E5-2696 v4 22core CPU for $150 or less

[wildpig1234]

2696 v4 for sale | eBay

I figure i might swing for a pair since it's a nice sent out for my dual 2011-3 MB as being the max upgraded path for it.
pretty decent price of $300 for 44core/88 threads total? would still being very capable for the price i reckon?
What yall think?
These 2696 v4 oem appear to be cheaper than the 2699 v4 which are non oem and might be more widely accepted in various MB? I don't think i will have problem running them with my asus Z10-PE D16 which accept ES cpus just fine

 

[amp88]

My 2 cents: this is an OK deal for these specific CPUs, but bad value anyway.

Check this comparison between the 2696 v4 and 2680 v4; the 14 core 2680 v4 is ~10% slower in single core (thanks to the 300MHz lower single core turbo) and ~27% slower in multicore (despite having only 64% of the cores of the 2696 v4). 2680 v4s are incredibly cheap at the moment; you can get them on US eBay from either US or Chinese sellers for only 20-30 USD each.

Unless you actually need the extra performance or you just want to have more cores for...reasons, then the SKUs at, or near, the top of the pile are almost always bad value, even on older platforms like his.

 

[wildpig1234]

Or should i spring for the epyc 7702p with 64 core/128 threads and a new MB. total price would be close to $1300....lol. I would imagine the epyc 7702p would be at least twice as fast as the dual 2696 v4?

Yes, 2680 v4 are definitely dirt cheap. You can actually get a pair for $30 now ...

 

[amp88]

Yeah, 2nd gen EPYC would be a massive step forward in raw performance, in addition to providing other benefits like many more PCIe lanes (128 Gen4 for 2nd gen EPYC vs 40 Gen3 for the E5 v4, per CPU).

One thing to keep in mind if you're looking at an EPYC system is that you should look for a motherboard which supports both 2nd and 3rd gen EPYC CPUs. For Supermicro, for example, the H12 line (including H12SSL-i) support 2nd and 3rd gen (may require a BIOS update). However, the H11 line (including H11SSL-i) only support 1st and 2nd gen. Being able to drop in a 3rd gen EPYC in a few years when the prices come down is probably going to be a cost effective upgrade in certain use cases, so that's something to consider.

 

[wildpig1234]

the H12 MB costs twice as much as the H11 MB...lol..

 

[amp88]

You can get bundle deals from China which are quite good value. For example (just the first hit, I have no affiliation with this seller or any others):

AMD EPYC 7402P CPU + Supermicro H12SSL-i + 2133P RAM multiple choices | eBay

AMD EPYC™ 7402P. The CPU Mall. 1x Supermicro H12SSL-i motherboard. Processor Type : AMD. Number of CPU sockets supported. Number of CPU cores. twenty four. Level 3 cache.

www.ebay.com

~1000 USD for an H12SSL-i + 7402P + 16GB of RAM.

 

[wildpig1234]

You can get bundle deals from China which are quite good value. For example (just the first hit, I have no affiliation with this seller or any others):

AMD EPYC 7402P CPU + Supermicro H12SSL-i + 2133P RAM multiple choices | eBay

AMD EPYC™ 7402P. The CPU Mall. 1x Supermicro H12SSL-i motherboard. Processor Type : AMD. Number of CPU sockets supported. Number of CPU cores. twenty four. Level 3 cache.

www.ebay.com

~1000 USD for an H12SSL-i + 7402P + 16GB of RAM.

7402p is very significantly slower than 7702p. Not worth the $300 saving imo. I would rather take the 7702p with the H11 MB than 7402p with the H12 MB

 

[amp88]

7402p is very significantly slower than 7702p. Not worth the $300 saving imo. I would rather take the 7702p with the H11 MB than 7402p with the H12 MB

Yeah, I was just using that as an example, intending for you to take a look and see if there were any other bundles that were a better deal for you. Obviously if you know about the limitation with the H11 and you're happy with it then go for it. It certainly makes more sense than dropping around $300 on the 2696 v4s anyway!

 

[chrgrose]

No one here is entirely wrong, but I wouldn't echo the sentiment that if you are going to upgrade to 2696v4's, you "might as well" just get an EPYC system. I think it's clear that a 2nd gen EPYC system is the real alternative here (second hand Xeon scalable is just far more expensive), but the 2696's might not be a bad choice. The first obvious reason is that a good EPYC system is still just going to be more expensive. It looks like you will pay a premium of about 2x per core for a good 2nd gen EPYC cpu. You can also easily get a whole 2011 platform (e.g., I rather like the T7810/T7910 towers) for around the cost of a 1P EPYC motherboard, maybe less. And the OP apparently already has a platform.

