New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
This disqualified you entirely now. AMD invented x64 as AMD64. Intel licenses it.
Funny. So you really want to try to put such a statement (basically half-cooked history) above technical facts? Funny.
Uhm ... just one example - > https://www.cpubenchmark.net/singleThread.html
In case you prefer to compare server CPUs only: a E5-2695 v4, 2.1 GHz, by no means a particularly fast Xeon, beats e.g. the 7302P and 7502P Epycs.
It may be (and obviously is) surprising to some but one can actually like AMD Zen (as I do. This is written on a Ryzen system) and still not consider intel processors sh_t.
I never said Intel does not beat AMD single threaded. They do, by a very minimal margin. At 3x power use.
You're not testing the processor, you're testing the configured virtual CPU and storage assigned by the hypervisor, things you don't know and can't know about since its configured by the providers and not publicly known. And then you also said "provided that a server is not brutally oversold", (wtf are you comparing to your home server that can't be oversold???) something again, you don't and can't know about. You can't control so many of the critical variables, it's useless.
You're not used to having peers review your work because many people here are telling you the same things that are just really obvious.
FYI I did not test storage, but hey why should you care about the facts ...
(a) the performance of a (not brutally oversold) VM isn't random, it's within a relatively small window related to the processors performance. But hey, why should an obvious lack of relevant knowledge stop you ...
(b) I can and do know about the processor and performance and my benchmark can in fact be parameterized to even compensate overselling to some degree.
(c) concrete example: The Xeon system just says it's qemu CPU but the provider mentioned some processor models and I could verify that it was indeed one of those and even which one.
Thank you, that's a particularly funny assertion, also because it comes from someone who knows virtually nothing about me and my professional life but also because, hint, wink, wink, nudge, nudge, formal verification is about the hardest review of my work, way harder than peer review. Finally it's also funny because you seem to know very little about actual real peer reviewing.
And btw you are wrong to assume that I'm not used to people telling me their views. I am used to it but I'm also used to the fact that the noisiest ones (hint, hint) usually also are the most clueless ones. You seem to just not get it, so let me spell it out clearly: There is the right to tell an opinion and there is the qualification to form an opinion. Unfortunately the law and the society do not demand a sensible link between the two.
Or, in other words: I'm sure there are fields were you are qualified to form and tell an opinion and I'd listen attentatively - but IT is definitely not one of those fields in your case and you'd do everyone - incl. yourself! - a favour by just keeping your "pearls of wisdom" to yourself.
Single thread benchmarks USUALLY perform better in desktop environments due to "turbo" settings. At times the difference is not even in the same ballpark.
On a server which we should reasonably assume is under OTHER load, the numbers should be similar.
Yes, the comparison is flawed, still, IMO it does make some points.
I fully agree AMD is AGAIN ahead, this time will last more than a year I suppose and it is not only the wafer technology. When you are the underdog for decades, you continuously strive for better performance and, eventually, money can no longer buy determination and business intelligence as well as customers.
I think Intel is not dead, they will do some moves eventually, but time is ticking away, AMD can now rise cheaper and faster capital, the old policy of drowning them in secret deals and discounts is finally failing for Intel, if there is no radical change within a year, real one, not a press announcement, Intel might be "also featuring" in the market.
But, what is really interesting, is the ARM race(s). I know it was a long time coming, but all market winds are into ARM sails. There is that time when tools are here for cross compiling, when energy frugal "things" are the hype and when licensing and externalization/specialization beats "in-house everything".
I largely agree, minus the
Look at the big manufacturers. One still can have a hard time to find systems based on the most current AMD processors while Xeon based servers and blades still seem to largely dominate the market. Hell, even the taiwanese companies needed lots of time to build Zen systems and quite a few giants still hardly or not at all build Zen systems.
Plus I think that we here at LET focus mainly on the data center market, but as a friend at a large client recently reminded me: "We have 2 or 3 (public facing) servers in a colo - but we have hundreds here in our own DC". And it seems that intel still dominates that market.
Yep. Pretty much all giant players (like Amazon) now either bet on Arm or even have their own Arm processors in their DCs. And quite a few of those "Monster Arms" (e.g. with 80 and soon 96 and more) cores are pretty close to x86 performance or even surpass it, at least in the field they were designed and created for (typ. DC).
That said, Arm has amassed lots and lots of baggage too and is a rather yester-decade Architecture too. THE coming new king IMO is Risc-V. Unfortunately it'll still take a couple of years but once there are 64+ cores Risc-V processors available X86 and Arm (minus MCUs) will have a very hard time to put it mildly.
