By: Anon (no.delete@this.email.com), August 15, 2019 3:20 pm
Room: Moderated Discussions
hobold (hobold.delete@this.vectorizer.org) on August 15, 2019 7:10 am wrote:
> Anon (no.delete@this.email.com) on August 14, 2019 11:39 pm wrote:
> [...]
> > Having said that,
> > so far Ryzen has been good to me - however it is early days.
> >
> Early days for you; third generation for AMD and 3rd parties. I admit that Zen is far from established
> maturity; it is still too young for that. But consider that Intel has to completely rethink their processor
> design, too, because their mature line of products turned out to have a rather fundamental flaw.
>
> A few more teething problems will be waiting for us on either side.
> And then those metaphorical kids will be grown up before we know it.
Perhaps we have different definitions of 'early days' :) (I still an an AMD 80286/20 sitting on my shelf of memories FWIW).
My issue is not really the CPUs, in general I have had few problems with these, it tends to be in the surrounding support (chipset+drivers).
I admit the situations where these machines end up is very niche - doing reasonably hard real time work with required glitchless uptimes ranging from hours to months.
I will say that so far, Zen has been good, and I greatly hope things are improved.. My point was more than a great CPU is one thing.. drivers, support, and unfortunately history also matter.
Having said all of that, Intel are hardly blameless - their IGPUs have a long LONG history of disappointing. they pretty much gave up on promises after the first few generations, however they are getting a bit more bullish again now - we will see.
Intel chipsets (and generally software) however has usually had that wonderful feature of being 'boringly reliable'
I am just running crystalmark under windows 7 on an oldish install of a Kaveri machine with a software raid-1 of 1.5TB HDDs, sequential reads around 135MB/s, which is fine.
sequential write, around 28MB/s, less nice. 4k random is 0.5 and 0.7 MB/s (read/write).
Now, on a clean install, I know from experience that those random numbers will improve to 2-3 times these figures, and the write will approximately double. I never see this on Intel 'matrix raid'.
Yes, very very unscientific... but unpleasant.
Now, this is not a critical issue at all, and yes I could drop a single SSD in there and it would fly, however RAID-1 has saved our arses enough times (yes, even the fakeraid) that we simple always require it.
As a side not, I REALLY wish that these was a nice simple way of running SSD+HDD fakeraid where the read (and possibly write, configurable) loads were optimised for SSD throughput. Easy enough with a pricey raid controller. Raid-1 is a very cheap insurance against storage issues, purely from the point of view of having a faster recovery time than 'restore from backup' (and, of course, is never a replacement for backup). Intel and AMD seem to be pushing their 'cache on SSD' solutions for 'consumer' boxes, but not really addressing redundancy, which is a pity.
Anyway, none of it is really 'critical path', but such annoyances get in the way with some regularity.
> Anon (no.delete@this.email.com) on August 14, 2019 11:39 pm wrote:
> [...]
> > Having said that,
> > so far Ryzen has been good to me - however it is early days.
> >
> Early days for you; third generation for AMD and 3rd parties. I admit that Zen is far from established
> maturity; it is still too young for that. But consider that Intel has to completely rethink their processor
> design, too, because their mature line of products turned out to have a rather fundamental flaw.
>
> A few more teething problems will be waiting for us on either side.
> And then those metaphorical kids will be grown up before we know it.
Perhaps we have different definitions of 'early days' :) (I still an an AMD 80286/20 sitting on my shelf of memories FWIW).
My issue is not really the CPUs, in general I have had few problems with these, it tends to be in the surrounding support (chipset+drivers).
I admit the situations where these machines end up is very niche - doing reasonably hard real time work with required glitchless uptimes ranging from hours to months.
I will say that so far, Zen has been good, and I greatly hope things are improved.. My point was more than a great CPU is one thing.. drivers, support, and unfortunately history also matter.
Having said all of that, Intel are hardly blameless - their IGPUs have a long LONG history of disappointing. they pretty much gave up on promises after the first few generations, however they are getting a bit more bullish again now - we will see.
Intel chipsets (and generally software) however has usually had that wonderful feature of being 'boringly reliable'
I am just running crystalmark under windows 7 on an oldish install of a Kaveri machine with a software raid-1 of 1.5TB HDDs, sequential reads around 135MB/s, which is fine.
sequential write, around 28MB/s, less nice. 4k random is 0.5 and 0.7 MB/s (read/write).
Now, on a clean install, I know from experience that those random numbers will improve to 2-3 times these figures, and the write will approximately double. I never see this on Intel 'matrix raid'.
Yes, very very unscientific... but unpleasant.
Now, this is not a critical issue at all, and yes I could drop a single SSD in there and it would fly, however RAID-1 has saved our arses enough times (yes, even the fakeraid) that we simple always require it.
As a side not, I REALLY wish that these was a nice simple way of running SSD+HDD fakeraid where the read (and possibly write, configurable) loads were optimised for SSD throughput. Easy enough with a pricey raid controller. Raid-1 is a very cheap insurance against storage issues, purely from the point of view of having a faster recovery time than 'restore from backup' (and, of course, is never a replacement for backup). Intel and AMD seem to be pushing their 'cache on SSD' solutions for 'consumer' boxes, but not really addressing redundancy, which is a pity.
Anyway, none of it is really 'critical path', but such annoyances get in the way with some regularity.