thanks,
so if I calculate linear performance increase 54,076*4,2\4,7=48,32 fps
so we are pretty much equal, within error margin
thats nice IPC wise, for 199EUR for 3600
we need to get that clock high
thanks,When set to the same Realtime setting as used by AMD, my 8700K at 4.7 GHz gets this:
I wasn't talking about you specifically, I just noticed a different kind of chatter on both forums and social media, and even though it comes from people who know to adequately discuss these subjects, it will soon radiate towards more... enthusiastic fellows. It's the nature of the beat I guess, we're forced to endure this kind of cycles with AMD as well.It's not really the focus and I'm not saying it to rain on Zen 2's parade - someone just said that Intel couldn't compete clock for clock with 10nm, which is likely not true. I already gave my thoughts on Ice Lake overall, it's irrelevant due to its clock regression and low volumes.
All will be revealed soon enough, IPC wise...
I asked this very question a few days ago and it got ignored.When set to the same Realtime setting as used by AMD, my 8700K at 4.7 GHz gets this:
then that someone should have simpy said that they meant intel couldn't compete clock for clock on the desktop and period.It's not really the focus and I'm not saying it to rain on Zen 2's parade - someone just said that Intel couldn't compete clock for clock with 10nm, which is likely not true. I already gave my thoughts on Ice Lake overall, it's irrelevant due to its clock regression and low volumes.
All will be revealed soon enough, IPC wise...
So does that mean it has much lower latency than Zen 2000 series ?
Well 3 samples are better than just one but still we don't have enough info on those benchmark runs.It's a nice result but nothing more than that (until sample size increases of course). One thing is for sure: Zen2's 2x L3 cache will come into play in a lot of common benchmarks and desktop workloads.Sorry if it was already posted and it's an honest question from me: is this the real passmark of the 3600??
https://www.cpubenchmark.net/compar...-7-2700X-vs-AMD-Ryzen-5-3600/3334vs3238vs3481
the score seems to be the average from 3 samples, where the only sample achieved 21000+ points was the one that used actually useful RAM (16 GB @ 3200) compared to the other 2 (8 GB @ 2666)
The OS sees a monolithic chip similar to previous Ryzen chips, just that the amount of CCXs now can be more than two (up to 16 in case of Epyc 2/Threadripper 3). In that regard latency is more uniform and predictable UMA, though the difference between cache accesses within a CCX and those between two CCXs remains.No - it means that latency should be more consistent across the cores.
Although - I would be concerned, as it is clear from the actual memory topology that it still is not uniform within the socket. So if the OS cannot see that and allocate accordingly, then performance will suffer - but I thought AMD had released information contrary to this already.
We know Core0 will have far quicker access to work that was done on Core1 via the L3 cache than it will on Core4 [which is different CCX].
I'm not sure whether Core0 accessing Core4's cache [same chiplet] is the same as Core0 accessing Core8's cache [different chiplet]. If both of those paths have to communicate across the I/O cip then it will be universal outside of local CCX. If CCX on chiplets can communicate then its another tier of latency somewhere between the two.
The OS sees a monolithic chip similar to previous Ryzen chips, just that the amount of CCXs now can be more than two (up to 16 in case of Epyc 2/Threadripper 3). In that regard latency is more uniform and predictable UMA, though the difference between cache accesses within a CCX and those between two CCXs remains.
New Sisoft scores for the 3700x
It's amazing to see how much better AMD's SIMD performance is now, and even more so with a significant clock speed deficit compared to the 9900K!
My point really is - is the system aware enough to schedule a 3 core task that requires much cross communication on Core0, Core1 and Core2 to take advantage of a common L3 rather than on Core0, Core4 and Core8?
They haven't/aren't going to put Sunny Cove on 14nm?Clock for clock, Sunny Cove is superior to Zen 2. It's not that a working 10nm alone would do the trick, it's that 10nm being a failure has been holding that "next-gen" back.
What? More cores won't beat Zen (in gaming), if Zen is clock for clock faster. That's been proven already by Zen 1 which was hamstrung by it's clockspeeds. However, if they're holding back Sunny Cove to put it only on 10nm, then that would make sense.What? Why do you think the 3000 chips have more cores? Higher density allows for more transistors in the same area. So Intel could put in more cores. You don't think that would matter? Do you think they would not add more cores?
