4 ms·

> with no IPC improvement Unacceptable. It's now 5+ years since major improvements. And they released a new socket too, breaking upgrade paths. Meanwhile, my

> with no IPC improvement Unacceptable. It's now 5+ years since major improvements. And they released a new socket too, breaking upgrade paths. Meanwhile, my Ryzen 1600 is overclocked at 4ghz and running smoothly with 6 cores/12 threads. Intel better release something worthwhile soon. AMD is well into its comeback.

RandomInteger4 9y ago
This is a random tangent, but I've been meaning to build my first PC and was thinking of going with Ryzen. Do you by any chance run Linux or Linux dev tools on it and if so have you experienced hard crashing issues that I've been reading about?
- floatboth 9y ago
  Compiling a ton of stuff on my R7 1700, never encountered any issues. (But with LLVM on FreeBSD, not gcc on Linux.)
- dragontamer 9y ago
  From my understanding, a later version of Ryzen fixes the crashing issue that was common in the first-chip revision. The firmware updates should be fixed in any new motherboard you buy today.
- zanny 9y ago
  It was a manufacturing error they fixed on new CPUs after week 25. If you bought one now, it would work fine.
  - kogepathic 9y ago
    So is it a new stepping? If so that would be remarkably fast for a bug to be fixed in silicon. Otherwise why couldn't AMD issue a microcode update to fix the issue on older processors of the same stepping? I guess I'm not understanding how it could be a silicon fix so quickly, or how any other manufacturing defect on the package would cause such an issue.
- barrkel 9y ago
  Rumour has it that it's been fixed, and only affects early versions - http://techreport.com/news/32459/amd-ships-revised-ryzen-cpus-with-a-compile-bug-fix http://techreport.com/news/32459/amd-ships-revised-ryzen-cpu... It's hard to get official word though.
- TwoNineA 9y ago
  When you buy your processor, make sure it's built after week 25. On the processor you see something like UA 17??xyz ?? being the week number and xyz 3 letters like PGT or SUS. If the week number > 25, you are fine.
dogma1138 9y ago
AMD took over 5 years to release something competitive. While it’s a comeback for AMD you won’t see any major improvements like that again. With Coffee Lake and Skylake-X the only real advantage AMD has is price there isn’t a single architectural advantage and if people expect another 50% jump in IPC from AMD with Ryzen 2 they will be sorely disappointed. Unless AMD or Intel will drastically change their architecture the improvement will come from clock speed and corecount with some minor improvements with specific instructions from time to time. If you look at the history of CPUs youll see a trend that started nearly 10 years ago where the additional transistor count doesn’t translates to actual general computing improvements in CPUs. As far as upgrades go it’s irrelvant for the most part. While AM4 will be supported for 2 more years to get the most out of the new CPUs new chipsets and motherboards will be required. With AM3 and FM3 we’ve seen this already where old motherboards could not run newer CPUs especially in the lower end of the spectrum and in some cases you would get issues like throttling due to lack of proper power delivery in the socket, memory compatibility issues and a plethora of other issues. And while the 1600 is a great value 4.0ghz stable isn’t reachable on every chip 3.7-3.8 is a more realistic scenario, and in either case I wouldn’t compare it with the 8700K in pure performance I’m holding on 5.2 now and likely will push it to 5.4 when I’ll delied it, at those frequencies it’s faster than even the 1800/1700 at 4-4.1ghz.
cfallin 9y ago
> Unacceptable. It's now 5+ years since major improvements. It's disappointing that the curve doesn't go up-and-to-the-right more, but "unacceptable" reads a little harsh to me -- I doubt the lack of big IPC gains for single-core workloads was a voluntary choice. Intel (and ARM, and Apple, and AMD, and ...) are full of engineers trying to come up with clever tricks in the instruction scheduler, or branch predictor, or prefetcher, or cache replacement policy or whatever, that obtain fractions of a percent IPC improvement. It's really hard to improve on something that's been optimized to death for the past ~25 years (since the first CPUs with out-of-order execution), and they'll take whatever they can get. I do agree that the competition with AMD will certainly be interesting to watch, certainly...
- Dylan16807 9y ago
  > I doubt the lack of big IPC gains for single-core workloads was a voluntary choice. It seems to me like they voluntarily focused on power consumption over performance. It's a valid choice but it wasn't forced on them, and raw performance is more important for desktops.
  - dogma1138 9y ago
    >It seems to me like they voluntarily focused on power consumption over performance. It's a valid choice but it wasn't forced on them, and raw performance is more important for desktop Because people wanted thinner and lighter laptops. The Kabbylake-R "8th gen" mobile CPUs are absolutly amazing, 4 cores at over 3ghz with 15W TDP is absolutly bonkers: https://ark.intel.com/products/124967/Intel-Core-i5-8250U-Processor-6M-Cache-up-to-3_40-GHz https://ark.intel.com/products/124967/Intel-Core-i5-8250U-Pr... It's also important to state that it's not like AMD did anything better. Zen is a very attractive architecture but IPC wise it's slightly behind at about the same power envelope as Intel (both AMD and Intel measure TDP differently but the power consumption of the CPU per core is more or less identical for the same frequency (with intel slightly lower still) for non AVX loads). Laptop sale outnumber desktop sales, and it's not like high end consumer and enthusiast CPUs for the desktop are that power conservative cooling 100-200W of TDP isn't easy even with water, I'm running a full custom loop on all my computers and cooling is still a problem when reaching high clocks at full stress. So while Intel could've pushed Coffee Lake 1-2 years earlier with some sacrifice (especially in profit margins) as far as IPC improvements I really doubt they had any choice even if they would crank their CPUs a notch or two when power consumption goes the best the would've gotten is slightly higher clocks. As far as the core count goes, yes Intel was lazy, but for AMD that was literally the only option a 4 core Ryzen wouldn't interest anyone since AMD can't get their process above 4.0ghz reliably even on low core count CPUs, on the IPC front it's behind, branch prediction is still quite behind Intel (one of the reasons why SMT give slightly higher relative advantage on AMD CPUs is all them stalls). AMD has been operating at a near constant operational loss for the past 4-5 years, their only profitable quarters were when the console revenue cheque came in and in the past 4-5 years they maybe had 2, they had to pay MSFT back a truck of money because GloFo cocked up their 20nm process which caused Xbox One S to be delayed and had to cancel Skybridge and a few other projects. Intel has been rocking on at near 60% gross margins for years now, and while it's easy to paint them greedy if I was a corporation I wouldn't likely do anything differently. The 8th gen is quite a big deal when it comes to Intel vs AMD more that some people would care to admit. Intel has a the fastest 6 core CPU out there with a built in GPU, it managed to squeeze 4 cores into a 15W part on the mobile side, this time around both mobile i5's and i7's will come with 4 cores no more 2 core CPUs for the mainstream and these are the U parts the ones that were traditionally dual core even for i7's, the HQ parts will likely bring 6 cores to mobile and with 9th gen CPUs launching for the ultra low voltage platform later this year AMD will have even more of an uphill battle where it counts laptops and other mobile devices. AMD needs to have a viable product for the mobile market, and with raven ridge coming for mobile this year they can only pull 4 cores with their APUs and how power efficient is it going to be no one still knows.

