If this happens, and it provides the performance of an octacore Haswell for a resonable price I'll be all over it.
I expect Zen to have IPC roughly equivalent to Haswell; lower than SKL/KBL. I also do not expect it to come anywhere close to the 4GHz+ stock speeds that SKL does.
3GHz for 8 cores would be a solid achievement, IMO.
If this happens, and it provides the performance of an octacore Haswell for a resonable price I'll be all over it.
Yes, I think 8 core Zen will be faster than 4 core Skylake & KabyLake in MT performance. Don't you?Clearly you expect Zen 8 core to faster and cheaper than Skylake or Kabylake.
If Zen would really aim at rather low clock frequencies like 2-3 GHz, they wouldn't use a 4 cycle latency for the 32kB L1 D$, as each additional cycle hurts.At stock speeds, due to a combo of lower clocks and likely lower IPC, 8 core Zen will be at least 25% off 4 core Skylake/Kabylake ST performance.
Many hints point to small Zen cores, and AMD will likely sell the server dies also as FX for DT, so 8-core CPUs are not that unlikely. Especially as clock frequency and TDP/core can't be scaled too well, while adding ~5mm² cores is cheap.It is far from clear if AMD will even be offering an 8 core Zen for desktop users in 2017.
At improved power efficiency (aside from more advanced pwr mgmt tech), 14nm CPUs could sustain higher clocks with the same number of cores active.SKL is hardly better than Haswell so it won't be bad if Zen doesn't hit Skylake.
You also appear to expect it to be cheaper, as though FX8350 pricing will carry on to 8 core Zen.Yes, I think 8 core Zen will be faster than 4 core Skylake & KabyLake in MT performance. Don't you?
You really ask for a source when the only product out is a 500Mhz DDR part?
And you forget the latency DDR1 had when you fetched more data. So you quickly end up around 30-40ns instead of 10-15ns.
Little disappointed yet more 28nm chips are coming out but I get they want to mature the platform first. I doubt DDR4 will make much of a difference for APU gaming over existing Godavari 7870K with a good set of 2400Mhz G.Skill Tridents but we'll see. Perhaps colour compression will help a bit. At least power consumption is looking good.
If Zen would really aim at rather low clock frequencies like 2-3 GHz, they wouldn't use a 4 cycle latency for the 32kB L1 D$, as each additional cycle hurts.
And rememberPollack'sGlew's rule, where 2x the cores at the same TDP could still run at ~70% the clock frequency.
http://pc.watch.impress.co.jp/video/pcw/docs/489/233/p5.pdf
This old graphic doesn't account for possible boosting modes, which are common nowadays.
You also appear to expect it to be cheaper, as though FX8350 pricing will carry on to 8 core Zen.
Personally, I'm expecting the flagship Zen to debut at $599.99 -- or roughly half the cost of the i7 5960x. That should give them a nice profit margin -- and reasonably brisk sales as long as performance meets most users expectations.
Personally, I'm expecting the flagship Zen to debut at $599.99 -- or roughly half the cost of the i7 5960x. That should give them a nice profit margin -- and reasonably brisk sales as long as performance meets most users expectations.
(Of course, it could offer twice the performance of Skylake and would still be deemed a disaster by about half the people on this forum).
I was reading the other day a breakdown of Intel's desktop CPU sales and apparently 99% of Intel processors sold come with an IGP, so how big is the HEDT market anyway?Personally, I'm expecting the flagship Zen to debut at $599.99 -- or roughly half the cost of the i7 5960x. That should give them a nice profit margin -- and reasonably brisk sales as long as performance meets most users expectations.
(Of course, it could offer twice the performance of Skylake and would still be deemed a disaster by about half the people on this forum).
Based on the tests I´ve made in the past the A10-7870K (8 CUs & 2 RBs @ 866MHz) would have ~33% performance benefit in GPU performance from 6Gbps GDDR5 (with 128-bit bus). At 4Gbps the benefit is already ~24%..
What do you think most user's expectations are? Even outside the extremely pessimistic, most people seem to think Haswell IPC at similar clocks would be doing quite good. Even if Zen gets Haswell levels of IPC and clocks, they would have a hard time competing with Intel's current quad channel HEDT platform. If Broadwell-E does debut with a 10 core Extreme processor and pushes the 8 core down to the spot occupied by the 5930k now, $600 would be a disaster waiting to happen especially considering BW-E will have been out for a half year at the minimum by the time Zen ships.
I was reading the other day a breakdown of Intel's desktop CPU sales and apparently 99% of Intel processors sold come with an IGP, so how big is the HEDT market anyway?
Zen will also have to be competitive against the i7 5930K(and successor processors that replace the i7 5930K).
Personally I maintain that single core performance will determine whether Zen is successful or not.
There are a number of applications where the FX-8350 is faster than i5's, but that hasn't swayed many people and for good reason.
The FX' main problems in MT (games with just a few threads and DX11) are also a result of the now in comparison really low ST performance, as MT just scales this up. The remainder gets lost due to low power efficiency under a TDP cap.Personally I maintain that single core performance will determine whether Zen is successful or not.
There are a number of applications where the FX-8350 is faster than i5's, but that hasn't swayed many people and for good reason.
Personally I maintain that single core performance will determine whether Zen is successful or not.
There are a number of applications where the FX-8350 is faster than i5's, but that hasn't swayed many people and for good reason.
HBM1 wont be faster than it is.
Overclocking doesn't count.
HBM2 will double the clock, still be far behind in latency.
And no, its still not better latency than DDR2/3/4.
pseudo random access
AMD isn’t discussing how much the latency was reduced
Mostly because of gamers who won't touch FX. DX12 should solve AMD's single core problems in games though, so if gaming performance is nearly identical and you get 8 cores instead of 6 or 4, it's a no-brainer choice. Nobody is going to buy Intel because of SuperPi.
Intel doesn't have much leeway in pricing because they already have everything so perfectly segmented. it's all downside for them if Zen is anywhere near decent performance. AMD can only lose if Zen is worse than Bulldozer.
Mostly because of gamers who won't touch FX. DX12 should solve AMD's single core problems in games though, so if gaming performance is nearly identical and you get 8 cores instead of 6 or 4, it's a no-brainer choice.