Anyway, I think the choice of getting the 2696v4's vs. going EPYC is a simple question of how much performance you need, and how much you are willing to spend. If your goal is to maximize performance while spending almost nothing, amp88 had the right idea, although I might suggest the 2697v3, which can be had for about $15/CPU right now.

Also, end of the line for dual socket 2011-3 is probably not the 2696v4, but the 2699Av4, but those are twice as expensive. This also doesn't include the 2699Pv4, but that appears to be only a wikipedia legend.

 

[Cruzader]

2683v4 -15$ area
2690v4 - 25-30$
2695v4 - 40-45$

Unless you have a load that you are hitting the ceiling with those, i dont see the point in taking the larger price leaps beyond them.

 

[chrgrose]

By the way unless your workload uses accelerators, Cinebench r23 scores might be interesting (from hwbot):

2x Xeon 2699v4: 12594
2x EPYC 7702: 28577

Indicating about 2.3x the throughput for almost 3x the cores, which means about 30% higher performance per core from the xeon. Wow, is that really the case or is 7702 perf/core better and r23 is a really bad metric?

 

[ano]

7402p is very significantly slower than 7702p. Not worth the $300 saving imo. I would rather take the 7702p with the H11 MB than 7402p with the H12 MB

workload dependant though, base clock is much lower

 

[wildpig1234]

No one here is entirely wrong, but I wouldn't echo the sentiment that if you are going to upgrade to 2696v4's, you "might as well" just get an EPYC system. I think it's clear that a 2nd gen EPYC system is the real alternative here (second hand Xeon scalable is just far more expensive), but the 2696's might not be a bad choice. The first obvious reason is that a good EPYC system is still just going to be more expensive. It looks like you will pay a premium of about 2x per core for a good 2nd gen EPYC cpu. You can also easily get a whole 2011 platform (e.g., I rather like the T7810/T7910 towers) for around the cost of a 1P EPYC motherboard, maybe less. And the OP apparently already has a platform.

Anyway, I think the choice of getting the 2696v4's vs. going EPYC is a simple question of how much performance you need, and how much you are willing to spend. If your goal is to maximize performance while spending almost nothing, amp88 had the right idea, although I might suggest the 2697v3, which can be had for about $15/CPU right now.

Also, end of the line for dual socket 2011-3 is probably not the 2696v4, but the 2699Av4, but those are twice as expensive. This also doesn't include the 2699Pv4, but that appears to be only a wikipedia legend.

By the way unless your workload uses accelerators, Cinebench r23 scores might be interesting (from hwbot):

2x Xeon 2699v4: 12594
2x EPYC 7702: 28577

Indicating about 2.3x the throughput for almost 3x the cores, which means about 30% higher performance per core from the xeon. Wow, is that really the case or is 7702 perf/core better and r23 is a really bad metric?

i think there's a bit of an error in your stat?

1x EPYC 7702p (128threads) cpu-z score: 25866 : CPU-Z Benchmark for AMD EPYC 7702P (33T) - CPU-Z VALIDATOR (x86.fr)
1x 2699 v4 (44threads) cpu-z score: 9122 : CPU-Z Benchmark for Intel Xeon E5-2699 v4 (33T) - CPU-Z VALIDATOR (x86.fr)
1x 2696 v4 (44threads) cpu-z score: 9001: CPU-Z Benchmark for Intel Xeon E5-2696 v4 (33T) - CPU-Z VALIDATOR (x86.fr)

So a single 7702p cpus is actually 2.8 time the single 2696 or 2699 v4. A dual 2696/9 v4 setup would still be significantly slower than one 7702p. Dual 2696/9 v4 (44core/88threads) is still down in both core/thread counts as well as ipc and other refinements compared to just one single epyc 7702p (64core/128threads) ...lol... Also the fact that you have to use the slower QPI between the two v4 cpu in a dual cpu setup will also make it even slower. dual 96/9 v4 won't get anywhere close to 18000. probably more like 17000 or less.

