I7-3820 to E5-2670

[M76]

Is that a worthwhile upgrade?

I was looking at passmark data and it's strange. Single threaded it's 1600 vs 1900 for the I7, but multi-threaded it's only 8000 vs 12000 for the xeon. With twice as many cores I'd have expected a bigger difference. My CPU is oced to 4660 so the difference would be even less for me. I'm not sure if the upgrade would be worth it even if I use all 16 threads. It's not about the price as they go for $60-80 on ebay. It's more about is it worth the hassle.

 

[arestavo]

If you didn't over clock your original CPU it would be an upgrade. Since you over clock it isn't going to be an upgrade for gaming.

If you use pro software it will help, but again for games it will actually hurt you a little with reduced minimum FPS and more FPS dips.

No, the 2XXX series (or 4XXX or 8XXX series) Xeons cannot be overclocked. Well, not more than a tiny bit with Bclk.

The E5-16XX series can be overclocked. And if your board supports it, there is a 6 core or an 8 core version that you can get and overclock just like an i7 K series. I've got an E5-1650 V2 Ivybridge in my X79 board overclocked to 4.5GHz. It runs much cooler than my old 3930K at 4.3GHz.

1650 /1660 V2 are 6 core.
1680 V2 is 8 core.

 

[M76]

Yeah I know they can't be overclocked. The E5-16xx series on the other hand can't be bought for cheap. For the price of an 1650 v2, or less I could get a 4930K locally. But I don't want to spend that much on a slight upgrade.

Thanks for the reply.

 

[M76]

Depends on whether you need a V1 or a V2 series. V1 is maybe around $150 for a E5 1650 and around $200 for a E5 1660. V2 prices will be much more than that.

It's not really a need, it's more of a want. The V1 has the same architecture as my current CPU, so there'd be no additional benefit apart from the two extra cores, and if it can't be overclocked that well I might end up where I started. I don't know how these xeons average on overclock. anyway with postage fees and tolls my price will probably end up around 220 even for a V1. And I can get a second hand 4930K for 270 locally. So it seems like a better deal. But I'm not even sold on the whole upgrade deal yet.

 

[Starbomba]

In my opinion, it's a tie and sometimes even a small downgrade if you want to play games, and only play games. You'd be trading 4 fast cores for 8 slow cores. Hell, even at stock speeds, the 3820 without boost is 300 MHz faster than the max boost of the 2670 on 2 cores (8 core boost is only up to 3 GHz). Most games still cannot utilize properly more than 4-8 threads, so the rest would be ignored.

However, if you do plan to render, run VMs, BOINC/Fold, or do any other CPU-intensive task, the 2670 is very much worth it, especially for under $70.

 

[cdabc123]

on passmark i get 1800 single and 12000 multi however unless you are doing something that can use 16 threads its not worth it

 

[M76]

In my opinion, it's a tie and sometimes even a small downgrade if you want to play games, and only play games. You'd be trading 4 fast cores for 8 slow cores. Hell, even at stock speeds, the 3820 without boost is 300 MHz faster than the max boost of the 2670 on 2 cores (8 core boost is only up to 3 GHz). Most games still cannot utilize properly more than 4-8 threads, so the rest would be ignored.

However, if you do plan to render, run VMs, BOINC/Fold, or do any other CPU-intensive task, the 2670 is very much worth it, especially for under $70.

The thing is that I want to do both. Render and game at the same time. Well I mean not at the same time, but on the same computer.

 

[M76]

So I pulled the trigger on a 2670 today. If nothing else for the kicks of it, to see what it can do. I haven't touched my PC in years, so I was itching for some tinkering for quite some time now.

I'd like to compare performance of these two with some benchmarks, any ideas/suggestions on what tests to run apart from the obvious like passmark and cinebench. I'll probably run 3DMark too, even if it's not considered [H] anymore.

 

[caltech31]

The thing is that I want to do both. Render and game at the same time. Well I mean not at the same time, but on the same computer.

What do you currently score in Cinibench R15?

I know that the 2670 "only" pulls in about 1050points, so that should give you a frame of reference for encoding performance. Single threaded performance is I think 120pts on the 267y0.

 

[M76]

I get 792 at 4625Mhz and 160 for single threaded.

 

[caltech31]

I get 792 at 4625Mhz and 160 for single threaded.

Yeah, that's a tough trade off - about 30% increase in multi threaded performance, but a 30% decrease in single threaded performance.

