|
I'd wait to see for similar benchmarks before classifying that as not an anomaly, but this sort of backslide is nothing new (take Haswell-E vs Broadwell-E for instance, and that has the excuse of being a new process) and if I had to take a stab at the reason it'd be Kaby Lake's more aggressive throttling than Skylake acting up when it's not supposed to at all, ie desktop chips
|
#
?
Jan 4, 2017 20:05
|
|
|
|
| # ? May 3, 2024 04:07 |
|
|
The best way to look at Kaby Lake is thinking of it as Intel handing AMD a big, and sorely needed, opportunity. AMD is well known to make good use of those historically, of course 
|
|
#
?
Jan 4, 2017 20:14
|
|
|
SwissArmyDruid posted:...the gently caress? I'd agree with Anime Schoolgirl's assessment that Kaby Lake (and any future architectures, most likely) aren't really intended for desktops. Just like with Haswell, Broadwell, and Skylake, mediocre desktop gains are what enthusiasts notice and complain about while the 4.5W and 15W chips make pretty big strides forward. The clock speed increases are small on desktop chips and pretty drat big on the ultra low TDP parts!
|
|
#
?
Jan 4, 2017 20:21
|
|
|
Kaby Lake seems to be a minor respin of Skylake, and would've warranted maybe a new stepping back in the '90s. Now, they're trying to claim it's a whole new SKU because the designs got so huge that even bugfixing them takes a couple thousand engineers like a year.
|
|
#
?
Jan 4, 2017 22:42
|
|
|
penus penus penus posted:The best way to look at Kaby Lake is thinking of it as Intel handing AMD a big, and sorely needed, opportunity.
|
|
#
?
Jan 4, 2017 23:43
|
|
|
lDDQD posted:Kaby Lake seems to be a minor respin of Skylake, and would've warranted maybe a new stepping back in the '90s. Now, they're trying to claim it's a whole new SKU because the designs got so huge that even bugfixing them takes a couple thousand engineers like a year. Well, you are right - Kaby Lake is the first physical manifestation of Intel's formal acknowledgment that development cycles and product cycles are diverging because large improvements (relative to historical ones) on a yearly scale are not feasible. They already kind of did it once since desktops got Devil's Canyon and not much of Broadwell, but this time they're getting rid of tick-tock entirely instead of just adjusting it. The question for me is why so many people seem to think a ~5% improvement over last year is news or a surprise instead of status quo for desktops for ~5 years now. A lot of comments I've seen are on the lack of IPC increase in particular, but given that their strategy this time was to focus on power dissipation improvements to enable higher clock speeds that also shouldn't be a surprise. e: Really, think about Northwood->Prescott. If I recall correctly that revision increased power consumption and at least in some workloads decreased IPC all in the name of higher clocks. At least this time they can get them with a process refinement instead of adding lots of stages to the pipeline. Eletriarnation fucked around with this message at 23:48 on Jan 4, 2017 |
|
#
?
Jan 4, 2017 23:43
|
|
|
Eletriarnation posted:e: Really, think about Northwood->Prescott. No. You can't make me.
|
|
#
?
Jan 5, 2017 05:14
|
|
|
Also while Prescott was a hot mess, wasn't it also the move to x64 for the P4? So you got a hotter potentially slower chip, but now you could go above 3G of ram.  Which was huge when Vista dropped and it loved Ram. Ran ok on 1.5G+ for the most part (Threw it on a 1st Gen Netbook with that and it ran "fine" for what an Atom could provide), but run it on 1G and it was just passable. Try and run it on 512 and it was futile as it never stopped thrashing.
|
|
#
?
Jan 5, 2017 06:35
|
|
|
Eletriarnation posted:Well, you are right - Kaby Lake is the first physical manifestation of Intel's formal acknowledgment that development cycles and product cycles are diverging because large improvements (relative to historical ones) on a yearly scale are not feasible. They already kind of did it once since desktops got Devil's Canyon and not much of Broadwell, but this time they're getting rid of tick-tock entirely instead of just adjusting it. Actually if we exclude the P4, the overall IPC gains from 1995-2006 pre Conroe (that's including a massive 90% gain from P5 to P6) vs 2006-current is roughly the same. The good old days was mainly due to brute forcing clocks and TDP upwards.
|
|
#
?
Jan 5, 2017 06:55
|
|
|
EdEddnEddy posted:Also while Prescott was a hot mess, wasn't it also the move to x64 for the P4? Yes, the 600-series Socket 775 Prescott chips supported it and the 500s as well as all of the Socket 478 models didn't. It didn't mean much yet, though. 4+GB of RAM was still a ton at that point, I built a Nehalem system at the end of 2008 with Vista and still only put 3x1GB in it to start. Palladium posted:Actually if we exclude the P4, the overall IPC gains from 1995-2006 pre Conroe (that's including a massive 90% gain from P5 to P6) vs 2006-current is roughly the same. The good old days was mainly due to brute forcing clocks and TDP upwards. Yeah, a lot of the vaunted improvement of those years was made by mostly ignoring performance-per-watt since we went from Pentiums that barely needed active cooling to ~100W with Prescott. Still, it was a huge improvement even considering IPW to increase clock speeds from 66MHz with the P5 (that is, original Pentiums) to 2GHz+ with Pentium M and it's just not possible to do that again within any power budget which is a lot of what people are griping about. Eletriarnation fucked around with this message at 16:40 on Jan 5, 2017 |
|
#
?
