If the primary job of the fabric is to convey cache snoop traffic (which it probably is) then having it tied to the L3 cache speed would sound reasonable. However they did it, I can assure you it would be on the back of a hell of a lot of workload simulation.Originally Posted by Corky34
Well if the designer of Zen didn't see a need for the data fabric clock to be (adjustable) running faster than the IMC then I'd like words with that designer.
Unless I'm mistake the data fabrics primary job is to carry all communication data going into and out of the CCX's (the 4 cores and their L1, L2, L3 cache) so (afaik) that's data going to and from the RAM, L3 cache on the other CCX, PCIe devices, and other CCX's.
I was just looking at how AMD describe the data fabric and they do call it 'scalable' data fabric so i would assume it's more scalable than RAM clocks.
From what i can tell it's similar to Intel's QPI.
Last edited by Corky34; 15-03-2017 at 11:29 AM.
I think Intel is screwed look at the R5 1400 pricing.
Also some more investigation of gaming performance and SMT in gaming:
http://www.tomshardware.com/reviews/...ming,4977.html
https://www.purepc.pl/procesory/amd_...wylaczonym_smt
More news on the Ryzen APU and Zen 2:
https://videocardz.com/67362/amd-pin...o-8-zen2-cores
Channel partners received first roadmaps showing long term plans for AMD Zen architecture for CPUs and APUs. AMD Raven Ridge APUs are currently expected to launch this year with up to 4 Zen CPU cores on FP5 BGA package (Mobile). This year the new APUs will be offered alongside existing Bristol Ridge APUs (desktop), but in 2018 only Raven Ridge will be present int the portfolio for both mobile and desktop platforms.
Moving on to Pinnacle Ridge. This is the Zen2 architecture, refined and matured Zen core will likely be offered as Ryzen 7/5/3 2000 solutions (should we call it Ryzen 2000?). While the roadmap lacks many details, it does ‘confirm‘ that Zen2 will also be 8-core chip, so we should at least be expecting higher clocked chips.
Last edited by CAT-THE-FIFTH; 15-03-2017 at 02:30 PM.
I'd argue, but:
Based on the results in that review you're right. There are very few games that see a meaningful difference between the i5 and the two i7s, suggesting that all those extra cores don't make a vast difference once you get over 4. That could put the 1400/1500X somewhere around R7 1700(X) performance at < $200... hello mainstream domination....
And dat Deus Ex result
AMD game deliberately hampering competition processors initAnd dat Deus Ex result
No, that is an intriguing result.
Interestingly the Hexus results for the game seem different.
Its why I wish testing could be done ideally in two,maybe three different locations for a game.
Hexus use an EVGA 1080 FTW rather than a founder's edition, so should have slightly higher results ... OTOH they also test at Ultra rather than Very High, so should probably have lower results. That Hexus' results are significantly higher than Tom's yet display a level of GPU-binding that isn't present in Tom's results ... well, *shrug*
Does it have a DX11 mode? Are Hexus testing in that, instead of DX12?
My understanding (which could be incorrect) is that, like DanceswithUnix talks about, the clock is tied architecturally to the IMC. That's how it appears in the material I've seen anyway. If that's the case, it wouldn't be an insignificant change to decouple it. It's a curious choice nonetheless, but don't forget AMD had to make this fabric scale across MCM chips too, so having the clocks to high might make it unworkable in those situations.
Remember how Intel went from their ring bus being tied to L3 (and therefore core) clocks in SNB, which was a trade-off in itself as it allowed predictable and consistent latency (in cycles) for L3 access, but at the same time meant the cores had to run at full speed for even the IGP to access the max potential bandwidth, which caused higher power draw. In Haswell, they decoupled the L3 clocks from the cores, allowing for the ring bus to clock more flexibly, but complicating communication between cores and the L3.
That's some epic competition right there! I think more than a few of us have been criticising Intel's desktop pricing for a long time now, so it's great to see AMD challenging it like this. Perhaps I'm nitpicking, but it would be nice to see one of the quads reaching 4GHz clocks. I know they probably don't want to risk competing with their more expensive parts but it would likely make it even more competitive against Intel's offerings for a more affordable gaming system!
It's a bit disappointing that we won't be seeing Zen on desktop APUs for a while, but at least they're planning a mobile launch.