All things considered though, a 2670@3.3Ghz (I think that's what it turbo boosts to) will be more than enough for whatever games you play when paired with a 980ti. Outside of benchmarks, you probably won't notice a thing.

For what it is worth, I have seen people get overclock their xeons to 3.5ghz via BCLK. Not a huge jump, but every little bit helps.

 

[M76]

The cpu arrived today. After quickly installing it, it does 1002 in in cinebench and 108 in single thread. That's more like a 50% decrease in single thread performance. But raid rebuilding is going in the background that might interfere with the test slightly. I'll post benchmark comparisons later when I can do them properly. And I'll try the bclk to see if it goes anywhere.

 

[Beaflag VonRathburg]

I can understand your desire to tinker, but that's pretty much what was expected. For 99% of users your 3820 at 4.5 was the way better choice.

 

[cageymaru]

I can understand your desire to tinker, but that's pretty much what was expected. For 99% of users your 3820 at 4.5 was the way better choice.

Hobbyists need cheap things to experiment on. $60 just to have a reason to poke around with his system is an awesome investment. That's the equivalent of a movie date with food.

M76 Let us know how your OC goes.

 

[M76]

I can understand your desire to tinker, but that's pretty much what was expected. For 99% of users your 3820 at 4.5 was the way better choice.

Thankfully 99% of people is not me. As i mentioned I use my computer for rendering as much as I use it for gaming. Yesterday I played with XCOM2 and if I didn't know about the CPU change I would not have noticed it. At least it runs cool. My 3820 @ 4625 required 1.48V to be stable. And it ran in the high 80s under load. Now I get to 60 tops. Of course this is still stock voltage and clock.

 

[Matthew Kane]

My 3820 @ 4625 required 1.48V to be stable.

I hope that 1.48v is a typo, either you didn't have the right settings or (though unlikely) have a really really bad chip.

My last 3 3820's did 4.625 (just a blck bump) at 1.32 - 1.3625v range no problems IBT, OCCT 3.1 and Prime stable for 20 hrs. Cooled by my Noctua D-14, temps never went past mid 70's under load.

 

[M76]

I hope that 1.48v is a typo, either you didn't have the right settings or (though unlikely) have a really really bad chip.

My last 3 3820's did 4.625 (just a blck bump) at 1.32 - 1.3625v range no problems IBT, OCCT 3.1 and Prime stable for 20 hrs. Cooled by my Noctua D-14, temps never went past mid 70's under load.

No, it's not a typo. It needed 1.40-42 for 4.5 already. I read that below 1.5 it's still safe for Sandy Bridge e-s. And since I've been running it at either 4.5 or 4.625 for the last 3 years it seems to be true. I've had a D-14 too, but it couldn't handle 4.625, only 4.5. I could only g to 4.625 after switching to a CM Nepton 140XL.

I've tried every trick in the book, this chip needs a ton of voltage to be stable. If lowered it just to 1.47 I already got slight instabilities. Altough prime95 didn't report errors, I'd get lots of bluscreens and applications crashing to the desktop with less voltage.

Edit I've been running it for so long at 4.625 that I don't remember exactly. It might have been 1.38V at 4.5.

 

[M76]

But back on topic. The 2670 is obviously not made for overclocking. The BCLK can go to 106, anything beyond that is a no go. Between 106 and 107 the system becomes highly unstable, it doesn't even get into windows. and it randomly wouldn't boot the video bios. Anything above 107 it outright refuses to start the post procedure. My MB has a diagnostic display to show what part is it at in the post. But it won't even start above 107. If I try to force the 33 max turbo multipler for all cores it goes into a perpetual reboot sequence. So that's a no go either. If I disable speedstep and set it to 33 for all cores. The bios reports it's running at 3.5, but in reality it is locked to 30x which is only 3180. It's better than 2600 obviously, but in single thread it would not be faster than on default clocks. So the best thing I could do is leave it on default turbo settings, so it can turbo properly and up the bclk to 106. Then I get 6% increase in performance across the board.

The games I've tried so far I couldn't detect the difference with my own senses, so that's good enough for me as far as gaming goes. I'll try to post benchmark results tomorrow. That darned Raid rebuild is still ongoing.