Jan 5, 2017 15:53
|
|
|
edit: whoops nvm
|
|
#
?
Jan 5, 2017 16:29
|
|
|
Apparently the MSI x370 mobos are going to be released at the end of February.
Arzachel fucked around with this message at 17:48 on Jan 5, 2017 |
|
#
?
Jan 5, 2017 17:40
|
|
|
SwissArmyDruid posted:So much wasted space around that socket, man. If you look closely at the surface of the board it's ALL circuit traces in the open space. Traces to the RAM and traces to the PCIe slots.
|
|
#
?
Jan 5, 2017 19:41
|
|
|
Arzachel posted:Apparently the MSI x370 mobos are going to be released at the end of February. So we should be expecting some kind of event in mid February? What's the usual order of board releases, low to high or high to low?
|
|
#
?
Jan 5, 2017 19:43
|
|
|
uh motherboards with no cpu? Are they going to hard confirm launch dates today?
|
|
#
?
Jan 5, 2017 20:05
|
|
|
Bristol Ridge exists
|
|
#
?
Jan 5, 2017 20:10
|
|
|
FaustianQ posted:So we should be expecting some kind of event in mid February? What's the usual order of board releases, low to high or high to low? Risky Bisquick posted:Bristol Ridge exists A300/A320/B300/B350 Are the low to mid range chipsets and were shown off back in October, X300/X370 are the high end and the guess is that they likely will be released alongside Zen.
|
|
#
?
Jan 5, 2017 21:53
|
|
|
Prescription Combs posted:If you look closely at the surface of the board it's ALL circuit traces in the open space. Traces to the RAM and traces to the PCIe slots. ....uh. Begging your pardon, what did you *think* connected the RAM and PCIe slots to the processor slot?
|
|
#
?
Jan 5, 2017 22:00
|
|
|
Risky Bisquick posted:Bristol Ridge exists It better be cheap too, like AM1 replacement cheap or I'll laugh in AMD's face if they charge more than 100$ for the best SKU.
|
|
#
?
Jan 5, 2017 23:39
|
|
|
I didn't notice somehow that Kaby @ 5ghz loses to the 6700K @ 4.2ghz in the bench I posted earlier. Wowzers. Whatever Intel counters with, hopefully AMD will have revenue to do the R&D to get HBM integration at a competitive cost for the next round. I don't think any chip can compete against something like that as long as they're stuck with the (relative) latency and throughput limitations of DDR4.
|
|
#
?
Jan 6, 2017 00:08
|
|
|
Bloody Antlers posted:I didn't notice somehow that Kaby @ 5ghz loses to the 6700K @ 4.2ghz in the bench I posted earlier. Wowzers. The bench you're talking about is using the iGPU and that didn't get changed so all of the results are random noise, they say this right below the graphs. In general Kaby Lake/Skylake IPC is identical so virtually all the differences are proportional to clock speed. There does appear to be something odd going on with the overclocked Kaby Lake results in their Geekbench Memory test, though. Eletriarnation fucked around with this message at 00:16 on Jan 6, 2017 |
|
#
?
Jan 6, 2017 00:13
|
|
|
Good news everyone, Ryzen was running at 3.6Ghz base, 3.9Ghz turbo at CES, apparently F4 stepping? And newer samples are hitting 4.0ghz boost, all under the 95W TDP. http://wccftech.com/ryzen-ces-2017-3-6-ghz-base-clock-f4-stepping-4-0-ghz/ 4.4 to 4.8Ghz is looking rather attainable (OCing of course), so it's really looking like it's all riding on price now. EDIT: Something a little better than WCCFtech - https://mobile.twitter.com/CPCHardware/status/817044837358780416?ref_src=twsrc%5Etfw EmpyreanFlux fucked around with this message at 00:59 on Jan 6, 2017 |
|
#
?
Jan 6, 2017 00:48
|
|
|
Would I buy that for $300? $350? If it truly is Haswell performance and 8c? ......probably.
|
|
#
?
Jan 6, 2017 01:28
|
|
|
Potato Salad posted:Would I buy that for $300? $350? If it truly is Haswell performance and 8c? ...yeah. I get to buy a mid tier video editing box at work later this year and if AMD prices are reasonable I'll just put some 8c monstrosity in there instead of the usual i7.
|
|
#
?
Jan 6, 2017 01:40
|
|
I CANNOT EJACULATE WITHOUT SEEING NATIVE AMERICANS BRUTALISED!
|
Potato Salad posted:Would I buy that for $300? $350? If it truly is Haswell performance and 8c? 6900Ks are over $1000 MSRP and base clocked at 3.2. If they're selling something better than a 6900K, it's not going to be under $500.
|
|
#
?