BTW is Zen2 just a community-given name? The slides are still calling the cores in Pinnacle Ridge 'Zen'. Have AMD given any details on this publicly yet?
I really think on OcUK,they have swallowed the whole you NEED 8 cores for gaming bumpf - tried telling them most gamers will be on 2C/4T and 4C/4T,and people need to stop thinking enthusiast forums are typical cases of gaming hardware. It didn't go down so well I said the 4C and 4C/8T versions would have most of the performance of the 8C/16T for gaming,and would help push more gamers off 2C/4T and 4C/4T which is more important than the 8C version.
Plus its news to me that a £170 to £200 CPU is budget......LOL.
Don't get more wrong games will push towards more cores,but there is a big difference about "better performance" and "needing loads of cores",which I think is being lost currently.I'd argue, but:
Based on the results in that review you're right. There are very few games that see a meaningful difference between the i5 and the two i7s, suggesting that all those extra cores don't make a vast difference once you get over 4. That could put the 1400/1500X somewhere around R7 1700(X) performance at < $200... hello mainstream domination....
And dat Deus Ex result
Apparently the R5 1600X comes with the same cooler as the R7 1700,so unless the parts need a lot of voltage,I think a slight overclock will do.
Also they are 65W TDP parts too.
Last edited by CAT-THE-FIFTH; 15-03-2017 at 09:20 PM.
Wraith Spire on the R5 1500X and R5 1600.
bsodmike (16-03-2017)
Yup, looks good. I'm surprised that they've got the 1600 listed as 95W - I was pretty sure that was meant to be a 65W part....?
The 1500X under a Wraith Spire should get a pretty decent all-core overclock ... I reckon the 1500X could be worth the extra money just for the better cooler!
You can already see that that's the case in benchmarks where a CCX is disabled. Of course, different games vary, and some like BF seem to like more cores, but for the price that's a pretty substantial win!
I know the exchange rate at the time had a lot to do with it, but I remember £150 buying you a very capable CPU or GPU!
Horses for courses. I think some people forget that - there are different areas that the CPUs perform best. We're now back to being in a market where the answer to 'what's the best CPU' is 'it depends...' Yeah, the 1700X is a bit more expensive than the 7700k, but that x264 performance
I did hear some rumours (albeit before release IIRC) implying that the 4C CPUs would have randomly-chosen cores rather than always being one whole CCX. The cache size in those slides implies otherwise thankfully.
Edit: It would have been nice for them to include a cooler with all of the 8C models IMO. I mean you can sometimes use an existing cooler with a bracket (like in my case), but for a new platform there will be a lot of people without a compatible cooler, and there are still a lot of coolers on the market that either don't support it or need separate brackets. Not everyone overclocks...
I think the case is you could an R7 1700 for £320 or a R5 1400 for £170 and £150 is like the difference between a GTX1070 and a RX480,or a RX460/GTX1050 and a RX480 and an SSD.
BTW,the R7 1700 comes with the Wraith Spire cooler.
Edit!!
Also if AMD keeps AM4 compatible with later Ryzen CPUs,the upgrade path is there if you need it.
Yeah it's just the X versions that do without. I know it's no different to what Intel are doing with the k series, but personally I like to have a stock cooler as backup anyway - you know it's going to work, even if it's not going to be as quiet as an aftermarket one (though the Wraith ones are supposed to be decent anyway?) That, and it gets you up and running without having to invest in an aftermarket cooler which may be of no real benefit if the stock one is quiet enough and you're not overclocking.
I tried the stock Intel cooler on my i3 and really wasn't impressed - it might not read much on reviewers' test instruments but I found it incredibly annoying, subjectively. Scythe Shuriken FTW!
Yep, I do hope they stick with AM4 for a while - it seems to be a solid platform, so aside from new features etc I don't see why they'd be forced to break compatibility in the short term at least.
People are getting good overclocks on the Wraith Spire and the R7 1700 - its not even the top version of the stock cooler.
Looking at the R5 1500X and R5 1600,they seem to have the Wraith Spire cooler,so that means with one of the better B350 motherboards you are good to go.
There are currently 3 users browsing this thread. (0 members and 3 guests)
Posting Permissions
LinkBack URL
About LinkBacks
Reply With Quote
