Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
That is a big number in that market.I think they could even get even some few percentages more (e.g. 7%).
Q: If, as Borat's countryman claims, Intel has been shipping product for 10 months, wouldn't you expect it to be optimized, something that will be happening with Epyc in the coming months?
Is that due to some specific performance issues for the type of software you use?
We had the hardware for some time now, running tests, testing OS's, drivers etc...even posted a benchmark a while back...just because you didn't have access doesn't mean others didn't have access.
On March 2016's Broadwell-EP review, E5-2699v4 scored 33.3 on 403.gcc.
On the July 2017 Skylake-SP vs EPYC review, E5-2699v4 scores 23.7 on 403.gcc.
We knew that AT EPYC review (PUN intended) managed to get both Skylake and Broadwell Xeons slower than Haswell Xeons, but thanks to Ricardo B from RWT, we know the exact magnitude:
That is a huge 40% drop in performance.
Which hardware did you have? AMD's or Intel's? It is not clear from your post. Are you saying both have had hardware out there for 10 months so both are as optimized as you would expect 10 months out?We had the hardware for some time now, running tests, testing OS's, drivers etc...even posted a benchmark a while back...just because you didn't have access doesn't mean others didn't have access.
So are you implying that Intel's CPU is as optimized as it gets, while we are yet to see what further improvements AMD can make?
I am not sure how positive it is that Intel's CPUs have been out there for so long. Intel's Data Center group revenue is up only 4.5% YOY in latest quarter, down from 8% YOY in 2016. Clear deceleration. NVIDIA's data center revenues are up 186% YOY in latest quarter. Accelerating from up 145% for 2016.
1. AVX512 loads and stores don't trigger the AVX512 DVFS curve, only FMAs (and IMUL) do. They (and other low-power AVX512 instructions) trigger the AVX2 curve.
2. Skylake-SP actually scores about 1.38X higher than Naples on GCC. It's 1.58X higher on a per-core basis.
See https://www.spec.org/cpu2006/results/res2017q3/cpu2006-20170627-47389.html and https://www.spec.org/cpu2006/results/res2017q2/cpu2006-20170529-47127.html
So I'm pretty sure that something is horribly wrong in Johan's testing.
David
What's extremely interesting about this announcement vs. AMD's dog & pony show for Epyc is that the only numbers AMD could post were intentionally-fudged SPEC benchmarks based on their own in-house testing of old Xeon parts compared to their own in-house-tuned Epyc scores.
Intel just shows numbers from third party customers who are actually using Skylake in real-world workloads instead of intentionally gimping Epyc in some in-house bakeoff competition. Kind of shows you why they are most certainly #1 in the real world.
For that to be true, the cloud would somehow need less xeons to accomplish the same thing. No scratch that, to accomplish a lot more, given the growth in data, analytics etc. It is a lot more plausible explanation that computing power is shifting away from xeons.If you know the numbers I assume you have seen the reason but somehow ignored it. The reason is the enterprise with on premise is in rapid decline because they go cloud. So while cloud grows fast, enterprise with its decline drag the numbers down.
Nvidia is up so much because they add a new segment with AI and comes from a relatively small revenue position.
Which hardware did you have? AMD's or Intel's? It is not clear from your post. Are you saying both have had hardware out there for 10 months so both are as optimized as you would expect 10 months out?
For that to be true, the cloud would somehow need less xeons to accomplish the same thing. No scratch that, to accomplish a lot more, given the growth in data, analytics etc. It is a lot more plausible explanation that computing power is shifting away from xeons.
AT doesn't even try to be objective anymore. I would expect a so called journalist/writer for one of the most important tech websites to be at least somewhat less biased (looking at you Johan De Gellas).
Sure...
I think exactly the contrary. Zen will sell better in desktop, thanks to fanboys and people that can be easily fooled by the usual biased reviews, whereas enterprise people are more difficult to fool because many test things by themselves.
And the claim that Ryzen captured 20% market share is wrong. RyZen impact is about 5%.
