Intel Xeon E5-2600: Doing Damage With Two Eight-Core CPUs

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
[citation][nom]cangelini[/nom]Gigabyte told me my -UD5 would work, but no such luck. Drop in the -2687W and it fires up/shuts down in a loop. Guess they were secretly hoping it'd work and counting on my processor not getting damaged? :p[/citation]
I only know one guy that has direct contact with Gigabyte, and he may have access to a beta BIOS. To my knowledge I've only seen ASRock post with the E5. Otherwise, even if the E5 posts it's not going to run properly without proper BIOS support.

Frankly, I was really surprised that the ASUS P9X79 WS didn't have an updated BIOS for the E5; they normally will offer Xeon support and IMO it's in the works; ECC, Non-ECC is supported. Example, the P8P67 WS Revolution lists the E3's. Therefore, IF you have 'P9X79 WS' on hand contact your 'ASUS guy' and ask for a beta BIOS.

Good Luck! :)

PS - Please let me know, I'd be VERY interested to know how it OC's and compares! Thanks!
 

willard

Distinguished
Nov 12, 2010
2,346
0
19,960
[citation][nom]leeashton[/nom]for Data mining CPu cores matter, if interlarlgos has 40 core they would kill the Xeon[/citation]
But Interlagos doesn't have 40 cores, it has 8. And it barely keeps pace with the now two year old Xeons, and is absolutely slaughtered by the current lineup. It's not like we haven't seen Interlagos yet, we have. It's here, and it's faring even worse than Bulldozer.

I just don't get the point of you posting that. Of course Interlagos would beat the Xeon if it had 40 cores. So would a Pentium 4. But this is reality, and you don't get to change hardware platforms through wishful thinking. Saying "But if Interlagos were better than Xeon it would beat the Xeon" is about the most pointless thing you could say.

It's not better than the Xeon, it's much, much worse.
 

jgutz2006

Distinguished
Jul 7, 2009
473
0
18,810
[citation][nom]Rock_n_Rolla[/nom]-- How about info on its overclocking performance??? This review would be much informative if the reviewer included infos on how far can these new E5 Xeons can go when heavily overclocked and this would give good insight to people specially those who are into (including me) multi-threaded digital content creation, editing, 3d animation and rendering where very fast processing power are needed in order to get the job done in the soonest possible time where on-time project deliverables is at most priority. As far as rendering is concerned, rendering a basic single 1920x1080 image of a living room interior with 1 ambient room light and 1 directional external light and with few assorted furniture, flooring, wall and figurine textures using Arion Render or Vray on a dual Xeon 5600 (westmere) processor server would take 20-40 mins alone andthe rendering would take much much longer (for hours) if the image is on a much higher resolution (specially designed for large printing purposes) and uses more complex lighting setup and textures for realistic rendering.I hope these new E5 Xeon processor would provide much much needed processing speed to aid Digital content creator and editor to make their job much faster and more efficient.[/citation]

When data centers have high density racks filled to the brim with dual to octo socket chips, they last thing they want to do is mess with the overclocking, extra heat not to mention taking a risk with stability/uptime and lifetime, not to mention the $$ they would be spending IT staff to manually work with one by one would be insane! Most of the vendors making dual+ socket boards arent making extreme overclocking focused boards, but stability and insane amounts of DIMMs.
That being said, it would be interesting to see if there was much of a boost or overhead for any overclocking at all, Workstation users rendering huge 3d models could potentially benefit from a little overclocking.
 

willard

Distinguished
Nov 12, 2010
2,346
0
19,960
[citation][nom]jgutz2006[/nom]When data centers have high density racks filled to the brim with dual to octo socket chips, they last thing they want to do is mess with the overclocking [...] That being said, it would be interesting to see if there was much of a boost or overhead for any overclocking at all, Workstation users rendering huge 3d models could potentially benefit from a little overclocking.[/citation]
Exactly. These are designed for high density computing, and the extra heat and electricity associated with overclocking puts a kibosh on the idea immediately. When you already have dedicated HVACs and electricity costs in the tens to hundreds of thousands, or even millions of dollars annually, you're not looking to overclock. Instead, you're going to be eying those lower TDP E5s to save yourself a bundle on electricity and maintain the same, or better, performance.

On the high end with people needing every bit of power they can, they'd probably be looking to simply fill more racks with the top end chips, rather than overclocking. Reduced lifespan isn't really an issue (typically you replace your hardware on a 3 year cycle), but the increased heat and power usage is. Numerous Tom's articles have shown that overclocking is usually an efficiency loss, and the electricity costs of these racks far outweigh the cost of components.
 

lilcinw

Distinguished
Jan 25, 2011
833
0
19,010
1. Chris, it must be nice to be the boss, you always get to play with the best toys.

