Separate names with a comma.
Discussion in 'HardForum Tech News' started by cageymaru, May 28, 2019.
I stand corrected
Your emphasis is not misplaced, if they said that it means these processors don't have any mitigation patches enabled. They're probably being bench-marked against older Intel parts that do have mitigation's enabled. This is a bullshit PR stunt to attempt to take thunder out of AMD's announcement because Intel has shit and they know it. A PR stunt so that Intel does not lose massive market share prior to releasing competitive products. I think they're screwed in the short term. If in 2021, AMD isn't releasing a grand slam architectural advantage.... Intel will recoup all their market share losses in about six months. Assuming they release something that is good.
That's basically the point I fell short of making, thanks. If these parts 10nm are more useful for mobile, where Intel has been innovating significantly despite being stuck on the same 14nm node with the same Skylake cores, than they are for desktop where TDP advances aren't as nearly as appreciated by most consumers, and they have limited capacity, then it makes sense to only release mobile 10nm parts, for now.
Indeed... a 3% total hit with full mitigation vs 16-18%
Their mobile APUs need to make a pretty big leap. So many more factors come into play for laptops, especially the nicer ones and the business ones, that as close as AMD has come on the desktop they're still a world away on mobile.
And this is also highly dependent on workload. Ultrabook customers are far less likely to be running over-provisioned mission critical VMs .
[I run VMs on both my Windows and Linux installs all the time, but I'm weird...]
Don't disagree... AMDs mobile chips have been lacking. Raven Ridge was a big step forward considering before RR they where selling 8 or 9 year old tech shrunk down and full of suckage, I am not shocked they haven't picked up major OEM support.
Which is why I say if their 7nm generation takes RR and improves a good bit on it... and 7nm shows some real power advantages. Its possible they could start getting more then the standard low end laptop OEM contracts.
As good as RR was... you rarely find them in anything but bottom of the barrel machines. Chip can be 10x better then Intels but if ever OEM is pairing them with the cheapest ram, slowest drives, crappies screens with shit build quality. Its not shocking most people think of Raven Ridge as shit as well.
I am hoping AMD will start winning some at least mid range laptop contracts with Zen2. It people start seeing good quality laptops much cheaper then Intels high end that compete in all aspects that are important in mobile. I do think AMD could pick up some market share in consumer mobile pretty quick.
We'll see... to be honest I still really like Intel for mobile. Intels Linux support is great... and in general a well made Intel laptop makes a great mobile workstation.
They need to get into XPSs. A monolithic APU into the 13, and perhaps one based on the desktop part with a GPU + HBM module swapped for the second CPU die into the 15.
The fun part here is that CPU performance means really very little; it's getting everything else right that counts the most. The CPU just has to not suck and not consume battery power too fast. So if AMD focuses on power draw, I'd be estatic with a 12hr ultrabook that can also do 1080p medium settings when desired.
The latest mitigation for MDS... hits context switching insanely hard. And yes that mostly means multiple VMs running on the same core. However it also means multiple bits of software running on the same core. I have little doubt most laptop users these days multi task as much as anyone on a desktop.
There is of course a good hit on anything multi threaded in general as well.... even things like Java script take a 10% hit.
Now perhaps someone running a laptop for non security / customer info type work can get away with running mitigation=off (at least in Linux) still AMD is offering security without penalty.
Sunny Cove will have many fixes in silicon, and thus not require microcode updates and O.S./application patches to address. You assume grand conspiracy, yet the simple fact is that Intel has had time to develop hardware fixes for many of the security vulnerabilities discovered in the last 18 months or so.
You couldn't sell me land in waterworld
I like the disclaimer only intel optimised tests and platforms like that is "real" world and AMD get prodded for using cinebench which Intel also used but everything Intel does is totally legit.
18% faster than skylake I guess is a high water mark for intel.
Yes, yes we do- but we also have more than one core .
No! Don't disrupt the narrative!
Hard reallity is that not only does AMD not have a real foothold into the mobile market, they also don't have a part that's competitive- here's hoping that they release one sooner rather than later.
Well, it is. Marketshare and install base are a thing.
...when that's all that got released? Yes, yes they were 'prodded'.