 

[caltech31]

But back on topic. The 2670 is obviously not made for overclocking. The BCLK can go to 106, anything beyond that is a no go. Between 106 and 107 the system becomes highly unstable, it doesn't even get into windows. and it randomly wouldn't boot the video bios. Anything above 107 it outright refuses to start the post procedure. My MB has a diagnostic display to show what part is it at in the post. But it won't even start above 107. If I try to force the 33 max turbo multipler for all cores it goes into a perpetual reboot sequence. So that's a no go either. If I disable speedstep and set it to 33 for all cores. The bios reports it's running at 3.5, but in reality it is locked to 30x which is only 3180. It's better than 2600 obviously, but in single thread it would not be faster than on default clocks. So the best thing I could do is leave it on default turbo settings, so it can turbo properly and up the bclk to 106. Then I get 6% increase in performance across the board.

The games I've tried so far I couldn't detect the difference with my own senses, so that's good enough for me as far as gaming goes. I'll try to post benchmark results tomorrow. That darned Raid rebuild is still ongoing.

The raid rebuild is probably knocking at least 15pts off of your single threaded score. Cinibench is super sensitive to background tasks - as an example, if I have Steam and Utorrent going, my scores drop by about 50pts in the multi threaded bench.

It's weird that you can't force the Turbo multiplier - have you updated to the latest bios for the AS Rock Board? Even those dual CPU server boards are capable of locking the turbo multiplier (that I have seen anyways)

 

[thesmokingman]

I hope that 1.48v is a typo, either you didn't have the right settings or (though unlikely) have a really really bad chip.

My last 3 3820's did 4.625 (just a blck bump) at 1.32 - 1.3625v range no problems IBT, OCCT 3.1 and Prime stable for 20 hrs. Cooled by my Noctua D-14, temps never went past mid 70's under load.

I concur, my 3820s needed very little voltage for 4.6. To get to 5ghz it was around 1.5v and a lil change under water.

 

[M76]

The raid rebuild is probably knocking at least 15pts off of your single threaded score. Cinibench is super sensitive to background tasks - as an example, if I have Steam and Utorrent going, my scores drop by about 50pts in the multi threaded bench.

It's weird that you can't force the Turbo multiplier - have you updated to the latest bios for the AS Rock Board? Even those dual CPU server boards are capable of locking the turbo multiplier (that I have seen anyways)

It is the latest bios.

 

[thesmokingman]

M76, what are your settings for your oc, blk etc?

I dug out my old 3820 cheat sheet. I have owned more than a few 3820s and clocked them all to roughly this sheets specs. I sold or put them into machines then sold those eventually.

i7 3820 - 4820 Overclock Club - Page 87

After going through the RIVE with a fine tooth comb, I settled on these outcomes. I didn't bother to take pics, just the raw info. Maybe it will help some of you. These voltages were consistent with the two 3820's that I had tested.

4.3 = 1.22v
4.6 = 1.3v
4.75=1.345v
4.87=1.4v
5.0=1.45v
5.1=1.51v


I still need to dial the 5.1ghz clock some more, but I really don't want to spend too much time fussing around at 1.5v+. Temps however are excellent, even at 1.5v+ the chip doesn't break 73c. My 2600k at 5.2ghz@1.52v would hit 80c in contrast.

 

[M76]

BCLK 125 x36 for 4.5, x37 for 4.625, and x38 for 4.75 I didn't touch anything else between the settings. I used the same guide linked in the page you cited that from.

Actually this board handles BCLK very uniquely so if I set BCLK to 125 it will automatically set the starp to 1.25, so reallly the bclk is 100 then.

 

[thesmokingman]

BCLK 125 x36 for 4.5, x37 for 4.625, and x38 for 4.75 I didn't touch anything else between the settings. I used the same guide linked in the page you cited that from.

Actually this board handles BCLK very uniquely so if I set BCLK to 125 it will automatically set the starp to 1.25, so reallly the bclk is 100 then.

Something must be off, because your chip is not following typical trajectory. 1.48v is higher than what most need to hit 5ghz. If you skim thru that thread, you'll find most chips will hit the cheat sheet I wrote, give or take. There must be something wacky about yer board or setup. Also strap 125 is not the same as straight bclk overclocking. The strap enables a divider for the pcie/etc bus to maintain 100 while the cpu bclk is 125. That said, shrugs it could just be a horrible overclocker I dunno w/o seeing your bios settings.

 

[M76]

Something must be off, because your chip is not following typical trajectory. 1.48v is higher than what most need to hit 5ghz. If you skim thru that thread, you'll find most chips will hit the cheat sheet I wrote, give or take. There must be something wacky about yer board or setup. Also strap 125 is not the same as straight bclk overclocking. The strap enables a divider for the pcie/etc bus to maintain 100 while the cpu bclk is 125. That said, shrugs it could just be a horrible overclocker I dunno w/o seeing your bios settings.

