AMD Readies Ryzen 5 Series and will offer six- and four-core processors

Yeah, the price is a big argument and it really makes it so that the two core i3 CPUs will become irrelevant (as they should). Plus the whole AM4 platform, that is supposed to be universal for AMD CPUs, which will again cut down on expenses in the future.

I'm sure the performance woun't lack so much behind the intel's i5, but from technological stand point it just would have been better if it was only one CCX and it could provide some iteresting results. Right now it seems like it will be slower in ST applications compared to Ryzen 7.

However, I do agree that APUs could be very interesting chips.

 

Maybe, but we don't know that. I'm pretty sure AMD would allow for a single CCX if it offered a measurable performance increase. Keep in mind, if any of the quad core parts are binned (which I'm sure they are), there is a possibility that one entire CCX could be crippled. As of what we know right now, those chips are not going to be in use, even though it would still make for half of a perfectly functional CPU.

Part of me wonders if the single-CCX chips are reserved for the Ryzen 3 series. It'd make sense, since they'd have to have, at highest, an 8MB L3 cache.

 

You didn't say that. If you want the fastest CPU for your very specific small-scale tasks, then sure, a different CPU is "faster" (not really, but since it gimps 6, 8 and 10 cores, it's 'perceived' as faster). But again, that's not what you said, you said you wanted the fastest CPU out there.

If i wanted the fastest vehicle out there, i would get a race car. But if that did not work for me, because i wanted/needed for it to be street legal, then i would not get a race car. And i would also not say i wanted the fastest vehicle out there, as that would be incorrect.

If you say you wanted the fastest CPU for Gaming at this moment, and have no care for the future, THEN your statement would make sense in being "disappointed" in AMD

 

The latency issue is hardware level. You can't fix an architectural design flaw in software. "AMD Infinity Fabric" suffers from high latency. It's a flaw in the architecture of the interconnect itself.

 

Intel's ringbus is slower than the CPU connection between the CPUs in a CCX too, it's not a flaw either. It's more of a compromise. Ryzen is most likely suffering from not being used properly on the compiler level right now. I really want to see some patched games in the near future.

 

It will be interesting if Ryzen can be ever used with something like 4266mhz ddr4. If so I could imagine it being 100% faster in ccx interconnect also then 2133mhz ram theoretically at least.

 

On the topic of RAM, why is DDR4 such a smaller leap than previous generations? I don't see any DDR4-4800 around while DDR3-2400, DDR2-1200, and DDR-600 were all readily available with 1600, 800, and 400 being very average/standard.

 

Just stumbled across this:

https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-34#post-38797798

Rumor mill of course.

Begs the question how reliable or realistic these claims are and further if a new stepping of Ryzen sorts the CCX bottleneck out then what of early adopters?

X399 looks mighty fine, though :infinity:

 

Pings :

same physical core :
Intel : 14ns
Amd: 26 ns

adjacent core:
Intel: 76 ns
Amd: 42 ns

Over 4 cores:
Intel: 76ns
Amd: 142ns

Overall intel is much better .
Ryzen also have some problems inside the CCX with data over 4 Mb , almost 3x latency . That should occur over 8 Mb , but no, on AMD occurs on both over 4Mb and 8Mb

Okay, AIDA may report incorrect values, but the delta is reported correctly

https://www.techpowerup.com/231268/amds-ryzen-cache-analyzed-improvements-improveable-ccx-compromises

 

And those ns is what still bellow 0.01ms ?not really affecting your gameplay.

 

In worst case scenario, those latencies may be per instruction latency. (close to impossible, but switching may happen thousands times per second)
That's reason why there is big difference if threads are locked to one CCX or allowed to jump like crazy between both of them.

I already saw few synthetic tests confirming that even there is measurable difference.
And I saw one more interesting thing around (not yet confirmed from 2nd source).

Windows addresses/assigns threads to CCXes differently based on way you disable SMT.
If SMT is disabled via windows, it may result in not assigning threads correctly (4 separate threads assigned to 2 real cores). If SMT is disabled via BIOS, everything is fine.

If you want to "park" 2nd CCX and prioritize use of 1st, you can't do it in windows. as Windows will "park" logical cores. So one has to disable cores fully in windows or BIOS (both require reboot).
- - - -
M$ stated that scheduler is fine, but that SMT thing confirms that it is not as fine as it could be. Secondly Windows scheduler does not distinct between CCXes like linux does.

 

Yeah, I have personally found this to be pretty irritating. We're limited by pretty much the same speeds as DDR3 while getting significantly worse latencies. DDR4 is hardly worth considering unless you get at least 2667MHz.

It's been available for over a full year and even Intel chipsets still require you to OC in order to reach higher speeds.

 

So, the 1600X has the same base and turbo speeds as the 1800X and costs half as much. Very intriguing. I'm waiting to see how the performance matches up. The 1600X may outdo the 8 core 1700.

 

So, AMD has 2x faster adjacent core communication, and Intel has 2x faster adjustent core complex communication. It isn't as bad as I thought it would be, and since the scheduler seems to be ok, it's most likely compiler optimizations and a bit of work from developers.

 

Has anyone tested Ryzen latency when PCI-E bus is loaded?

Intel latency numbers are from an 8 core right? How does it scale down to 4 cores? I would assume less hops = lower latency given the ringbus design?

Also kind of curious how both scale up. Intel switches over to some weird dual ringbus design IIRC. I wonder if AMD's design is better with higher core counts in terms of latency.

 

1600x looks intresting, I bit weary of the 95tdp it claims I wonder how offthat claim really is

 

Remember, these TDP ratings are based on thermals, not actual power consumption. Even the 8-core models proved to have different actual wattages.

If you intend to overclock, the advertised TDP is irrelevant anyway.

 

I know what it means, 1700 claims 65 TDP but more like 125 1700x/1800x are 95 but again like 145 not even intel exaggerates there TDP by that much Intel is usual off by about 20 or so these ryzen chips are off anywhere from 40~60+

OCing in irrelevant to majority of people, just like SLI. I look at TDP and expect stock max wattage to be close to the advertised TDP underfull load, not 40~60+ off

6700k is 95tdp it actual max wattage is 115

Like said I original said 95TDP 8c/16t cpu that actual ran about that with total watts would been amazing to me, ATM price the only thing I see as amazing.

 

So the 1700 at its base clock is really 125 and not 65? The fact it overclocks itself to a certain frequency, that overclocked frequency is what the TDP should state? It sounds more like if a cpu is going to be self-overclocking, the TDP should be stated for the base clock and the top boost clock. Effectively stating a TDP range. The 1700 would then be listed at a TDP range of 65 - 125 as an example.

 

From what I understand, the extra bandwidth doesn't matter in most games unless there's serious problem. The latency does however. Makes DDR4, especially at the current joke speeds, somewhat of a downgrade.