I take it you have read a list of CVE's addressed in CPU design? where are they? Where is this fantastical list you claim. They have had 3 generations of CPU's to get those fixes in to suddenly expect them now is just delisional. Until they say "Here is our list security professionals go and verify!" I don't believe they have fixed anything
To be fair I'm more than a little bitter over this I own TWO server generations of 18 core CPU's from intel that currently have 1/2 of their logical CPU's turned OFF because of these SHITTY CPU's.
I'm really wanting an EPYC based host node to test with.
Uh, those three 'generations' were all Skylake. Sunny Cove is quite specifically not Skylake.
You have to understand that CPU development cycles are roughly three years (or more) long. There are stages where fixes are possible, but at some stage changes are no longer possible. Also keep in mind that Skylake-X, Kaby Lake, Kaby Lake-X, Coffee Lake, etc. were all variants of Skylake. Virtually nothing changed. Architectural refreshes are closely tied to the development of the original architecture and can only be modified so much. It was unreasonable to expect a bunch of in-silicon fixes for some of these issues during that time. Its also not unreasonable to expect at least some in-silicon fixes for Sunny Cove.
Intel had a choice. Release the newer processor models or cancel those models and make nothing new. Intel simply made a business decision and Intel chose the former option.
Here's the list for Whiskey Lake:
Here is an even more comprehensive list:
Marketshare is changing we don't have to pretend that isn't the case. The impact is felt in Intel's bottom line.
At this point Intel marketing is more cringy
Is changing. The needle has barely moved today. So yes, you should benchmark against what's out there today.
All marketing is cringy unless you accede to bias.
Nothing new about Intel's marketing.
This is the U in their FUD marketing mode.
They can't do the F too much as they have way more vulnerabilities and are shit for security.
Intel's is good at FUD and it will mitigate market losses.
Reading those I wasn't able to correlate which of those fixes was with the hyperthreading speculative execution vulnerability.
Rouge Data cache load one is:
CVE-2017-5754 (Meltdown, Variant 3, Rogue Data Cache Load) is a microprocessor vulnerability that allows an attacker to overcome all memory isolation mechanisms offered by the microprocessor by causing it to speculatively execute code out-of-order that loads inaccessible information which end up changing the cache state of the microarchitecture, thereby leaking information through side-channel timing analysis.
So that being fixed will be a big one for people like me that are thinking from a Datacenter standpoint.
Well ya that is part of the problem. Granted you do more context switching running lots on one core. However modern operating systems don't restrict programs to one core. We have also been screaming at software developers for years now to better thread their software... which means lots of context switching. Which at least right now hits most Intel chips pretty hard. I don't think many users are restricting their well threaded software to one core on purpose.
Keep in mind, we dont't yet have a list of mitigations/remediations for Sunny Cove-derived parts. However, you can see a clear progression as the newer chips on the list receive more fixes in-silicon than the older chips. Cascade Lake (launching now in server market and later this year for HEDT) receives hardware remediation of BTI aka Variant 2. It would make sense to expect further in-silicon fixes for e.g. Ice Lake-U/Y/S than say Coffee Lake Refresh.
I just wish intel didn't segment their instruction set so much. 18% faster than skylake (desktop without avx512) or skylake (server with avx512) or skylake-R (Crystal Well small form factor with edram) ?
By varying things from there the extrapolated outcome can probably sway another 10-15%.
It actually looks good especially if they equal AMDs decoder's flexibility.
What gets to consumers will always be at the whim of intel's artificial segmentation.
Either way my money is on a new incompatible chipset coming soon. ChaChing
18% eh.. So in terms of real-world impact the ST performance of my 6700 will be functionally outdated in... oh 5 more years.
Wait till AMD feels like they can command their own prices, if they ever get there again...
Sunny Cove will only come soldered in a laptop, so you're buying the motherboard regardless.
...by a 15w part in an ultrabook, i.e. Macbook Air...
They already have. We'd have a 16 core part if they weren't holding it in their pants.
Way to gloss over what I said and focus on an irrelevant implication. The point was artificial segmentation. For the desktop more often than not, with new intel chips comes new motherboards.
I think this image from anand exemplifies things. Lets play spot the difference.
I may be wrong but this segmentation is over 2 lanes and one letter. Is this graphic even necessary?
Edit: Oh they dropped ddr4 3200 support. For shame.
Might read the topic.
With AMD, you'll want a new motherboard and you have to research whether what you have will work, and how well. With Intel, it works.
