So, in theory this should be a big deal for computer people. This is the first time in around 6 years since Sandy Bridge (2nd generation) from Intel, that there's been anything of note to talk about in the consumer CPU market.
It was just announced a little bit ago, so benchmarks are sparse, and all we have is a few videos from the likes of:
What we see in these CPU's is a lot more cores for the dollar (all with hyperthreading), however the cores themselves can't compete with Intel. For example, in the 3rd video I linked, we see that the 1800x was directly competing with the 6900k, and we see it had identical single core performance.
The 6900k however is a broadwell processor, so we can roughly expect a 10% increase in core performance in the 7700k at the same clock speeds. The other important thing to note is the 7700k is clocked at 4.2Ghz base, while the 6900k is clocked at 3.2Ghz base. This sample math to me suggests that the 7700k should have roughly 44% higher single core performance than the 1800x without any OC's. That said, the Ryzen 7 chips have twice as many threads, so if multithreading optimization is perfect, in theory they should be 38% faster on tasks that can utilize all these cores.
Anyway, from a consumer standpoint, the 1600X and the 1700 will likely be the sweet spot, and from looking at the specs, I can safely say that they are the best performance for the dollar. There's a lot of information we still need about motherboards and such, but what we can say is that the 1700 (priced at $330), will be around 30% faster for a similarly priced i7 7700k for tasks that can utilize all those cores. However, for whatever reason, a lot of games wont even use more than 4 threads, and just about none will use more than 8, so is there any reason of choosing this over Intel?
This video here is an excellent demonstration of cores and threads you need to play most games with a mid-range graphics card:
Thoughts on Ryzen? How will this shape our custom rig building in the next month to follow?
March 2nd is when Ryzen chips will go on sale, you can preorder them now.
In my aged opinion, AMD was more into future proofing their chips. When I built mine (15 years ago), it still was up to par 2-3 years later. I think AMD is including all of the cores and multithreading so as to keep it relevant when VR and 4k-8k comes more consumer friendly.
As such, I expect the old battle lines to be drawn in a few months once these bad boys are out and people get a good chance to play with them and see how they perform. I'm looking forward to knowing more about what mobos and gpus are going to be compatible. I've been seriously considering building a new PC for work and some light gaming and AMD may have just solved my problems.
I've been looking to upgrade my i5 760k. I held off on upgrading because there was a chance that AMD could make a great CPU for cheap or atleast force a price drop from Intel. So far, from what we've heard, Ryzen sounds pretty compelling. I'd like to see some motherboard pricing. I'm looking forward to seeing some benchmarks when these things get released, especially on some of the lower end stuff like the 1300. It's priced similarly to the i5 7400, is slower but has 8 threads. Really curious how that thing is going to perform.
If the B350 motherboards are priced right, some of this lower end stuff could be killer. It's very possible that an overclocked Ryzen 1100 or 1200+mobo+cooler would beat a locked i5+mobo at a lower price (which would be great since I was considering getting a locked i5...).
Maybe I'm just getting too overly excited and setting myself up for disappointment, but you know... the worst case scenario is the CPU sucks and I just get an i5 for high prices, which is what I was going to do anyway
The only thing that I wish would be standardized are mobos and chips. Somehow create mobos to accept either chip so that way if we find that Intel doesn't work, we can swap in an AMD without having to send everything back and exchanging.
AMD has been putting their stuff into the budget sector and Intel has dominated most every other aspect. I'm looking forward to the competition and joining the old debate on who is better.
So I've recently decided to build a new pc for myself the first time and I'm waiting for this to release to see its effects on the market. Do you guys know how long it takes for accurate performance benchmarks will be available and how long it'll take for Intel to respond which price decreases? I know Intel is releasing a new version of their chips at the end of the year but I'm not gonna wait that long.
I want to see proper independent benchmarks. Just hype from AMD. Also, the NDA for reviews apparently doesn't lift until the release, which I personally find a but quesitonable
but what we can say is that the 1700 (priced at $330), will be around 30% faster for a similarly priced i7 7700k for tasks that can utilize all those cores.
Should be a lot more especially when you take into account OCing
---
However, for whatever reason, a lot of games wont even use more than 4 threads, and just about none will use more than 8
but what we can say is that the 1700 (priced at $330), will be around 30% faster for a similarly priced i7 7700k for tasks that can utilize all those cores.
I think from the numbers we've seen, 30% sounds right on point.
Also, nice wikipedia article, all I know is that for our CFD simulations, almost everything could be parallelized. We ran a cloud computing server where we used some 80 cores, and there was significant performance gains. Most processing power is needed on the physics engines of these large studio games, and currently parallelization isn't very hot.