I don't know it never went over 4625, and was the same with every bios version before, where the options were quite different than with the latest bios. With earlier bioses I couldn't even oc if I didn't disable speedstep and turbo.

 

[Matthew Kane]

Probably something to do with the vdroop or your LLC setting on that particular mobo you're using. Awfully high for a 3820 @ 4.6 but again lots of variables at play here with CPU batch, mobo bios/setting/batch/revision etc.

That said I did go from an ES E5-26XX Xeon series hexacore at stock clocks 2.6ghz (can't remember the turbo) to a 3820 and for multi-core benchmarking (Cinebench) the Xeon was about 15-18% higher in scores compared to the 3820 @ 4ghz and single core numbers proved the 3820 to be faster performer. That is until I had the 3820 @ 4.7+ clocks the multi-core rendering scores was very close against the Xeon. The Xeon also had a lower TDP and naturally my temps was lower.

You will see a nice little gain with multi-core rendering if your Xeon's stock clock is 3ish ghz. For gaming the 3820 could prove to have the edge on certain games with a larger gap and some games the gap smaller.

 

[M76]

Probably something to do with the vdroop or your LLC setting on that particular mobo you're using. Awfully high for a 3820 @ 4.6 but again lots of variables at play here with CPU batch, mobo bios/setting/batch/revision etc.

That said I did go from an ES E5-26XX Xeon series hexacore at stock clocks 2.6ghz (can't remember the turbo) to a 3820 and for multi-core benchmarking (Cinebench) the Xeon was about 15-18% higher in scores compared to the 3820 @ 4ghz and single core numbers proved the 3820 to be faster performer. That is until I had the 3820 @ 4.7+ clocks the multi-core rendering scores was very close against the Xeon. The Xeon also had a lower TDP and naturally my temps was lower.

You will see a nice little gain with multi-core rendering if your Xeon's stock clock is 3ish ghz. For gaming the 3820 could prove to have the edge on certain games with a larger gap and some games the gap smaller.

Load line calibration was set to the best setting.

If anything it might have to do with RAM, since my ram is the cheapest kingston avaialable, and the four sticks are not even from one batch. And it's not kit ram, but individual sticks.

I had X58 before so I had three sticks in that, and when I switched to x79 I just added a fourth. Of course I couldn't get one that's exactly the same as the others. It's the same brand, same product line, but it's still very different.

Even the three I already had is made up of a dual kit, plus one that I purchased separately to save money.

 

[M76]

So here are the numbers:



So it's not as bad. About 25% less single thread performance but 30% increase in multi thread.

I'll post 3DMark results at a later time.

 

[M76]

The results are in. Even if nobody is interested. It's surprising to me. It seems that 3DMark tests very heavily favour more threads. I expected a very different result. I hope this means that the 2670 is more future proof for gaming than the 3820.

 

[Matthew Kane]

Try some individual game benchmarks to verify if it is true. The more modern 3DMark indeed favor more threads/cores than raw clock speed. Games on other hand completely different story.

 

[Deleted member 238539]

I have been following this thread for a little and I also traded my 3820 for an E5-2670. I haven't benched anything but I have to say I noticed a difference, not in speed, but just in general power. On the 3820 and 2133 DDR3 I could not seem to record on a CPU codec without drops in frame rate. But, on the Xeon with the extra cores and cache no issues at all, and no frame loss. I also can give some VMs more resources which makes them feel more like actual physical machines.

It was a great budget upgrade that I do not regret, also the upgrade to 32GB of RAM was nice as well.

 

[M76]

Try some individual game benchmarks to verify if it is true. The more modern 3DMark indeed favor more threads/cores than raw clock speed. Games on other hand completely different story.

The only game I had installed with a built-in benchmark was rise of the tomb raider. But that was totally useless. It gave wildly different results when I ran it twice. So I didn't bother.

I'm not going to put back the I7 now, you should've suggested titles when I asked what should I run 2 weeks ago.

Actually I used to run unigine valley, but I completely forgot about that this time.

 

[M76]

Something must be off, because your chip is not following typical trajectory. 1.48v is higher than what most need to hit 5ghz. If you skim thru that thread, you'll find most chips will hit the cheat sheet I wrote, give or take. There must be something wacky about yer board or setup. Also strap 125 is not the same as straight bclk overclocking. The strap enables a divider for the pcie/etc bus to maintain 100 while the cpu bclk is 125. That said, shrugs it could just be a horrible overclocker I dunno w/o seeing your bios settings.