The graphic is one in a series between the two different 'levels' of Sunny Cove.
Note that Intel is only competing with themselves here- AMD doesn't have a part that hang in this segment, as much as I personally wish they did.
My post was on topic. Especially with regards to the implications of the IPC. You might want to reread my post and see how far you brought it off topic.
Haha. Research is free. Nah better just to have to upgrade everything every time. You really believe in blue crystals and will pay top dollar for them
It's fucking stupid. Defend it to your own peril.
Perhaps you enjoy drinking Intel's Cool-aid and failed to read the actual, written, disclaimer that Intel themselves posted :
Yeah, I agree, Intel has had the time to fix their problems (if they gave a shit to), however, nothing short of a completely redesigned processor core with nothing tied to the old architecture will fix the issues their processors have. Grand Conspiracy my ass. There are only 1-2 websites with the balls to post the actual impact of the mitigations on Intel Processors. Hm... Wonder why? Because they would look like shit if they factored it all in.
That is what Ice Lake (which Sunny Cove uses) is.
The topic is about Sunny Cove.
Well, the implications of IPC to 15w class mobile CPUs are very different than desktop CPUs. Totally different world.
There's a difference between 'these things are designed to go together' and 'wait until someone else tries your very specific configuration and try to replicate their settings and pray'.
It's one in a series that shows the difference between the two parts, which are physically two different parts. Infer stupidity or otherwise as you wish, it's obviously not for you.
My contribution was: It is hard to nail down the 18% improvement due to intel's instruction segmentation. You took it off topic and continue to derail it.
The only factor that can alter IPC externally is communication (ex. edram). Otherwise IPC should be identical. Unless one artificially segments their instruction set. Oh wait, now we've come full circle. Maybe reread my original post and stop derailing it.
It is for me. I love laughing and commenting on stupid things. (double entendre)
Well, sure doesn't sound like the new architecture has fixed the issues inherent in it's predecessors. Especially if Intel has to mention disclaimers about not necessarily having all mitigations in place during testing. That's what I'm saying.
I'm hoping some website will do a generation to generation IPC comparison with:
1) Old generation (maybe a couple) IPC performance without any spectre/meltdown fixes
2) Old generation (same couple) IPC performance WITH all available spectre/meltdown/l1tf/mds/etc/etc mitigations fullly in place and enabled. This along with 1) will be interesting to see how much of a hit was taken due to cumulative mitigations.. I know for newer Archs it is a smaller hit, but older Archs' like Gen 5 Broadwell and Gen 6 took bigger hits.
3) Sunny Cove IPC performance WITH all available spectre/meltdown/l1tf/mds/etc/etc mitigations fullly in place and enabled.
So that we can see what all smoke n mirrors were used in the PR junk, but also to just truly see what the improvements are, and how well the in-silicon mitigations improve performance.
I will consider this, although I don't necessarily have examples of every architecture. I have Nahelem, Gulftown, Sandy Bridge, Ivy Bridge, Sandy Bridge-E, Devil's Canyon (I think) Skylake, Skylake-X (with no motherboard as its that shitty 7740X), Haswell, Haswell-E, Kaby Lake, and I think a Coffee Lake CPU. I'm missing Broadwell because I fried it.
It would be a ton of work though.
Yeah their 18% compared to Skylake could be because Skylake seems to be more affected than newer generation by mitigation from the quick search I did. So although at release they have same IPC, once patched it's another story.
Thank you ! You could test only key CPU (Like Skylake in this scenario, the current latest and greatest and maybe one older like Sandy/Ivy). That should be great ! With 2700X and newer AMD too for sure.
Oh yeah. That was good times. I was still rocking a celeron 333Mhz cranked up to 450 then. I was poor lol. Shortly after that I got an athlon xp. Wasn't too interested in Intel's nonsense. Not much has changed.
It's looking pretty damn good to me - https://browser.geekbench.com/v4/compute/4094642
That would be amazing Dan. I think Haswell, Skylake or Kabylake, would be a good sampling of the older architectures. I can see it would be a bit of work though, as you would need both old and new Bios's for the older boards, plus a hardrive with older install of windows to ensure it was pre-patch. Summer of 2017 build should be old enough. Pretty sure that by Fall/winter 2017 Microsoft had already slipped the first round of patches in, even before the vulnerabilities were made public.