RandomInteger4 9y ago

This is a random tangent, but I've been meaning to build my first PC and was thinking of going with Ryzen. Do you by any chance run Linux or Linux dev tools on it and if so have you experienced hard crashing issues that I've been reading about?

floatboth 9y ago
Compiling a ton of stuff on my R7 1700, never encountered any issues. (But with LLVM on FreeBSD, not gcc on Linux.)
dragontamer 9y ago
From my understanding, a later version of Ryzen fixes the crashing issue that was common in the first-chip revision. The firmware updates should be fixed in any new motherboard you buy today.
zanny 9y ago
It was a manufacturing error they fixed on new CPUs after week 25. If you bought one now, it would work fine.
- kogepathic 9y ago
  So is it a new stepping? If so that would be remarkably fast for a bug to be fixed in silicon. Otherwise why couldn't AMD issue a microcode update to fix the issue on older processors of the same stepping? I guess I'm not understanding how it could be a silicon fix so quickly, or how any other manufacturing defect on the package would cause such an issue.
barrkel 9y ago
Rumour has it that it's been fixed, and only affects early versions - http://techreport.com/news/32459/amd-ships-revised-ryzen-cpus-with-a-compile-bug-fix http://techreport.com/news/32459/amd-ships-revised-ryzen-cpu... It's hard to get official word though.
TwoNineA 9y ago
When you buy your processor, make sure it's built after week 25. On the processor you see something like UA 17??xyz ?? being the week number and xyz 3 letters like PGT or SUS. If the week number > 25, you are fine.