We knew that AT EPYC review (PUN intended) managed to get both Skylake and Broadwell Xeons slower than Haswell Xeons, but thanks to Ricardo B from RWT, we know the exact magnitude:
That is a huge 40% drop in performance.
when results don't suit people they scratch and claw, seen this on both sides and exhibited in spades here. Real World TEchnologies, couldn't even have a real website, seems legit though.
results don't suit your inuendo, accuse them of bias.
If you could exhale helium, you could have a job at the state fare, blowing up balloons for kids.
when results don't suit people they scratch and claw, seen this on both sides and exhibited in spades here. Real World TEchnologies, couldn't even have a real website, seems legit though.
Correction: AT results don't agree with reality. AT tests EPYC and it is a lot faster than Xeon. Everyone else tests EPYC and it is a lot slower than Xeon.
I am not the one that fueled the Zen hype train up to stratospheric levels. Neither the one looking for weird excuses when the real product was released and didn't match the hype. Let us again know: how is going that silly 'theory' that current RyZen chips are engineering samples relabeled for consumer and that AMD didn't release real RyZen still?
Doesn't even deserve a detailed reply.
You got evidence shown and you reject it because its not what you wanted. No wonder you are up against the wall
What's next, more unicorns and fairy tales?
is it faster than xeon?
It is not. That is the reason why people is noticing the cooking/cheating made on the AT review, with Xeon performance crippled by huge amounts as 40% to make EPYC look competitive
https://hardforum.com/threads/intel...-lineup-leaked.1928015/page-4#post-1043107593
https://hardforum.com/threads/intel...-lineup-leaked.1928015/page-4#post-1043108651
The funny part is that AT marketing piece is useless, because datacenters test stuff by themselves and are rejecting EPYC massively. Intel already sold 500k+ units of Skylake Xeon and got three top500 systems based on Skylake.
Why dont you explain to us, why their review doesn't match past performance and is completely off compared to other reviews?
Are they simply too incompetent at AT?
when results don't suit people they scratch and claw, seen this on both sides and exhibited in spades here. Real World TEchnologies, couldn't even have a real website, seems legit though.
Yes, Anandtech will cheat for AMD when they could make more money punting intel, yup makes total sense.
Do you typically get a few servers to evaluate b4 you order? And, if you don't have AMD alternatives what do you evaluate against? The old generation?Nice strawman...cute, but boring.
We have had the new Intel CPU's for a while...it's very common for us to have hardware before it is offically launched...hence why I often laugh a rumors threads when we have had the actual hardware for some time, despite it not being officially "launched".
That is for Intel though.
AMD is very abscent...no servers containg AMD CPU's are planned, but we have plans for new servers with the new Intel CPU's.
My bet is that this time next year...we will have +5K servers with Intel CPU's...and zero with AMD hardware.
AMD has got an uphill battle to get into data-centers...and PR slides will do them no good at all...this is corporate-land, not PR FUD hype fanboy-land.
Like I said I think one of our vendors has a AMD file server on their roadmaps...a single server...guess how that shapes AMD's odds of getting into our datacenters?
I have just computed the amount of cheating on Anandtech review. Broadwell cores were crippled by 1.4x. Skylake scores were crippled by 2x.
I have just computed the amount of cheating on Anandtech review. Broadwell cores were crippled by 1.4x. Skylake scores were crippled by 2x.
Do you typically get a few servers to evaluate b4 you order?
And, if you don't have AMD alternatives what do you evaluate against? The old generation?
Limited lifespan you say but it seems that a lot of people are buying used servers and run them for years after the big data centers have sold them. Shouldn't the decision be made on the basis of performance over operating costs?Yup...physical servers have a limited lifespan, we are always phasing out a generation and embarking on a new generation across several server vendors.
Limited lifespan you say but it seems that a lot of people are buying used servers and run them for years after the big data centers have sold them. Shouldn't the decision be made on the basis of performance over operating costs?