Yes it does, due to old deals, discounts, unsold inventory and whatnot. When the rest of the pack has to actually manufacture their product, Intel has it sitting it in their wh from last year and even 2 years ago. It is similar to what happens in the car industry, the huge demand for newer generation chips forces them to use old ones, sometimes offering a guarantee they will upgrade it later for free, once they have the necessary parts.
I was saying this can last at most for one more year. If Intel does not make a dramatic move in 12 months, the juggernaut will have the things that prop it up now running against it, inventories still have a market now, they wont in 12 months so there will be a major write-off, huge profit hit, stock tumble, cash disappear and such a move will no longer be possible. It will be only down from there at least for a few years.
This is another reason that we can't draw any solid conclusions from this -- clock boost behaviour is controlled by the host. We don't know if they're running on a power save or performance profile, or if other cores are currently running on boost - most CPUs can only boost a small number of cores at once.
Another point is this seems to measure some form of memory access speeds (not clear what ProcMem SC actually measures)? If the host numa configuration is not optimal this would have an impact - also not visible to the VM. It's also more likely that an AMD VM would have problems compared to Intel, as AMD usually has more (smaller) numa nodes than intel does - there's no way to tell if a host has configured for that.
@jsg in the academic world this would be a starting point for further research, but no conclusions can be drawn from the data we've got right now.
Disclaimer: $AMD gang, bought in at $22.
I hope you are joking there
@jackb
This is the performance a users application would see too. Yes, the memory influences the result too as do diverse processor settings, but (a) no matter, that's the same for users applications, (b) my benchmark is not the only one showing some Xeons having better single core performance.
For those particular machines at those particular hosts at the load level present when you did your test - there's no way to tell if this is indicative for the hardware in general.
True, but other benchmarks about Xeon Vs Zen aren't what this thread is about.
i like william and jsg so i will refrain from commentary
Did anyone see the AMD keynote yesterday? Intel is dead now pretty much.
Not sure what I'm missing? Anand isnt giving any good info either.
Never mind. Found it
Careful there. Dead should not be predicted but checked post factum. Especially when the presumably (virtually dead) has billions of $, quite many fabs, and a very large and loyal (or "loyal") customer base.
Yea, obviously Intel will not just die, they have insane amounts of assets alone (like a multi billion dollar campus in Israel) but the tech gab between them and AMD gets bigger and bigger....
I did not expect what happened with Zen but its just what happened to AMD before also in reverse
It won't die... There are large-scale companies that work closely with Intel, and some that are in a contract/commitment for new hardware (from what I've heard from a local campus). Intel's done too much in the past 2 decades to just die off. Neither CPU is perfect, each has its major flaws.
First: I apologize because I just saw that I had forgotten an all important 'not' (...should not be predicted...). Corrected it. Sorry.
Re your comment: Yes, actually I think that the cards aren't bad for us that both will prosper, which would be great.
I would not bet on it. Intel has an extremely strong brand and a loyal following. A lot of big corporations and datacenters will run Intel for years to come no matter how superior AMD gets, simply because "it works™".
AMD may take big marketshares in the lowend and the more agile businesses but the old, slow moving dc's and corporations will stick to Intel like glue. This inertia will of course not last forever, but probably long enough to give Intel time to catch up tech-wise.
https://azure.microsoft.com/en-us/partners/directory/amd-corporation/
I have come across issues that until this days are not resolved. Intel and AMD even if they both follow the same architecture, they implement their instructions differently.
For example, gaming industry relies heavily on fpu math (floating point), but in certain situations, the result of such an fpu operation might be different from amd to intel.
SSE2 implementation is different and amd lacks some instructions, therefor resulting in broken games/applications. SSE2 is the default fpu math library when compiling x64 apps and the fpu x87( fld, fadd, fsub) instructions can be used only with x86 applications.
Example of mass effect being broken on amd cpus ( note that it's not related to graphics cards but to cpu math embedded into the hardware)
https://cookieplmonster.github.io/2020/07/19/silentpatch-mass-effect/
Now imagine if such thing will happen in the financial sector, where they also use new instructions to make their code faster. Amd has some issues to solve before they can be integrated with enterprise.
I'm not factually wrong, it's a true statement. Your response should have been that storage was irrelevant to your specific testing.
BS.
BS.
So they all ran the same hypervisor with all the latest CPU patches installed, same OS in the VM, same kernel version as well?
There is irony, here.
Ad hominem, got it.
Your bias experience is what you call "facts"? I'm loud and noisy in that little post with specific issues pointed out?
No, it's very clear from your writing style and hostile responses that you don't regularly get feedback.
Irony, again.
And you are trying to say what?
TL;DR: Intel is not really dead, but is set for a long disease and perhaps even longer recovery.