review a moment of zen with david kanter the tr podcast 190

A moment of Zen with David Kanter: The TR Podcast 190

Duration: 1:05:33

Hosted by: Jeff Kampman

Special guest: David Kanter


RSS (MP3) | iTunes (MP3)

Show notes
The Tech Report Podcast made a guerrilla trip to San Francisco a couple weeks ago to talk about AMD’s Ryzen 7 CPUs and the Zen architecture, as well as the upcoming Ryzen 5 CPUs. Editor-in-Chief Jeff Kampman and special guest David Kanter also talk about the GeForce GTX 1080 Ti and upcoming developments in the graphics card market.

Follow us on Twitter – JeffDavidThe Tech Report

0 responses to “A moment of Zen with David Kanter: The TR Podcast 190

  1. The bulk of those gains will probably translate well to Vega, though. My main point was that Doom is one of the increasingly rare games that won’t see big gains from Vega’s Maxwell-style rasterization tech, and if it works as well as Maxwell’s, Vega might yet approach GP102 in median workloads.

  2. Doom was also extensively hand tuned for the GCN consoles, for instance sees one of the biggest benefits from async (three to five ms a frame, that is massive for a single feature).
    Winning the two big console design wins didn’t mean everything was instantly super optimized for AMD, but it did eke out some wins.

    [url<][/url<] They also said that some frames, all available cores were completely pegged. No wonder it's the one thing that has made my PS4 span spin up audibly accross the room!

  3. Drivers matter very much for a new CPU architecture, especially if a particular vendor isn’t supporting it correctly.
    [url<][/url<] Right now, Nvidia has some dx12 performance issues with Ryzen. This can't be blamed on the CPU, because dx11 is faster, and AMD does not experience the same slowdown that nvidia does in dx12. Why do I consider this important? Well, because if a Tech website decides to solely test in dx12 using Nvidia cards at this exact moment in time, those benchmarks will be skewed by a driver bug that screws up Ryzen's performance. Scenarios like this should be avoided at all costs, unless you are deliberately trying to fudge the numbers.

  4. I was specific to say ‘the TR way’, not necessarily TR. And drivers shouldn’t matter if we’re measuring a performance delta; same for game settings, unless they are so absurd as to not be remotely reflective of real-world gameplay.

  5. I wouldn’t count on seeing any updated TR benchmarks unless there is some new hardware released to review. Maybe like a motherboard paired with AMD certified 3200 ram.

    Also, Nvidia needs to fix their drivers before doing any new Ryzen benchmarks with their hardware, considering the weird performance issues when compared to AMD cards.

    TR doesn’t go out of their way to address early platform issues like that, so I’m fine with waiting for new hardware, and leaving the platform updates to other sites that care more about it.

    Frametime data is also meaningless if you bench with a bad driver, game, or hardware settings, and it’s not like other sites don’t ever use it. Depending on TR for every single data point is missing the forest for the trees. You just end up getting their limited view of an issue.

  6. This needs to be tested the TR way to be confirmed.

    (and I’m not saying that this is a bad thing- I’d forecast that the improvement would be realized in real-world gameplay- but the claim doesn’t bear full weight until verified with frametime data)

  7. Guru3d did a memory performance test, and showed that using low latency 3200 ddr and overclocking can boost performance up to [b<]29%[/b<] in some games with the 1700. Of course that isn't going to apply to every title, but it certainly brought Ryzen's general gaming performance up. Benchmarks that don't use that configuration aren't going to reflect those gains, so it's not like Intel is actually that much better, more so that Ryzen is being benched on a suboptimal platform. Ryzen's launch was really subpar, for a number of reasons. Poor ram compatibility, OS issues, driver/game issues, Bios issues, etc. Once those are dealt with, I don't see it worth going Intel for only a [i<]slightly[/i<] better single thread performance. I'd like to see some benchmarks of the 1600X with 3200 ddr. That CPU may be the best bang4buck ryzen chip for gaming, while the 1700 is better for multitasking. Bit of a toss up for me until I see some reviews, but am leaning toward a 1700 build atm. Waiting for better availability, updates, and benchmarks before making a decision though. Can't go wrong with extra cores, especially when there is little to no difference at 1440p, and those extra cores mean background tasks aren't going to conflict with playing a game.