I replaced the 3820 for other reasons. But now you can see my bios settings for 4500. Any less voltage and it becomes unstable. Below 1.35 it won't even post at 4500.

 

[Deleted member 238539]

M76

Question for you. How did you get your 2670 to lock @3.5? I normally just enable turbo and turn off speed step to get it to run at max frequency.

 

[M76]

M76

Question for you. How did you get your 2670 to lock @3.5? I normally just enable turbo and turn off speed step to get it to run at max frequency.

It's not locked to 3.5, I mentioned it in this thread that it's not possible to lock to 3.5 at least not with this board. So it goes up to 3.5 with turbo, but it's not locked.

 

[thesmokingman]

I replaced the 3820 for other reasons. But now you can see my bios settings for 4500. Any less voltage and it becomes unstable. Below 1.35 it won't even post at 4500.

Holy Carp! You have way too high of volts on your VCCSA, should only be 1.25v on the high side. VTT is low-ish. There's a formula for keeping your VTT and VCCSA within range of Vcore but I don't remember it off hand. As a rule of thumb I keep them both around 1.2v unless of other reasons like running quad GPU so then I'd want more VCCSA. High memory speeds will require more VTT.

What crazy memory are you using that requires 1.65v at DDR3 1600? Also, what is the BCLK Mode? It must be a new bios option as i don't see it in bios screens from a couple years back.


Assrock is a pain overclock wise, personally I never enjoy overclocking on their boards.

 

[M76]

The overclocking guide specifically mentioned that you should keep VCCSA as close to VCORE as possible. It used to be lower for me too, I only raised it recently when I read that part of the guide, but it didn't make much of a difference. It didn't say anything about VTT, apart from leave it alone.

BCLK mode basically can be manual or AUTO, it is the strap setting. In auto it will strap to 1.0 1.25 or 1.66 whichever is closer to the BCLK you set.

As you can see, the ram needs 1.65 according to the XMP profile. It's some Kingston HyperX Blu. Basically it's the cheapest RAM I could pick up at the time in late 2012 I think.

 

[thesmokingman]

The overclocking guide specifically mentioned that you should keep VCCSA as close to VCORE as possible. It used to be lower for me too, I only raised it recently when I read that part of the guide, but it didn't make much of a difference. It didn't say anything about VTT, apart from leave it alone.

BCLK mode basically can be manual or AUTO, it is the strap setting. In auto it will strap to 1.0 1.25 or 1.66 whichever is closer to the BCLK you set.

As you can see, the ram needs 1.65 according to the XMP profile. It's some Kingston HyperX Blu. Basically it's the cheapest RAM I could pick up at the time in late 2012 I think.

Hmm, not as close as possible, as close to a range or within a range. You should never run VCCSA at VCORE, that's freaking crazy. VCCSA is essentially the pcie bus, you only need it high if you run quad gpu. Btw, running high volts on the secondary voltages increases heat a lot and adds to degradation needlessly. I wouldn't set ram to 1.65 for 1600mhz. Any ram should be able to handle 1.5v for slow speed. Also your LLC is set rather low, or is that high? Is there a difference when you set the BCLK Mode one way vs the other?

 

[M76]

Hmm, not as close as possible, as close to a range or within a range. You should never run VCCSA at VCORE, that's freaking crazy. VCCSA is essentially the pcie bus, you only need it high if you run quad gpu. Btw, running high volts on the secondary voltages increases heat a lot and adds to degradation needlessly. I wouldn't set ram to 1.65 for 1600mhz. Any ram should be able to handle 1.5v for slow speed. Also your LLC is set rather low, or is that high? Is there a difference when you set the BCLK Mode one way vs the other?

The ram requires 1.65 for 1600. It's not me who is decided to set it arbitrarily to 1.65 because I thought it would be good. It automatically sets ram voltage to 1.65 when I enable the xmp profile. I found one guide where it says the vcssa should be .3 - .4 to the vcore, but that seems very low to me.

Level 1 is the best setting for load line calibration meaning the least vdrop under load.

What do you mean is there a difference? If I set it manually to 1.25 there is no difference to it being on auto with 125BCLK. If I set it to 1.66 I can't get the ram to run at 1600, because it'll either run at 1333 or way above 1600 that this cheap ram can't handle.

 

[M76]

So far I never seen any guide mention vtt voltage. Now I set it to 1.25, and VCCSA too. Let's see what happens.

Update: No, there is no magic. Still needs at least 1.47 for 4.6. to not instafreeze in prime95.