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

[BackupProphet]

I just did a benchmark with Clickhouse, dual 2690 v4 vs dual 2699 v4, 28 cores vs 44 cores.

The dataset fits in memory, and mitigations=off as you get at least 10% performance increase

The 2x 2699 v4 result:
1 row in set. Elapsed: 5.762 sec. Processed 14.13 billion rows, 226.09 GB (2.45 billion rows/s., 39.24 GB/s.)
Peak memory usage: 58.88 MiB.

The 2x 2690 v4 result:
1 row in set. Elapsed: 6.581 sec. Processed 14.13 billion rows, 226.09 GB (2.15 billion rows/s., 34.35 GB/s.)
Peak memory usage: 58.60 MiB.

The Intel 2690 v4 is only 13,5% slower. But the price difference is 9x.
Most likely the bottleneck here is caused by the QuickPath Interconnect (QPI)

The Intel 2690 v4 is really good value for the money today

 

[ano]

The Intel 2690 v4 is really good value for the money today

we ran 2690v4 in compute/hypervisors for years, the 6130, 6154 that came after was a dissapointment, need to goto like 6354 for intels to be impressive agian

 

[BackupProphet]

I have tested the Intel Gold 6148 20 cores, with 2666mhz memory instead of 2400mhz and with UPI (3 lanes) which is significantly faster than QPI and 2 lanes. I don't have the numbers here now, but I did get on average 20-25% better performance than the 2690 v4

 

[wildpig1234]

I just did a benchmark with Clickhouse, dual 2690 v4 vs dual 2699 v4, 28 cores vs 44 cores.

The dataset fits in memory, and mitigations=off as you get at least 10% performance increase

The 2x 2699 v4 result:
1 row in set. Elapsed: 5.762 sec. Processed 14.13 billion rows, 226.09 GB (2.45 billion rows/s., 39.24 GB/s.)
Peak memory usage: 58.88 MiB.

The 2x 2690 v4 result:
1 row in set. Elapsed: 6.581 sec. Processed 14.13 billion rows, 226.09 GB (2.15 billion rows/s., 34.35 GB/s.)
Peak memory usage: 58.60 MiB.

The Intel 2690 v4 is only 13,5% slower. But the price difference is 9x.
Most likely the bottleneck here is caused by the QuickPath Interconnect (QPI)

The Intel 2690 v4 is really good value for the money today

I finally got my 2699 v4 pair as well as a pair of 2690 v4. And yes i can also confirm what you say about it from running y cruncher. The gain between the two cpus on y cruncher is also marginal.

It's really weird because on cinebench and on cpuz benchmarks, the 2699 v4 is significantly faster than 2690 v4. But i am thinking that maybe cinebench and cpuz are more cpu bounded whether as y cruncher also very dependent on memory access speed. The benchmarks for y cruncher using epyc cpu seems a lot faster and maybe that's because epyc has 8 channel memory.

Between 2680 v4 ($15/cpu) and 2690 v4 ($21/cpu) , I think that given the low total price, it's probably better to get the 2690