I think from the numbers we've seen, 30% sounds right on point.
+30% performance with 2x the core count in a nearly-100% parallel task would be awful; that would mean that the core performance was 1.54x weaker than Kaby Lake which is quite clearly not the case.
The actual multithreaded performance would be 2x (due to 2x core count) minus the difference in core performance due to the architecture and achievable clock speeds - twice as many cores with 80% performance-per-core = 1.6x performance (0.8*2) for example.
A number like 30% faster than a 7700k is probably coming from a test that was not scaling well across cores and/or from a large frequency gap that may not hold up when you can OC both CPU's (the $329 bin has low stock clocks)
I think from the numbers we've seen, 30% sounds right on point.
+30% performance with 2x the core count in a nearly-100% parallel task would be awful; that would mean that the core performance was 1.54x weaker than Kaby Lake which is quite clearly not the case. They've broken some world records already which were held by 8c16t Haswell/Broadwell CPU's and Skylake/Kaby is not a massive leap beyond that~
The actual multithreaded performance would be 2x (due to 2x core count) minus the difference in core performance due to the architecture and achievable clock speeds - twice as many cores with 80% performance-per-core = 1.6x performance (0.8*2) for example.
On February 24 2017 06:55 Cyro wrote: It varies program to program but many are functionally 100% at these core counts. x264 on many settings runs literally twice as fast (or like 1.99x faster) when you double core counts, many other usages do too.
Yeah because it's basically one unique task spread across cores the entire time, but afaik you can't scale these gains linearly on such programs for ever (performance starts decreasing above a certain core count due to bus width afaik? not really sure), and for programs such as games you won't have a program that is a giant parallelized process the entire time.
But since the quote is about tasks, not games or "real" programs that do more than one task, yeah you can indeed double gains
I think from the numbers we've seen, 30% sounds right on point.
+30% performance with 2x the core count in a nearly-100% parallel task would be awful; that would mean that the core performance was 1.54x weaker than Kaby Lake which is quite clearly not the case.
The actual multithreaded performance would be 2x (due to 2x core count) minus the difference in core performance due to the architecture and achievable clock speeds - twice as many cores with 80% performance-per-core = 1.6x performance (0.8*2) for example.
A number like 30% faster than a 7700k is probably coming from a test that was not scaling well across cores and/or from a large frequency gap that may not hold up when you can OC both CPU's (the $329 bin has low stock clocks)
Sorry, when I said perfectly parallelized that was a mistake on my part, what I meant is the best optimized programs, so like 95%-97% parallelization?
On a per core performance, I expect the 7700k to be 25-30% quicker (by looking at the per core performance of the 6900k vs 1800X benchmark, and then comparing 6900k vs 7700k).
On February 25 2017 05:04 Cyro wrote: Cinebench ST for OC'd Kaby hits about 220 while Ryzen's probably reaching about 180 at OC (162@4ghz)
That's a 22.2% difference, but we'll see. My expectation of the OC on air is 4.3Ghz or so, seeing as they said that 1800x can go even 4.1Ghz, and a bit higher with better cooling with the dynamic overclocking thing.
So my that's right on point of what I'd expect and agreed with my numbers. 162*4.3/4.0 = 174.15, then 220/174.15, that's 26.3%. But anyway, this is the 1800X you're comparing to, what is more appropriate is the 1700, as that's $20 cheaper than the 7700k.
Those I don't expect to see to be OC'ed as high (though I don't know the exact differences now, like what is the actual difference between the 1800X and 1700X, or is it just a scam? But let's say that OC's 300Mhz to 4.0Ghz turbo, and now you have Kaby Lake be 35.8% faster per core than Ryzen (assuming your original numbers are correct, I don't know what Kaby Lake OC is, but I'm assuming somewhere around 4.8Ghz).
So if we use that 35.8% figure, that's going to be significantly higher performance in games, and in a game 100% absolutely perfectly parallelized you'd get a 47.2% performance improvement from Ryzen. Anyway, I think the highest anything real world will achieve is say 30-40% percent, and Intel will blow games out of the water. But anyway, that's why I'll reserve my final judgement until we see benchmarks through the board.
Either way, it's a great job done by Ryzen, but in no way does make Intel obsolete, opposed to what I've been reading in some other places on the internet. It's just a tighter race now, and from my preliminary calculations and assumptions, I'd still go with Intel if buying my next CPU with the intention of gaming (at this budget).
I think the 1500 and 1600X are going to warrant some attention though, I think that they could outperform the i5's quite well at a similar price range.