List of Intel Xeon processors (Broadwell-based) - Wikipedia

The 2699v4 and 2696v4 seems to be very very similar. Both has 22cores/44threads and nearly identical boost speed. If anything, the 2696 v4 is actually slightly better due to the higher 150W tdp compared to 145 for the 99 part. The 96 part is oem while the 99 part is not which means that the 96 part might not be as widely compatible with some MB as compared to the 99 part? So for all intents and purposes, the 2696 v4 is really equivalent to 99 as the end point for 2011-3 MB i think. Right now you can get one -96 v4 for around $150 vs $200+ for a -99 v4 so i think -96 v4 is better if your MB has wide compatibility which i expect most of the ones from asus and supermicro and most MB not from dell, hp, etc would be ok.

performance/$ wise for me, the dual 96 v4 cpu would cost $300 for a likely cpuz benchmark of 16000 compared to 7702p costing $1300 with a new MB and getting 25800 for the cpuz mark with also more pcie lanes.....

decision decision..... maybe i should save up to do both? lol

 

[chrgrose]

I'm not sure if there is an error in my stats, but it could be that Cinebench r23 is not as good of a metric. Although I'm certainly not familiar with what r23 and CPU-Z are computing. The CPU-Z benchmark is much more generous to epyc. But still, your scores indicate that the cores on the 2699/2696 are 'faster' than the EPYC cores, although now only by about 1-3%. Hence, one of the benefits of epyc is in having more cores, not faster cores (at least when all threads are working). So, unless you need more/faster pci lanes or higher memory bandwidth, going epyc might be a bit of a waste unless you get at least one 64 core cpu. So then the main question is still probably if you need more power than the 44 cores, and whether it is in your budget to get that extra power. The other question is what will depreciate faster, but that's not clear to me lol.

You mention the QPI penalty, which is true, but you would know more than I about whether there are significant core-core latency problems in your parallel work. I don't think there is any such penalty in the CPU-Z benchmark. I've seen 8 socket 8894v4 (192 cores) score 86329, which in terms of perf/core is about 10% faster than these 7702/2699v4 cpus. This despite only 1 QPI link between processors and even multiple hops needed for core-core communication.

You are right the 2696 and 2699 are basically identical when all threads are operating (2696v4 has higher single-thread boost). But maybe they behave differently when not boosting? 9122 and 9001 are pretty close anyway.

 

[RolloZ170]

The 96 part is oem while the 99 part is not which means that the 96 part might not be as widely compatible with some MB as compared to the 99 part?

ASRock X99M Killer:
E5-2699v4 TDP=145W, TDC=189A, working
E5-2687Wv4 TDP=160W, TDC=189A, working
E5-2696v4 TDP=150W, TDC=199A, NOT working

Intel S2600WT series:
E5-2699v4 TDP=145W, TDC=189A, working
E5-2696v4 TDP=150W, TDC=199A, NOT working

HP DL380 G9
E5-2699v4 TDP=145W, TDC=189A, working
E5-2696v4 TDP=150W, TDC=199A, working

 

[nutsnax]

2696 v4 for sale | eBay

I figure i might swing for a pair since it's a nice sent out for my dual 2011-3 MB as being the max upgraded path for it.
pretty decent price of $300 for 44core/88 threads total? would still being very capable for the price i reckon?
What yall think?
These 2696 v4 oem appear to be cheaper than the 2699 v4 which are non oem and might be more widely accepted in various MB? I don't think i will have problem running them with my asus Z10-PE D16 which accept ES cpus just fine

IMO this is a question of use-case. What will this be used for?

I have a dual 2698 v3 that can outperform my single 7532 if I'm spawning lots of threads to do something simple like pull down API data or something. But if I need lots of powerful threads the 7532 crushes my dual 2698v3.

Edit: just remembered this is in great deals section. May want to move it somewhere else as this is kinda also build advice. Just my $0.02.

 

[wildpig1234]

I have the 2696 v3 and it seems like 2696 v4 gives a small bump in speed but would have to go 7702 to get a lot more drastic speed increase..lol..

 

[GregTaylor]

The 14nm v4's are significantly less expensive to keep cool than their 22nm v3 counterparts. They are far better in workstations where performance can be thermally constrained.

 

[chrgrose]

Where has the large difference between v3 and v4 been shown?

 

[BackupProphet]

The biggest difference I've noticed is https/ssl. I think V4 is around 30% faster than V3 here, especially noticeable on file servers