2. How did I miss the link to the giveaway contest? I know it has to be in there somewhere. :D
 
G

Guest

Guest
Where is the contest to win this build? Just wishing. Great review.
 

jgutz2006

Distinguished
Jul 7, 2009
473
0
18,810
I can personally say that i LOVE server chips for workstations as i first setup my primary desktop with a dual socket Xeon 5560 setup for total of 8 cores and 16 threads and then i graduated to a dual socket 5660 setup with 12 physical/24 threads which is what i currently run with all 12 DIMMS (6 per Socket) filled with DDR 3 1066 ECC RAM. Interestingly enough, My 240gb Vertex 2 SSD OS drive is the weakest link on the Windows Experience Index as i'm at 7.8 across the board but 6.9 on primary disk.

Yes its obviously overkill for a personal computer/workstation so why bother? well because i can, but i do have an architect wife who can bring home CAD drawings for rendering which can be completed in hours whereas their work computers would still be churning away the next morning if left running overnight.
 

silverblue

Distinguished
Jul 22, 2009
1,199
4
19,285
[citation][nom]suryasans[/nom]Of course, AMD admits that Its own products are rubbish> Even, its engineers have been leaving from AMD because of shrinking AMD's cash due to AMD buyout of Seamicro.[/citation]
Really? How would you know that?

I heavily suspect that you're sans2212 from Anandtech and, because you got moderated rather heavily over there over the past day or so, you've settled here instead.

I think AMD, more than anybody, is acutely aware of how poor Bulldozer is for anything but heavily threaded scenarios, and even then, it's not consistently ahead of the 2500. I suspect that it's not decoding enough instructions per clock (actually, less than PII) and that the entire cache architecture needs an overhaul (it's clocked far too slowly for a start), though I'm definitely no expert. It's entirely possible that an L3 cache-less Piledriver with such improvements would monster Bulldozer. In any case, the first iteration of an architecture isn't always as fast as what people want - K6 and P4, for example, started off poorly and generally improved with each revision. Not every new CPU family is that far ahead of its predecessor; definitely behind in some cases. Coincidentally, both P4 and Bulldozer did better at video and compression work than pretty much any other workload you threw at them.
 
[citation][nom]willard[/nom]But Interlagos doesn't have 40 cores, it has 8. And it barely keeps pace with the now two year old Xeons, and is absolutely slaughtered by the current lineup. It's not like we haven't seen Interlagos yet, we have. It's here, and it's faring even worse than Bulldozer.I just don't get the point of you posting that. Of course Interlagos would beat the Xeon if it had 40 cores. So would a Pentium 4. But this is reality, and you don't get to change hardware platforms through wishful thinking. Saying "But if Interlagos were better than Xeon it would beat the Xeon" is about the most pointless thing you could say. It's not better than the Xeon, it's much, much worse.[/citation]

Interlagos has up to 16 cores at up to 2.6GHz stock clocks with over 2.4 or 2.5GHz Turbos. A 40 core Netburst CPU would probably NOT beat the CPUs listed here. Valencia is limited to 8 cores. Interlagos does have some decent performance for it's price, but yes, it doesn't compare in performance to the more expensive Xeons. I just did a performance comparison between the sub $700 SB-EPs and they aren't as fast as the sub $700 Interlagos CPUs with fairly similar or even lower prices, all comparisons done through Newegg.

For example, the $620 SB-EP Xeon E5-2630 (the closest CPUs to it in it's family are both $200 lower and $300 higher) probably doesn't beat to the $400-$500 twelve core Interlagos Opterons, let alone the $600-$700 16 core Interlagos Opterons. Of course, AMD doesn't compete well after those price points because they don't have faster CPUs, but in the places where they do compete, they seem to have low enough prices to be considered for highly parallel workloads. The eight core FXs are neck and neck with the quad core i7s so it seems that a similar number of threads including hyper-threaded threads and Bulldozer cores, when at similar clock speeds or with Bulldozer a little higher, have fairly comparable performance.

Intel has closed the gap in memory capacity/channels (AMD quad channel vs. Intel triple channel probably wasn't favoring AMD too well anyway, Intel has better controllers) AMD seems to be owning all fronts for highly threaded price/performance. AMD's problem with the desktop is that it doesn't need highly threaded performance as much as lightly/single threaded performance, of which AMD currently sucks at. However, for low/mid end servers, AMD seems to be doing okay. Their problem with servers is currently market adoption as far as I can tell. I can say that workstations and such that can't use the amount of cores that AMD needs to be competitive have little use for AMD, at that point the highly threaded performance is not obtainable so price/performance drops sharply on the AMD side whilst Intel is largely unaffected.
 