  8. So, to go back to the “streamer” part of the discussion:

    I’m thinking such a usage scenario involves running a game on one window while tracking encode/decode work in another along with viewer/subscriber related functions and so on.

    Does such a usage scenario not lend itself to utilizing the extra threads? When would it make sense? Strategy (Civ6 etc) gaming oriented channels perhaps?

    Not to say there’s anything wrong with using the product in general as an alternative, its performance is quite respectable even in single thread scenarios, but just like to understand the potential for areas of strength better.

  9. You should check out TR’s reviews of the Ryzen CPUs and of various GPUs. TR tests smoothness as well as pure “frames per second” performance in game tests, and typically prefers a smoother experience at a lower framerate (say, 50 FPS) over a less consistent experience at a higher framerate (say, 60 FPS). That’s a generalization, but it represents the fact that hitches and stutters have a worse effect on enjoyment of the game experience than a somewhat lower average framerate.

    The problem with saying that Ryzen [i<]necessarily[/i<] has "good enough" gaming performance is that when you pair it with a really fast GPU you end up with an experience that just [url=<]isn't quite as smooth as the Intel offering.[/url<] As David has said himself, client tasks (that is, 99% of everything you do on a desktop PC) are much more responsive to system latency than to compute throughput, and the Ryzen 7 chips land further on the compute side of things than Intel's do. So, [i<]specifically for gaming[/i<], a Ryzen 7 chip might be "good enough", but it's still unquestionably inferior, and more to the point, the Intel chips that are better are the same price or cheaper. That's why David's not that impressed.

  10. Great to hear a new podcast with The Kanter! Thanks for making it happen!

    Stuff like Kanter’s discussion on GDDR5X not being as much of a benefit as it seems due to underlying architectural reasons is fascinating and not discussed, let alone mentioned anyplace else. Cool info like this is what makes these podcasts shine.

    Sounds like there’s room another another podcast post-Vega and once more GDDR6 details are known discussing GDDR6, Vega’s HBM2, and GDDR5X.

  11. Higher speed memory support would be nice, but if they’re going to change something, it should be to work on that inter-CCX link first- it needs to be unhitched from memory speed and latency needs to be improved, assuming they’re going to keep the four-core-per-CCX building blocks.

    Ryzen is already pretty bandwidth efficient overall.

  12. Jeff,

    Thank you for the podcast! I’ve missed them and missed the ones with David Kanter in particular.

    For me, this made it likely that another incremental upgrade is all that is warranted right now, as much as I really would like to build a wholly new system. In doing so, my listening has paid for the annual subscription. The one thing that would change my mind is to compare nvme against my current sata ssd’s, as I don’t have m2 slots on my aging system.

    Thanks again.

  13. Considering price an aspect of architecture strikes me as odd, as price mostly isn’t determined by architectural factors. Architecture has a lot of bearing on costs to the manufacturer (marginal and otherwise), but marginal costs clearly aren’t the biggest factor in setting the prices that consumers see, and trying to recoup R&D is complex enough that we can’t tell much about it from here.

    It probably still costs AMD as much or more to fab a Zeppelin die than it would for Intel to fab an 8C whatever-lake equivalent. It just doesn’t matter because Intel’s marginal profits are so massive that AMD can make money by undercutting them on price anyway.

    Despite that, I have to disagree on Broadwell being all-around superior. The problem is that AMD is still doing that thing where they factory OC everything to the moon, and it’s still wrecking their perf/watt. Pitting a down-clocked Zeppelin against a Xeon-D looks like a solid perf/watt win for AMD, and that’s the voltage range more relevant to servers and laptops.

  14. I think AVX’s future relevance is not exactly a foregone conclusion in most markets, given the absurdly cheaper FLOPS/$ of SIMT-driven GPUs on non-branchy code.

