https://www.servethehome.com/amd-epyc-7601-dual-socket-early-power-consumption-observations/With the level of power/ performance of the new systems, you can essentially replace four Intel Xeon E5-2600 (V1) servers with a single dual socket EPYC node and get more performance (in most cases) in a single node that uses half the power. That is absolutely stellar. The AMD EPYC platform is still seeing major updates to BIOS for power and performance which is why we are calling these preliminary results. At the same time, we are already seeing some impressive figures.
I agree we need real reviews, but if the performance pans out, that means equal performance to slight edge to AMD. But, that is also against v4.Looks like AMD reduced Intel numbers do to the apparently unfair compiler advantage on Intel scores if TH is correct. I could see it proper to some degree but not 46%. Wacky.
Need real reviews.
https://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed/2
Funny estimation scheme. Not to say hilarious, but certainly funny.
Thank god they finally killed libquantum, though.
So, availability is 2nd half 2017 if I read that correctly. So if they want to call it a launch then it's a paper launch. Should have called it an announcement.
http://www.anandtech.com/show/11551...w-7000-series-cpus-launched-and-epyc-analysisToday’s announcement of the AMD EPYC product line sees the launch of the top four CPUs, focused primarily at dual socket systems. The full EPYC stack will contain twelve processors, with three for single socket environments, with the rest of the stack being made available at the end of July. It is worth taking a few minutes to look at how these processors look under the hood.
Wow, is that true? AMD flat out lied about the Intel benchmark results?
Wow, is that true? AMD flat out lied about the Intel benchmark results?
Intel's response to the server launch is indeed EPYC
"With our next-generation Xeon Scalable processors, we expect to continue offering the highest core and system performance versus AMD. AMD’s approach of stitching together 4 desktop die in a processor is expected to lead to inconsistent performance and other deployment complexities in the data center."
I guess that is what intel thinks about infinity fabric -
The issue is shared hosting/cloud computing where VMs are supposed to be isolated from each other. This regularly fails which is very worrying in a time where computation is regularly offloaded to AWS, Azure, GCE etc. Transparent encryption at the hardware level like SEM and SEV (or SGX and MEE in Intel's case) are an important part to harden the isolation between VMs.You know I get the worry about data security. It's a big thing. But really what is the potential for these types of attacks vs. an insider with access to the data leaving with it or some forward facing server with some unprotected share ala Sony. Basically most data is stolen in the most simple ways and not somehow intercepting a VM during internet copy (which most would be on an encrypted VPN anyways) or hacking an reading RAM to get encryption keys.
You wouldn't expect Intel to say anything else would you? If they weren't concerned they wouldn't have said anything.Intel's response to the server launch is indeed EPYC
"With our next-generation Xeon Scalable processors, we expect to continue offering the highest core and system performance versus AMD. AMD’s approach of stitching together 4 desktop die in a processor is expected to lead to inconsistent performance and other deployment complexities in the data center."
I guess that is what intel thinks about infinity fabric -
That reply stinks of desperationYou wouldn't expect Intel to say anything else would you? If they weren't concerned they wouldn't have said anything.
You wouldn't expect Intel to say anything else would you? If they weren't concerned they wouldn't have said anything.
Well, they did say they were estimates, not actual numbers . Besides they are fairly accurate estimates, putting aside the fact that GCC -O2 is not something you would use outside of debian packaging.
Though looking at how magically results (on Naples) for GCC -O2 are higher than every other benchmark posted today... Huh
Wait, wait, wait. There is no anywhere near 45% delta between GCC -O2 SPECint and ICC SPECint, in fact, it looks to be around 10% at most. Assuming (for AMD's sake) techreport screwed up as well with their chart (multiplying results by 0.575 instead of 0.7), Naples advantage suddenly looks much more.. uhm, how do i put it... humbler.
Sounds like FUD from Tom's. There will be plenty of systems out in the wild very soon, and it will be easy enough to see the real numbers. So I'll add a large grain of salt to Tom's thing.Tom's is saying AMD took Intel's officially published Spec numbers and subtracted 46%. Did their marketing team screw this up again? I have no idea why AMD wouldn't run whatever compile of Spec they wanted on real Intel hardware and use those numbers.
This smells of intentional gimping / willful misleading.
I get what it's supposed to do and in the end approve. Just saying that we go deeper and deeper into isolation, encryption, protective co-processors. When there is just so much low hanging fruit out there that 99.9999999% of the data security issues exist. Sometimes it seems like some of the investment spent in that tech could go to customer training of best practices and have a ton better affect of securing data.The issue is shared hosting/cloud computing where VMs are supposed to be isolated from each other. This regularly fails which is very worrying in a time where computation is regularly offloaded to AWS, Azure, GCE etc. Transparent encryption at the hardware level like SEM and SEV (or SGX and MEE in Intel's case) are an important part to harden the isolation between VMs.
Considering that numerous software and kernel bugs repeatedly broke VM isolation in the past years customer training wouldn't have helped securing any data there. Unless you want strictly keeping data local or on dedicated (non-virtualized) servers be part of said customer training.I get what it's supposed to do and in the end approve. Just saying that we go deeper and deeper into isolation, encryption, protective co-processors. When there is just so much low hanging fruit out there that 99.9999999% of the data security issues exist. Sometimes it seems like some of the investment spent in that tech could go to customer training of best practices and have a ton better affect of securing data.
That reply stinks of desperation
Intel knows AMD is right there in the ballpark (performance and perf./watt wise) with this current core. And AMD is not standing still, this is very first of Zen iterations ,many things to improve upon, fun times ahead
Sounds like FUD from Tom's. There will be plenty of systems out in the wild very soon, and it will be easy enough to see the real numbers. So I'll add a large grain of salt to Tom's thing.
- Based on estimated SPECint®_rate_base2006 scores. 2P Intel Xeon E5 scores other than E5-2699Av4 were derived by AMD from the following ICC compiler-based test results published at www.spec.org, multiplied by 0.575 to convert from the ICC compiler to the GCC-02 v6.1 compiler used for EPYC testing: E5-2698v4=1620, E5-2695v4=1440, E5-2680v4=1270, E5-2650v4=1000, E5-2640v4=865, E5-2630v4=814, and E5-2620v4=683. The following EPYC 2P scores are projections made by AMD labs (measured test data for these pre-production products is not currently available): EPYC 7301=845, EPYC 7281=760, and EPYC 7251=485. All other scores based on AMD internal testing. 2P E5-2699A v4 in Intel Server System R1208WT2GSR scored 943 on 6/5/2017, with Ubuntu 16.04, GCC-02 v6.3, 512 GB (16 x 32GB 2Rx4 PC4-2666 running at 2133), 1 x 500GB SSD. 2P EPYC 7601 in Supermicro AS-1123US-TR4 scored 1390, with Ubuntu 16.04, GCC-02 v6.3, 512 GB (16 x 32GB 2Rx4 PC4-2666 running at 2400), 1 x 500GB SSD. The following EPYC 2P scores tested using AMD’s “Ethanol” reference system with Ubuntu 16.04, GCC-02 v6.1, 512 GB (16x32GB 2Rx5 PC4-2667 running at 2400), 1 x 500GB SSD: EPYC 7551=1345, EPYC 7451=1218, EPYC 7401=1120, EPYC 7351=939. Pricing ranges based on Intel recommended customer pricing per ark.intel.com, and AMD 1Ku pricing.
Looks like AMD reduced Intel numbers do to the apparently unfair compiler advantage on Intel scores if TH is correct. I could see it proper to some degree but not 46%. Wacky.
Need real reviews.