I remember that also. Anyway, it will be around then.I thought they were launching on July 7th (7/7) to tie in with the whole angle of being the first 7nm CPU. July 7th is a Sunday so perhaps that means no real availability at most places until the 8th, but I don't recall hearing the 10th for availability.
Would it be pretty bad for (laptop) power consumption to have a chiplet BGA socket directly linked to a chipset/northbridge+iGPU?
In the old days (i386-5x86 days) I think there were off-CPU northbridges, as well as onboard GPU's.
I think it was June 10th when more info would be made available at E3, then pre-order on July 1, with availability maybe on July 7.
I feel the power usage of IF is overstated since many people (including one notorious article here on AT) mix up IF with the whole uncore. Case in point are the Threadripper WX chips (with 24 and 32 cores) compared to the equivalent Epyc chips. You'll notice the former have much lower power usage at idle. How could that be, after all the IF has to be the same between both? The answer is that Threadripper WX has only half the IMC and half the PCIe lanes activated, and those are eating a big part of the power usage at idle. And if one worries that the actual IF connections on MCM eat power, by centralizing them all into the IOC Zen 2 has less IF connections than Zen 1 had per die/chiplet.The idea was getting at is that IF requires more power than a ring bus and a decently significant part of the CPU package power usuage. When dealing with IF over substrate TR and EPYC shows that, that power usage dramatically increases. So the more IF complexity that you the more of your power budget is going to non-clock or processing module (whether CPU or GPU) work. Zen has and maybe Navi does processing units can have great efficiency. It's why Zen has been so great at so many levels accept the high end performance till this point (assuming benchmarks concur). But power wise Monolithic designs will have an advantage on power usage over chiplet designs. Zen 2 is a great compromise on power vs performance for what it is on most levels and will still be if AMD decides to do a low voltage desktop Mataisse. But the power margin on laptop and other mobile devices I think the substrate IF power usage might be the straw that breaks the camels back when it comes to chiplet based APU's even ones where there isn't an intermediary IO chip where the IO is in the GPU portion (or CPU, but we know the CPU chiplets don't have it).
I feel the power usage of IF is overstated since many people (including one notorious article here on AT) mix up IF with the whole uncore. Case in point are the Threadripper WX chips (with 24 and 32 cores) compared to the equivalent Epyc chips. You'll notice the former have much lower power usage at idle. How could that be, after all the IF has to be the same between both? The answer is that Threadripper WX has only half the IMC and half the PCIe lanes activated, and those are eating a big part of the power usage at idle. And if one worries that the actual IF connections on MCM eat power, by centralizing them all into the IOC Zen 2 has less IF connections than Zen 1 had per die/chiplet.
Either fake or it isn't working properly with Zen 2.
Why ?Either fake or it isn't working properly with Zen 2.
Wow stuff is getting interesting.
X570... Amd can't win, punters screamed for advanced features, premium build, lot's of choices and lots of power available for OC, the gave it to them, now they complain!
Don't forget you can use the last gen boards.
Wow stuff is getting interesting.
X570... Amd can't win, punters screamed for advanced features, premium build, lot's of choices and lots of power available for OC, the gave it to them, now they complain!
Don't forget you can use tthe he last gen boards.
It's AMD they're damned if they do and damned if they don't.
It's AMD they're damned if they do and damned if they don't.
Why ?
Zen+ does 56ns on 4000MT/s
If you take the userbenchmark score of an 3600 with 2666MT/s ram and adjust it for memory freq gain it comes at 40ns.And that aida is on a x570 and oced chip.
I say plausible.Not that its certainly true, but plausible.
Like a month ago userbenchmark scores were 100ns on a 3200MT/s ram, now its 80ns at 2666.
Zen2 apparently supports JEDEC DDR4-3200 speeds. Does anyone know if this only applies to the X570 chipset or if this will be available on X470 boards as well? (And is there a release date for unbuffered ECC DDR4-3200 DIMMs?)
Unknown, I’ve been watching this however I’m not an expert.
3200 speed appear to be only for 2 dims per the Asus memory speed chart on the promo info regarding their X570 boards.
If I remember correctly they could be single or double sided.
Unknown, I’ve been watching this however I’m not an expert.
3200 speed appear to be only for 2 dims per the Asus memory speed chart on the promo info regarding their X570 boards.
If I remember correctly they could be single or double sided.
FWIW I heard the person who posted deleted it, and nothing further was stated. Unclear if it was real or not. Wow if true.
Thank you - not sure why I thought 3+3.The B450-DS3H is 4+3, it should be fine.
FWIW I heard the person who posted deleted it, and nothing further was stated. Unclear if it was real or not. Wow if true.
Here is it compared to a 95W 2600X i helped to set up (with faster memory to boot):New GB4 for Ryzen 5 3600
well AMD had a like 34% perf improvement for CS:GO on there slides for 3800x vs 2700x so that should be a pretty good indication. a 2700x should be clocking around 4.2 so it only has like a 9% clock advantage.The indications are that IPC is on par with or better than intel, but that means little to gamers if these chips are still affected negatively by latency which tends to diminish otherwise impressive CPU performance. Time will tell. I don’t see many comments about latency concerns. I find that interesting.
Yes, they have doubled the cache, which should also cut memory accesses significantly. Previously the L3 caches couldn't really share data between the two CCX. Geekbench seems to indicate (for a while now) that there are still 2 separate instances, but if they can snoop data from each other with low latency, we could be talking about up to 4x the effective L3 cache size.well AMD had a like 34% perf improvement for CS:GO on there slides for 3800x vs 2700x so that should be a pretty good indication. a 2700x should be clocking around 4.2 so it only has like a 9% clock advantage.
https://images.anandtech.com/doci/14407/COMPUTEX_KEYNOTE_DRAFT_FOR_PREBRIEF.26.05.19-page-022.jpg