    GPU acceleration is not completely ubiquitous yet, but among most heavier workloads it feels like it already has more traction.

  15. [quote<] an 8 core broadwell part [/quote<] So... an EE or Xeon? No, an 8 core broadwell part does not exist for the average consumer. AMD wins because they're the only one playing in the realm of higher than 4 core x86 CPUs for consumers.

  16. Intel [i<]Whatever[/i<]-Lake Can i quote you on that Jeff? 🙂 Seriously though, I really appreciate that you did the Podcast. I hope you got some well deserved rest afterwards. I'll recently re-subscribed.

  17. For folks that live in a “cost-is-no-object” world because it’s an enterprise application or whatever, then you’re right. Broadwell works. For cheap-ass folks who want a better bang for the buck, price *and* performance matter. In suboptimal workloads it’s equal or even still slightly better performance that high-end Broadwell. In optimal workloads, it’s far better.

  18. Flew to Sacramento then drove to San Fran, not bad not bad.

    Not quite Timmins to the southern Govi though 😉

  19. Broadwell is flat out a superior architecture except for product price (as of today’s date) on higher-end parts.

    Furthermore, an 8 core broadwell part is much more future proof than Zen since the AVX support is never going to get any better than what you have right now, and even without additional tuning Broadwell’s AVX support is already a major advantage in many high performance workloads.

    Anybody who thinks otherwise ought to actually read TR’s review.

  20. Double, not same. Ryzen gives double perf for the price. It’s much different from Broadwell: it’s better.

  21. Ryzen no doubt tremendously closed the gap with Intel, and if anything, I’m just amazed at how AMD catapulted themselves back into contention in one fell swoop. And with a budget the size of Intel’s quarterly snack bar expenses at that.

    But even so, it’s obvious Ryzen still has lots of rough edges and it’s far from reaching its full potential. This, coupled with the fact that I don’t need an upgrade at this time means I can easily pass up Ryzen 1.0. I also don’t like how it’s up to developers to optimize every game out there. Yes I know optimization is important and it’s necessary but the fact remains that at this point I don’t need to worry about these things on an Intel system and the fact is, not everyone will optimize. You guys know how partial I am to AMD but here is where I draw the line between being an AMD fanboy and sensibility, unlike some folks out there who’ll root for AMD no matter what. No doubt Ryzen is a very compelling CPU but I think it just needs more work.

    Here’s a link that pretty much sums up the situation for Ryzen too. Check it out.


  22. They say that Ryzen is memory bandwidth limited but AMD can improve higher memory frequency support to 3600Mhz or more in the future. Lets not forget that DDR4 memory is still not maxed out so there is room to play and remove this bottleneck that indeed exists on the 8-core Ryzen CPUs.
    Second thing, you insist on the fact that you cannot find a real scenario where you would need the entire power of the 1800X, but on the other end you praise 7700K for being the best for gaming. Well, that is a bit counter-intuitive, because the same can be said for the 7700K in games. Who would need more than 100FPS in any game, right? Just get the R7 1700 and enjoy your games.
    I think you are being a bit biased towards Intel because you expect that all of a sudden AMD will surpass Intel on every single metric. This is not possible or at least it didn’t happen with Ryzen. But at least you can be objective and see the bigger picture here:
    1. Ryzen has more than good enough gaming performance and it is comparable to 7700K even though in some games the gap is big, there is no case where you could say that using Ryzen CPU result in unplayable experience.
    2. At the end of the day, any buyer expects to get as much life as it can from its purchase. And in all fairness, with 7700K you get 1-2 years of maximum gaming performance (which is not very far off the Ryzen proposition), but after that you will see the lesser cores in a lot of scenarios and not only gaming. The same story was with quad cores when they started to appear. There was no actual need for them at that time, but in the end those who chose a quad core have it even today (see 2500K owners).
    All in all, I don’t see any reason to buy 7700K over the 1700. 1700 has almost the same performance as 7700k in many games , but trounces it in computing workloads, has the same power consumption, will be supported for longer (AM4 socket will live for at least 2-3 generations, so you can just upgrade CPU), whereas with 7700K you will have to change the MB when cannonlake appears, so more money thrown out of the window.
    I am actually surprised that David is not very impressed with what AMD has accomplished.