dogma1138 9y ago

AMD took over 5 years to release something competitive. While it’s a comeback for AMD you won’t see any major improvements like that again. With Coffee Lake and Skylake-X the only real advantage AMD has is price there isn’t a single architectural advantage and if people expect another 50% jump in IPC from AMD with Ryzen 2 they will be sorely disappointed. Unless AMD or Intel will drastically change their architecture the improvement will come from clock speed and corecount with some minor improvements with specific instructions from time to time. If you look at the history of CPUs youll see a trend that started nearly 10 years ago where the additional transistor count doesn’t translates to actual general computing improvements in CPUs. As far as upgrades go it’s irrelvant for the most part. While AM4 will be supported for 2 more years to get the most out of the new CPUs new chipsets and motherboards will be required. With AM3 and FM3 we’ve seen this already where old motherboards could not run newer CPUs especially in the lower end of the spectrum and in some cases you would get issues like throttling due to lack of proper power delivery in the socket, memory compatibility issues and a plethora of other issues. And while the 1600 is a great value 4.0ghz stable isn’t reachable on every chip 3.7-3.8 is a more realistic scenario, and in either case I wouldn’t compare it with the 8700K in pure performance I’m holding on 5.2 now and likely will push it to 5.4 when I’ll delied it, at those frequencies it’s faster than even the 1800/1700 at 4-4.1ghz.

cfallin 9y ago

> Unacceptable. It's now 5+ years since major improvements. It's disappointing that the curve doesn't go up-and-to-the-right more, but "unacceptable" reads a little harsh to me -- I doubt the lack of big IPC gains for single-core workloads was a voluntary choice. Intel (and ARM, and Apple, and AMD, and ...) are full of engineers trying to come up with clever tricks in the instruction scheduler, or branch predictor, or prefetcher, or cache replacement policy or whatever, that obtain fractions of a percent IPC improvement. It's really hard to improve on something that's been optimized to death for the past ~25 years (since the first CPUs with out-of-order execution), and they'll take whatever they can get. I do agree that the competition with AMD will certainly be interesting to watch, certainly...

Dylan16807 9y ago
> I doubt the lack of big IPC gains for single-core workloads was a voluntary choice. It seems to me like they voluntarily focused on power consumption over performance. It's a valid choice but it wasn't forced on them, and raw performance is more important for desktops.
- dogma1138 9y ago
  >It seems to me like they voluntarily focused on power consumption over performance. It's a valid choice but it wasn't forced on them, and raw performance is more important for desktop Because people wanted thinner and lighter laptops. The Kabbylake-R "8th gen" mobile CPUs are absolutly amazing, 4 cores at over 3ghz with 15W TDP is absolutly bonkers: https://ark.intel.com/products/124967/Intel-Core-i5-8250U-Processor-6M-Cache-up-to-3_40-GHz https://ark.intel.com/products/124967/Intel-Core-i5-8250U-Pr... It's also important to state that it's not like AMD did anything better. Zen is a very attractive architecture but IPC wise it's slightly behind at about the same power envelope as Intel (both AMD and Intel measure TDP differently but the power consumption of the CPU per core is more or less identical for the same frequency (with intel slightly lower still) for non AVX loads). Laptop sale outnumber desktop sales, and it's not like high end consumer and enthusiast CPUs for the desktop are that power conservative cooling 100-200W of TDP isn't easy even with water, I'm running a full custom loop on all my computers and cooling is still a problem when reaching high clocks at full stress. So while Intel could've pushed Coffee Lake 1-2 years earlier with some sacrifice (especially in profit margins) as far as IPC improvements I really doubt they had any choice even if they would crank their CPUs a notch or two when power consumption goes the best the would've gotten is slightly higher clocks. As far as the core count goes, yes Intel was lazy, but for AMD that was literally the only option a 4 core Ryzen wouldn't interest anyone since AMD can't get their process above 4.0ghz reliably even on low core count CPUs, on the IPC front it's behind, branch prediction is still quite behind Intel (one of the reasons why SMT give slightly higher relative advantage on AMD CPUs is all them stalls). AMD has been operating at a near constant operational loss for the past 4-5 years, their only profitable quarters were when the console revenue cheque came in and in the past 4-5 years they maybe had 2, they had to pay MSFT back a truck of money because GloFo cocked up their 20nm process which caused Xbox One S to be delayed and had to cancel Skybridge and a few other projects. Intel has been rocking on at near 60% gross margins for years now, and while it's easy to paint them greedy if I was a corporation I wouldn't likely do anything differently. The 8th gen is quite a big deal when it comes to Intel vs AMD more that some people would care to admit. Intel has a the fastest 6 core CPU out there with a built in GPU, it managed to squeeze 4 cores into a 15W part on the mobile side, this time around both mobile i5's and i7's will come with 4 cores no more 2 core CPUs for the mainstream and these are the U parts the ones that were traditionally dual core even for i7's, the HQ parts will likely bring 6 cores to mobile and with 9th gen CPUs launching for the ultra low voltage platform later this year AMD will have even more of an uphill battle where it counts laptops and other mobile devices. AMD needs to have a viable product for the mobile market, and with raven ridge coming for mobile this year they can only pull 4 cores with their APUs and how power efficient is it going to be no one still knows.