granted, I don't have the latest games on
 
yes you need ,your gpu should be 100% in all games and your cpu much less,what is your cpu and memory usage when playing games?

check this post

with lot of games on ultra high 2560x1920 i get 70-80% cpu and 100% gpu ,with battlefield v i get 90-95 cpu ,i am near the limits even with a 6 core cpu.

a 15$ buy that will worth it

*edit*
something of all 1366 users that i noticed,i use my pc ONLY for gaming ,no internet nothing else and noticed spectre patches made a bit slower the pc,solution is the program below to disable those microsoft patches

https://www.grc.com/inspectre.htm

and found somewhere that this dll is also causing slower pc ,so i deleted it.
C:\Windows\System32\mcupdate_genuineintel.dll

so far all works fine

slow down noticed here as well :


Retpoline restores most of the lost performance.
 
But 1.35 VTT is about as low as you can go with 6 sticks and 24Gb RAM.


nonsense
i have 48GB's across 6 sticks at 1.25v on a x5690 (21x200 @4.2 / DDR3 1600)
and 24GB's across 6 sticks at 1.26v(fluctuates to 1.28v) on a 5670 (19x200 @ 3.8 / DDR3 1600
Untitled.png

either you have dud xeons or don't know what you're doing.
 
Last edited:
nonsense
i have 48GB's across 6 sticks at 1.25v on a x5690 (21x200 @4.2 / DDR3 1600)
and 24GB's across 6 sticks at 1.26v on a 5670 (19x200 @ 3.8 / DDR3 1600)

either you have dud xeons or don't know what you're doing.

or i don't do low multiplier high fsb because i tested it and for what i do, higher multi lower fsb works better, but requires more VTT on my mobo...
 
Back
Top