jgutz2006

Distinguished
Jul 7, 2009
473
0
18,810
[citation][nom]blazorthon[/nom]AMD seems to be owning all fronts for highly threaded price/performance. AMD's problem with the desktop is that it doesn't need highly threaded performance as much as lightly/single threaded performance, of which AMD currently sucks at. However, for low/mid end servers, AMD seems to be doing okay. Their problem with servers is currently market adoption as far as I can tell. I can say that workstations and such that can't use the amount of cores that AMD needs to be competitive have little use for AMD, at that point the highly threaded performance is not obtainable so price/performance drops sharply on the AMD side whilst Intel is largely unaffected.[/citation]

Not only does AMD get waxed in single thread but its going backwards even which in unfortunate. AMD and higher core counts is not completely a horrible idea because Virtualized environments can better spread the wealth around all the cores and even dedicate 1 or more specific cores for that specific virtual machine.
You mention Xeons vs Opterons at the same price points where AMD does in fact take the lead but on top of the CPU's running cheaper, the accompanying chipsets are also cheaper lowering the overall cost of the entire system even more which could allow someone with a budget to add more RAM or Storage compared to a Xeon system at the same price point. I run a Magny Cours HP DL385 G7 Server at home for virtualization of other servers and use it to run test environments (Like Server 8) but i use the snapshot Virtual Machines as a consistent base install for when working on client machines remotely and the AMD system is wonderful as more cores is more important that a little more performance (or a lot more...)
 

ramicio

Distinguished
Nov 8, 2011
17
0
18,510
For x264 encoding, they recommend not going past 16 threads for quality reasons. The program is not going to just choose the maximum number of cores available x 1.5.

The SB-E with 6 cores just about perfectly utilizes 16 threads.

2-socket systems simply aren't utilizing what can be done. For the new Xeons it would shine by running 3 instances of x264 with 16 threads each, or 4 instances of 12 threads each.

Do you use a 64-bit version of x264 for testing?
 
G

Guest

Guest
uh, blazorthorn, according to anandtech, Intel wins in price/performance and performance/watt too now. The 2630 does beat the Interlagos 16 core. Nice post though. Just happens to be wrong.
 
G

Guest

Guest
From anandtech:

"For those who are more price sensitive, the Xeon E5-2630 costs less than the Opteron 6276 and performs (very likely) better in every real world situation we could test."

 
How could the 2630, six cores, hyperthreading, 2.3GHz.. possibly beat the 16 core Interlagos at 2.2GHz? That completely defies the performance scaling I've seen with FX-8xxx vs i7. The 8 core FXs are almost as good as a quad core i7. It's natural to assume that the 16 core Interlagos should be almost as good as an 8 core SB based Xeon. How could a quad core Xeon then go on to beat a sixteen core Interlagos? Performance scaling shouldn't be identical to the desktop side for obvious reasons, but it shouldn't be THAT far off.

The only way I see the Interlagos losing is if the software doesn't fully support 16 cores. In that case, yes, the six core i7s could win, but for highly threaded work, it should take the 8 core i7s and/or a higher clock frequency to top the 16 core Interlagos in the most highly threaded software. There should not be any way for desktop performance scaling to be twice as god as server scaling. Server scaling should be better for increased core counts.

Unless performance comparisons and scaling are radically different here (they shouldn't be that different), a six core SB-EP Xeon with Hyper-Threading should lose to an Interlagos Opteron with a similar clock frequency and 16 cores. Perhaps the Interlagos relying on two dies hurts scaling against a single die Xeon? I don't know.
 
Also, the Opteron 6272 is listed at $539.99, the Xeon E5-2630 at $619.99, so the 2630 is NOT cheaper than some 16 core Interlagos chips. Newegg doesn't have the 6276 listed, but it would need to be more expensive than the 2630 because the 6274 is at $649.99. The 12 core Interlagos should be comparable to a six core Xeon E5 in highly threaded performance.

Of course the Xeons win by a huge margin in anything that doesn't tax all cores of the Opteron, but in situations that use all of the threads, the 16 core Opterons should win.
 
G

Guest

Guest
blazorthorn, not sure what you are getting at. Go to anandtech and see how they benchmarked it. They used server based benchmarks that used multiple threads. I don't know what you have seen, but every benchmark site I have been to including Toms, Anadtech, CPUMark, cpubench, PassMark, etc. don't list the latest Athlons as close to the latest Intel CPUs. You are pretty much the only source I have on that one. Now, if you want to publish your benchmarks and your methodology, then feel free to change our minds. Interlagos isn't that fast I'm afraid. It is not the most efficient architecture out there which is why a 6 core CPU is beating it.


As far as price, the 2630 is $612. The 6274 is $639.

The proof is in the numbers, but again, feel free to publish your own results and prove it different. All the "should be better" talk really isn't saying much. All AMDs CPUs should do better than they are but the design is not very good right now.
 