If they had, I'd think we'd have heard about it by now. Intel aren't above pulling Emergency Edition-type launches to rain on AMD's parade when the need arises, and it looks like for once they genuinely have absolutely nothing (other than maybe a round of price cuts) to distract from Zen 2.They haven't put Sunny Cove on 14nm?
They haven't/aren't going to put Sunny Cove on 14nm?
I know I should probably wait for reviews but given that I've held off on upgrading since around December I'm all but guaranteed to pull the trigger. Since people here seem to be quite knowledgeable I'm looking for some input on the cpu aspects.
1. Should I get the 3600 or 3700x? 3600 will be sufficient for my uses, but 3700x will last longer (relevant since I'm still using athlon ii 630 as daily rig).
2. For a SFF case (considering Cerberus), would I benefit from an aftermarket cooler over the stock wraith? I'd have about 140mm of height to work with. I was looking at the Noctua C14, but its performance doesn't seem to be that much better than the stock wraith.
I'm hoping review embargoes lift at least a few days early to help with the choices, because I'm 99% sure I'll be at a microcenter on the 7th.
Allegedly, Francois Piednol really wanted Intel to do just that. Years ago. He was rebuffed. There's this thing called Rocket Lake coming up but uh . . . when, we don't know. And we don't know if it's Sunny Cove or if it's even meant to compete with Matisse.
1). Either one will be a massive upgrade over your current daily driver! That being said, the 3600 may actually OC better, making it potentially a better gaming/ST chip. 3700x is probably the worst-binned chip of all the launch-day Matisse SKUs. It should still do pretty well. 3600 might hit 4.8-5.0 GHz while 3700x may be stuck at 4.7 GHz. Just guessing here, but still. If you are okay with 4.5 GHz or lower on either chip, actually the 3700x would be the better buy . . . if it's in your budget.
2). Which motherboard are you looking to use here? Unless you are going for a bleeding-edge OC on a 3700x, I doubt the cooler will matter much. You will not get a great HSF with either the 3600 or 3700x since both are specced for only 65W coolers. 140mm doesn't give you much wiggle room either. Have you considered an AiO instead? EVGA CLC 280 isn't that expensive, it'll give you lots of cooling overhead, and it will probably fit the Cerberus if you're willing to be flexible about where you put the rad:
https://www.sliger.com/products/cases/cerberus/
What's your budget?
Or consider posting this in the Ryzen 3000 builders thread.
New Sisoft scores for the 3700x
It's amazing to see how much better AMD's SIMD performance is now, and even more so with a significant clock speed deficit compared to the 9900K!
The required topology info is know to Linux, and the latest Windows update finally introduces it as well. So both OSes are aware of the CCX topology now. How they spread threads across virtual and logical cores is still up to each task schedulers (and there likely is still much to research and improve the logic/"AI"), but usually the default is to keep threads of the same process within close locality (i.e. within one CCX) to make good use of the cache.My point really is - is the system aware enough to schedule a 3 core task that requires much cross communication on Core0, Core1 and Core2 to take advantage of a common L3 rather than on Core0, Core4 and Core8?
Conversely, is it also smart enough to schedule 3 core task that doesn't require much cross communication on Core0, Core4 and Core8 to take advantage of uncontested L3?
not anymore, but we also didn't expect this huge (i.e much higher than anticipated) IPC jump either, especially in workloads where intel was leading by a lot. So I'm quite OK with how things turned outPeople are still expecting up to 5GHz on all-core OCs? Sorry, but I'm calling it now. That isn't happening. Not until Zen 3.
The only Zen 2 chip I saw at 5GHz was on LN2 at 1.6V.
(I would be happy to be wrong)
1. Should I get the 3600 or 3700x? 3600 will be sufficient for my uses, but 3700x will last longer (relevant since I'm still using athlon ii 630 as daily rig).
2. For a SFF case (considering Cerberus), would I benefit from an aftermarket cooler over the stock wraith? I'd have about 140mm of height to work with. I was looking at the Noctua C14, but its performance doesn't seem to be that much better than the stock wraith.