  23. Kanter is right (as usual) with his analysis of RyZens plusses and minuses.

    As I said somewhat jokingly (but also correctly) RyZen is a “60% of the time it works every time” chip where certain workloads are basically at the same level as Broadwell but other workloads just don’t benefit as much.

  24. I really don’t think the AMD’s Ryzen 7 CPUs should be priced below the 7700k, um it blows it away in highly threaded work loads. Doesn’t Intel have its 8core 16 thread priced higher than 7700k…?? Seriously. The 4core 8 thread Ryzen CPUs will be priced lower than the 7700k. Plus, the 3200mhz RAM speeds have been showing an average of 10fps increase in games at 1080. We have also seen improvements with the latest BIOS updates. From a multi threaded stand point, the feel the Ryzen 7 CPUs are a clear win. Plus, gaming performance is not bad. It is actually quite good, especially compared to the previous gen CPUs.

  25. I expect vega to be slightly faster than 1080 with greater power consumption at the same price. I don’t think it’ll be as fast as 1080ti. I hope I am pleasantly surprised.

    Jeff and Kanter were quite concise in expressing ryzen performance too. depending on the workload ryzen fluctuates between ivy bridge, haswell, and broadwell performance. It is interesting. another thing I was wondering is if amd’s multithreading implementation is better than intels and when they talked about ring bus scaling that helped answer my question.

  26. [quote<][...] Game developers are not going to be able to really effectively use more than four cores until we get a new generation of consoles.[/quote<] A Jaguar core as seen in consoles has quite a lot less than half the power of a modern high-clocked big core. I think last time I ran the numbers, I found a PS4 to be similar in CPU power to a 2C4T Haswell at two-point-high GHz. That doesn't mean that a G4620 is automatically plenty to run a console port, though, because consoles tend to run games at 30 fps. If you've got a game that can use six or eight threads (being designed for PS4/XB1) and you want to run it at 100+ fps, the Zen strategy of trading a bit of single-threaded performance for more than four cores isn't inherently a bad one. One issue with this that I see all the time is that game logic and rendering loads are different. For instance, I've been playing some Borderlands with my brother lately over a consistent but highish-latency internet connection (which can increase CPU load too). I've got a G3258 at 4.2 - all the single-threaded performance, but only two threads. He's running Vishera, and has all the threads but not much single-threaded performance. It seems that Borderlands multithreads a lot of things, but not the core game logic that needs to be networked. When I'm hosting and a lot is happening in-game, my framerate gets choppy (the game logic thread displaces rendering work), but everyone else's experience is fine. When he's hosting, his framerate is fine (as he's got spare threads to handle it), but everyone else gets issues similar to but distinct from packet loss when his system has a lot on its plate. Point is, games do various kinds of work, and not all of them scale the same way. Throw memory bandwidth in as a factor, and it isn't too surprising that gains are so far iffy. That doesn't mean that games won't be able to make good use of 8C Zen or similar going forward - it definitely isn't as much work to make that happen as multithreading in the first place is. On Vega being 1080-class: wasn't Doom the main game they were testing that with? Thing is, Doom has a forward renderer. Nvidia has their tiling stuff to help out deferred renderers, and that leaves AMD + Doom looking like a good combo right now (there are more reasons, but that's a really big one). Vega is supposed to be introducing some of AMD's own stuff to help out deferred renderers though, right? I can see why they would test with Doom if they were still refining the tech, but it doesn't strike me as the best benchmark for Vega's real capabilities.

  27. Guys, I really hope you include power consumption of the r5 for the curious people out here! I appreciate the work!

  28. Either you got a better microphone or have been working on your radio voice. Nice and smooth. Like a good cup of joe. Keep on rockin in the free world.