[citation][nom]BenchMarkGuruII[/nom]blazorthorn, not sure what you are getting at. Go to anandtech and see how they benchmarked it. They used server based benchmarks that used multiple threads. I don't know what you have seen, but every benchmark site I have been to including Toms, Anadtech, CPUMark, cpubench, PassMark, etc. don't list the latest Athlons as close to the latest Intel CPUs. You are pretty much the only source I have on that one. Now, if you want to publish your benchmarks and your methodology, then feel free to change our minds. Interlagos isn't that fast I'm afraid. It is not the most efficient architecture out there which is why a 6 core CPU is beating it. As far as price, the 2630 is $612. The 6274 is $639. The proof is in the numbers, but again, feel free to publish your own results and prove it different. All the "should be better" talk really isn't saying much. All AMDs CPUs should do better than they are but the design is not very good right now.[/citation]

What are you going on about Athlons? I didn't say anything about them. FX-8150 is right behind an i7-2600. An Interlagos CPU is basically two Valencia dies on a single chip, Valencia is basically an Opteron version of FX. Xeon E5-2630 is a little different from the i7-2600, but if a quad core i7 is just ahead of an eight core FX with a similar clock frequency, then through easy logic, doubling the core count of either chip shouldn't change performance differences too much. That means that an 8 core i7 should inch ahead of a 16 core Interlagos with a similar clock rate. A six core SB Xeon should NOT beat a 16 core Interlagos at a similar clock rate.

Sure, Bulldozer should be better, but that's not what I'm going on about. Even with it performing as it does, the only way that the Interlagos could be beaten like this is if it has significantly less IPC than even FX and/or the Xeon has significantly more IPC than the i7s.

I'm not a source of anything beyond asking why the server chips are scaling so extremely differently than the desktop chips with the same architectures.

I'm not saying that Anand and the rest are wrong, only that if they are right then the Interlagos CPUs are even more messed up than the FX CPUs. Also, Bulldozer's problems go even beyond just the design. I'm not some idiot fanboy that thinks FX is great, it sucks unless you can use all cores and I already know this. However, the numbers on these server chips don't scale with the desktop chips and seem very misplaced.

EDIT: Also, the 2630 is $619,99, NOT $612. Not a big difference and it doesn't really matter there, but you were still wrong. Also, the 6272 is only $539.99, much cheaper, and it should be a little faster in highly threaded work, going off the real performance of FX. Will the difference be made up in power usage? I don't know, I do know it would take me several years for it to be made back up even if the system was running 24/7. Of course, this calculation only includes the CPUs, not the rest of the system, but the only other differing factor would be the motherboard anyway. So long as the Intel board isn't more than $50 cheaper (I'm no expert on server motherboards, but I doubt it will be cheaper), the AMD system would be faster and cheaper for ME for until it's time to be replaced, unless as Anand and others imply, Interlagos cores are even worse than FX cores or they simply fail to scale performance up correctly, or some other bottleneck I'm missing.

We've seen FX-8150 benchmarked against i7. We know it isn't to far off and if the i7 were a tri core (for comparisons for the Xeon in question being a 6 core), the FX would win. So, why is it after we double pretty much everything, the same is not true? Something is wrong.

PS: All prices from Newegg.com
 

nadavp3

Distinguished
Apr 3, 2009
45
0
18,530
to bad you didnt have windows 8 server for the test, and that you didnt, for the fun of it, tried out some game tests.. just for the sake of testing how good is top of the line server duel cpu work on games.
 
G

Guest

Guest
blazorthorn
You are quoting prices from 1 source, newegg, not from the manufacturer. Thus, it is YOU whom are wrong. You are also clueless. The benchmarks were posted and prove without a doubt the conclusion that the 6274 was slower than the E2630. Why are you even arguing? The FX8150 is behind the i5 2500k by the way. This is also server workload that will stress bandwidth. It isn't just a desktop test running Crysis.

If you want to own the AMD hardware, then go buy it. You have your head in the sand though if you think the price per performance or watt is better.
 
WTF are you saying now? The i5 doesn't win in highly threaded work against the FX-8120 nor the 8150 nor even the Phenom II x6s and this is a server talk, I was talking about highly threaded work. I've said that plenty of times. If it is one price at Newegg, then that is the price I care to look at. Intel and AMD can say the prices they want them to sell for all they want, it's the retailers and such that decide how much they sell them to us for.

Of course the i5 wins in lightly threaded work. Also, why are you mentioning anything about games? Crysis has nothing to do with anything I've said. All I've talked about is server loads.

The manufacturer can say prices all they want, it is the places where you actually buy the CPUs that choose how much they sell them to you for. Also, I didn't say anything about performance per watt being better on AMD. In fact, I already said it was worse. Also, saying whom doesn't make you look smart unless you bother to use proper grammar, especially when you type whom are when it's supposed to be who is. I don't normally grammar police, but the only reason for you doing that is to make you look more intelligent and credible than I am, but you failed at that.
 
Status
Not open for further replies.