Ryzen 1100 and 1200X doesn't impress me much, as I think the G4560 outperforms them at less than half the price, and the 1300 and 1400X are imo only marginally better for most things (than the G4560), and almost triple the price (kind of the same comparison as the original one, half the cores, the pentium cores are just as good if not marginally better, 4 threads will be just fine for most games as seen in the last video of the OP. Plus, the 6c/12t are only 30% more for 50% more theoretical power, so I think that upgrade is well worth it.
I don't know what Kaby Lake OC is, but I'm assuming somewhere around 4.8Ghz)
~5ghz @1.37v which is fine for air cooling. Some 4.9 and some 5.1 with good voltages but the OC variance is actually much smaller than usual so very few fall outside of that range
---
I am extremely interested in gaming & OC benchmarks plus just dozens of other benchmarks when the NDA lifts. So far i guess there is a substantial gap in ST performance between Kaby and Ryzen but the MT-perf/$ of Ryzen looks to be on another level~
Anyway, I think the highest anything real world will achieve is say 30-40% percent
Parallelization is basically a nonfactor for a lot of programs that can achieve values like 99% or 99.9% parallel and scale to hundreds of threads; 16 is nothing. Twice as many cores, twice as fast. Other times you have twice as many cores, 0% faster.
They've already shown Ryzen 8c16t @4ghz to be ~50% faster at cinebench MT than a 7700k @5ghz. Lower performance-per-core, huge frequency gap but twice as many cores doing work
On February 25 2017 05:45 FiWiFaKi wrote: Ryzen 1100 and 1200X doesn't impress me much, as I think the G4560 outperforms them at less than half the price, and the 1300 and 1400X are imo only marginally better for most things (than the G4560), and almost triple the price (kind of the same comparison as the original one, half the cores, the pentium cores are just as good if not marginally better, 4 threads will be just fine for most games as seen in the last video of the OP. Plus, the 6c/12t are only 30% more for 50% more theoretical power, so I think that upgrade is well worth it.
I'd like to see how an overclocked 1100 or 1200X performs compared to a locked i5. With the price difference between a Ryzen 1100 and say an i5 7400, an overclocked Ryzen may be better and cheaper than a locked i5.
The G4560 seems to kinda be in a class of its own. None of the Ryzen products seem to compete with it. I find it interesting that AMD isn't even trying to target the current budget king
I'm excited to see how Intel responds. They still have great products, so if we see a price drop in response to this, we all win!
Looks like ST perf roughly matches up with the leaks and overclocking is around 4ghz rather than 4.5
Very strong MT perf and MT perf/dollar but not very competitive ST and falling short in some games because of that
10% clock for clock + 20-25% freq difference = 1.32 - 1.375x slower on 1 thread.
Also curious to see if the memory latency thing is a software issue (not registering properly) or if it's actually broken in some way and could be improved
Several reviews talking about 5-15% performance loss when enabling SMT in games and we're still missing a lot of info, day 1 might not be the final story here.
I don't know what Kaby Lake OC is, but I'm assuming somewhere around 4.8Ghz)
~5ghz @1.37v which is fine for air cooling. Some 4.9 and some 5.1 with good voltages but the OC variance is actually much smaller than usual so very few fall outside of that range
---
I am extremely interested in gaming & OC benchmarks plus just dozens of other benchmarks when the NDA lifts. So far i guess there is a substantial gap in ST performance between Kaby and Ryzen but the MT-perf/$ of Ryzen looks to be on another level~
Anyway, I think the highest anything real world will achieve is say 30-40% percent
Parallelization is basically a nonfactor for a lot of programs that can achieve values like 99% or 99.9% parallel and scale to hundreds of threads; 16 is nothing. Twice as many cores, twice as fast. Other times you have twice as many cores, 0% faster.
They've already shown Ryzen 8c16t @4ghz to be ~50% faster at cinebench MT than a 7700k @5ghz. Lower performance-per-core, huge frequency gap but twice as many cores doing work
Honestly, I would love to see 7700k without the graphic part and instead more cores. IIRC(and it was a long time ago), 7700k has almost a half of it covered with graphic part that is useless to gamers. With 2 more cores(I don't think they would be able to add 4 cores and maintain the insane clock) it would be interesting. Probably not beating Ryzen in software application, but that gaming application could be awesome.
Well, I keep dreaming, in the meantime I can wait for new CPU war and wait for the response from Intel. (I don't need the upgrade now)
When we approached AMD with these results pre-publication, the company defended its product by suggesting that intentionally creating a GPU bottleneck (read: no longer benchmarking the CPU’s performance) would serve as a great equalizer. AMD asked that we consider 4K benchmarks to more heavily load the GPU, thus reducing workload on the CPU
From GamersNexus
they were doing this all of the time in their pre-launch benchmarks but i had considered it to be mostly due to incompetence rather than malice