Look at this, the 9590 is 1.2ghz higher, but it's slightly slower per core than the Athlon and the Athlon isn't even Zen.
AMD has been making small IPC improvements with every new revision of Bulldozer. The most recent Carrizo+ (Excavator) cores have a significantly higher IPC but they just don't really get fleshed out into higher end parts. The 845 is the only one, and honestly it's a bit nerfed by the lack of L3, low amount of L2, and aggressive dynamic clocking.
You see, that's what I'm talking about. It looks like a modified Laptop CPU design from the lack of Cache and it's doing well considering it.
It is a laptop CPU, just clocked higher.
1 Like
Just as I thought.
Also, Zen is supposed to have a 40% IPC improvement over Excavator, not Bulldozer.
Shame they did not have the cash to make an 8 core excavator with full cache on FM3+ that we could have now.
When I see gaming benchmarks between Haswell and Skylake I wounder if IPC is starting to show diminishing returns.
its not that its just intel hasnt been increasing the ipc much in gen
Yeah, that's what this comparison is about. the Athlon is Excavator based and it has a 1.2Ghz handicap along with less cache than 9590 and it out performs in performance per core.
Yeah, Skylake was a big disappointment. Skylake was suppose to be a tick and a tock, but at best the gains over Devil's Canyon were 5%.
yeah but still the athlon has only 2 modules with each 2 integer cores sharing cache and a fpu.
1 Like
Although you really gotta keep in mind the sample size there. I would imagine yes, it is likely a faster CPU. But it may not be as fast as it looks because the sample size for the athlon x4 is only 15 whereas the sample size for the 9590 is over 1000.
honestly it's a bit nerfed by the lack of L3
Well, in the current AMD architectures (not including Zen) there is no point to having L3 cache since the L2 cache is the DRAM facing node, but a larger L2 could certainly help. Adding L3 would only add unnecessary complexity to the die meaning more power draw and heat dissipation. And since the L2 and L3 would be at the same speeds there really is nothing to gain, but adding L2 in that architecture makes sense. The only limiting factor with adding more L2 is die space.
I think people might understand cache levels better if the numbering were reversed, but what can you do... \o/
-------------------------------------------------
Physical CPU with no HyperThreading, like current gen AMD.
DRAM
.|
.+- Memory Controller / NUMA Core
.....|
.....+- L2 Cache
.........+- CPU Core 0
.........|...|- L1 Code Cache 32KB
.........|...+- L1 Data Cache 32KB
.........~
.........+- CPU Core X
.............|- L1 Code Cache 32KB
.............+- L1 Data Cache 32KB
Since there is no third level component (eg. no HyperThreading), there is no advantage
to adding L3 cache in this scenario. In fact, since the L3 and L2 cache would have
the same speed and latency characteristics, there would be more of an advantage to
increasing the size of the L2 cache then adding an L3 cache level.
-------------------------------------------------
Physical CPU with HyperThreading, like latest Intel or AMD Zen.
DRAM
.|
.+- Memory Controller / NUMA Core
.....|
.....+- L3 Cache
.........+- CPU Core 0
.........|...+- L2 Cache
.........|.......+- Thread 0
.........|.......|...|- L1 Code Cache 32KB
.........|.......|...+- L1 Data Cache 32KB
.........|.......|
.........|.......+- Thread 1
.........|...........|- L1 Code Cache 32KB
.........|...........+- L1 Data Cache 32KB
.........~
.........+- CPU Core X
.............+- L2 Cache
.................+- Thread 0
.................|...|- L1 Code Cache 32KB
.................|...+- L1 Data Cache 32KB
.................|
.................+- Thread 1
.....................|- L1 Code Cache 32KB
.....................+- L1 Data Cache 32KB
In this scenario with HyperThreading, the L3 cache becomes necessary since the path
to the actual processing core is deeper.MHZ Doesn't matter as much as # of cores, and IPC. Sadly # of cores has been retarded for ages by now, but with dx12. It seems cores finally gets a saying rather then MHZ. Even the "simplest" AMD 8 core cpu will dominate Intel with dx12, atleast for gaming. since 8*4,5Ghz > (8/2)*4Ghz. Albeit older AMD cpus has some caveats, the mere core count just dominates.
I actually threw down the money for the Athlon x4 845, and made a thread about experimenting in overclocking it:
On my Gigabyte F2A-A88X-UP4, I had a weird issue with some games with some fluctuating clockrates during certain games - probably due to the beta BIOS. I actually had a back-and-forth with Gigabyte over the issue, although it has been about a month since I have received any reply about any updates. If/when Gigabyte resolves this issue, I plan on making a follow-up thread.
2 Likes
i dont think so. ☺
ipc and per core performance will still matter even with dx12.
Depending on the games you play and how you play them.
There arent manny sources wenn it comes to dx12 cpu benchmarks yet.
But we have seen some ashes of singularity tests in the past.
Current AMD chips cant dominate intels counter parts in gaming.
But the new upcomming ZEN might be somewhat compatitive again.
also worth noting, intel is catching up to core count
1 Like
Yeah also that.
But according to this older article on pcper it shows cpu and gpu scaling in ashes of singularity on both dx11 and dx12.
And as you can see, the per core performance and ipc still obviously matters in this game,
no-matter dx11 or dx12.
I think we will probably see a situation where the top tier amd zen offers 80-90% of the gaming performance of the comparable i7 (Intel will probably be 100 more) and the 95-110% of the gaming performance of the comparable i5 (I can see the amd costing around the same as the i5 this time) . I'm basing this on claimed performance gains.
I think it will be a similar situation (maybe a little better for amd) as when the 8350 first came out
Their core count is slowly rising, but only on the higher end of things. We have been stuck with the same configuration for the i3 i5 and i7 for quite a while now. Would have thought that they would eventually upped the core count on those chips or used a new naming scheme. Quad core ought to be the standard by now.