Jan 6, 2017 01:47
|
|
|
FaustianQ posted:Good news everyone, Ryzen was running at 3.6Ghz base, 3.9Ghz turbo at CES, apparently F4 stepping? And newer samples are hitting 4.0ghz boost, all under the 95W TDP. 4GHz 8 core out of the box with Broadwell like IPC? If they price it right, they've got a real winner on their hands...
|
|
#
?
Jan 6, 2017 02:03
|
|
|
Careful on that IPC guess https://twitter.com/CPCHardware/status/817130693759303680
|
|
#
?
Jan 6, 2017 03:47
|
|
|
Maybe someone already mentioned it and I missed it, but it doesn't look like Ryzen is coming out in Q1 2017
|
|
#
?
Jan 6, 2017 03:50
|
|
|
I don't get AMD's secrecy one bit. There's no point not showing their hand now, paper launch/ES preview/whatever, when they have gently caress-nothing to lose by this point. Heck, even most of my local DIY stores has already stopped carrying AMD CPUs.
|
|
#
?
Jan 6, 2017 04:15
|
|
|
I actually kinda see what they might be doing. If Vega is as good as some people think it's going to be, there might be a pretty big value to holding back Zen and either tuning it as much as they can or just waiting to launch them both at the same time, if they think that they can simultaneously beat the top Intel CPUs and the top NVIDIA GPUs in one big event, that'd be bigger for their momentum than putting them out separately.
|
|
#
?
Jan 6, 2017 04:22
|
|
|
SourKraut posted:Maybe someone already mentioned it and I missed it, but it doesn't look like Ryzen is coming out in Q1 2017 I took that to mean that they are aiming for earlier than the last day of Q1
|
|
#
?
Jan 6, 2017 04:24
|
|
|
Lolcano Eruption posted:I took that to mean that they are aiming for earlier than the last day of Q1 Same. I also heard "end of February" discussed elsewhere as a possible timeframe.
|
|
#
?
Jan 6, 2017 04:29
|
|
|
SourKraut posted:Maybe someone already mentioned it and I missed it, but it doesn't look like Ryzen is coming out in Q1 2017  No AMD, there is nothing mainstream about loving Bristol Ridge, stop it. Stop it now. I'm reaching here, but bear with me. It looks like AMD is aiming to have the cheapest Zen CPU be about 150-180$, so the 4C/8T looks to be directly taking on i5s (based on current FX-8000 pricing). I have no idea how high it will go but if you want people buying your poo poo they might not go higher than 499$? Maybe 399$?
|
|
#
?
Jan 6, 2017 04:44
|
|
|
SwissArmyDruid posted:....uh. Begging your pardon, what did you *think* connected the RAM and PCIe slots to the processor slot? I guess I don't understand what you're getting at with your wasted space comment. What would they put there?
|
|
#
?
Jan 6, 2017 07:59
|
|
|
FaustianQ posted:
AMD might not even with bother with 4c/8t Zen which would explain having Bristol Ridge up to mainstream in that table.
|
|
#
?
Jan 6, 2017 10:50
|
|
|
Word on the street for Bristol Ridge is retail/OEM availability is even worse than the usual already hard to find AMD APUs. And let's not kid ourselves nobody would buy AM4 for Bulldozer derived garbage.
|
|
#
?
Jan 6, 2017 13:30
|
|
|
Palladium posted:
Manufacturers like HP and Acer will buy them in bulk. 4c/8t Zen will certainly exist at some point but since they've never mentioned it, it might be 6 months or a year.
|
|
#
?
Jan 6, 2017 14:35
|
|
|
FaustianQ posted:Careful on that IPC guess
|
|
#
?
Jan 6, 2017 14:56
|
|
|
Twerk from Home posted:6900Ks are over $1000 MSRP and base clocked at 3.2. If they're selling something better than a 6900K, it's not going to be under $500. what people really have to watch for is if they'll come out with 6-cores (where Intel is selling for $400) and their 4-core pricing possibly undercutting i5-ks. Anime Schoolgirl fucked around with this message at 17:34 on Jan 6, 2017 |
|
#
?
Jan 6, 2017 17:31
|
|
|
|
| # ? May 3, 2024 04:07 |
|
|
Prescription Combs posted:I guess I don't understand what you're getting at with your wasted space comment. What would they put there? Traces can be routed anywhere you please, those don't have positional requirements like say, mounting holes do. I'm saying that the mounting brackets and the RAM slots are like the big rocks in the jar, and the traces are the sand that's poured in afterwards. Had they just gone and aped Intel's mounting hole positions, they'd probably have freed up a lot more space to bring those RAM slots in closer. SwissArmyDruid fucked around with this message at 17:37 on Jan 6, 2017 |
|
#
?
Jan 6, 2017 17:33
|
|