Discussion: AMD Ryzen

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


wow. Didn't think AMD was THAT far behind.
 
Do we know or have guesses on zen pricing? I was thinking of getting a Haswell i5 or i7 in about of month but if zen is going to be cheaper it might be worth waiting for?
 


Well, our guesstimates put the 8 core version of Zen at -E 6 core levels. That is ~USD$550. And it will trickle down from there. The 6 core versions of Zen could land at 400 within range of the i7 non-E.

Keep in mind there is no information currently that even points in that direction, so this is based on projected performance and marketing from AMD.

Cheers!
 


Aren't you ignoring the performance boost from L3 cache in Zen that is not present on the Athlon?
 


Is not the 40% quoted by AMD already considering the L3?
 


Well, the truth is, excavator is much closer that the 40% behind, but it is not a platform for desktop. If you consider this, the last platform AMD put out on AM3+ was actually Piledriver, and a refresh. Piledriver itself is now as old as Ivy Bridge.

What I expect from AMD from Zen, based on what I hear, and some extrapolating is the following:

Flagship 8 cores 16 threads
base clocks between 3.3-3.5 (moderate overclock ability, not exceptional, not terrible either)
Turbo core between 3.7 and 3.9
95W TDP
Single thread performance that is very similar to Broadwell-E
Multi-thread performance that is very similar to -E series processors of similar cores clock for clock with lower stock TDP than E series.
Will probably be priced competitively for performance/value.

AM4 MBs will also be much less than E-series boards on the low end. High end boards will probably be equal to cost as high end intel options for non E series processors (as quad channel ram is not a thing in this generation of Zen, it may be later, but I have no knowledge of such things at this time).

Zen+ will be an iteration that will likely get AMD even closer in single thread performance, multithread should be quite competitive out of the gate from Zen.

AM4 will support DDR4, DP 1.4, and all kinds of cutting edge tech.

 


Both the X4 845 CPU and the Bristol Bridge APUs for AM4 are Excavator for desktop.



Others expect base clocks in the 2.6--2.8GHz range, turbo somewhat around 3.5GHz and very small overclocking room due to strict VRM requirements and process node limits.



Official data given by AMD (40% higher IPC) suggests a lower performance even when considering benchmarks that favor Zen, such as CB; see computation of scores above, where I assumed that Zen can reach 3.8GHz. The IPC seems more close to Sandy Bridge than Broadwell.

I say benchmarks as CB are favoring Zen because CB only uses half of the 256bit execution units in Haswell/Broadwell). If we use benchmark/applications that can use full 256bit vectors then Broadwell-E will be much faster than Zen, which is expected because Broadwell is a 32FLOP/core microarchitecture whereas Zen is only 16FLOP/core. It is fair to remark that many desktop users will not use applications that can fully exploit 256bit vectors, but other users will do and will see the difference.



Assuming that the Orochi-Zen slide given by AMD means 2x faster than FX-8350 (as many tech media supposed) this still means that 8-core Zen has performance similar to 3.4GHz 8-core Ivy Bridge (see computation of scores made above).

It is also worth to remark that we cannot compare TDPs directly because Intel and AMD measure it differently. Check the next graph. The 8-core Intel chip rated at 140W TDP is more efficient than the 8-core AMD chip rated at 125W TDP.

Power.png




Maybe or maybe not. What would be cheaper are the AM4 boards.



Assuming that Kabylake or Cannonlake don't improve enough.



Agree for DDR4. Source for rest of your claims?
 
OK. Everything you all said sounds terrible. If I was the marketing guy from AMD, this sh*t would give me nightmares.

That is why I am so confused. If it is supposed to be that terrible then why do they even bother? I mean, after so much time since they first teased Zen, why release a fail product? That sounds like the final nail in the coffin. Unless the pricing is super aggressive.
 


Actually zen could be faster than skylake.

If it is true that mainstream zen CPUs are going to pack 8 cores and 16 threads, then it is very possible that Zen could beat an i7 6700k. But they probably will still be behind broadwell-E. But this is only for heavy threaded apps.
 


I wouldn't call it mainstream but enthusiast, but yes it is 8 cores and 16 threads

IMG_0978-1024x768.jpg
 


You should understand a few things first...

The difference between ivy bridge and skylake clock for clock is about 20%...over 3 generations of chips...they gained 20%.

Zen is going to gain 40% in one leap. Yes...40%...which is quite unheard of in modern times.

Intel has hit a wall with increasing IPC, so they focus on clock speed. AMD is slightly behind there, but as the process matures, they will get closer with Zen+. Even a 15% gain over regular Zen would put them in the same ball park as skylake. Kaby lake/cannon lake will not be a huge step forward in IPC, only gains in clock speed mostly from what I have read.

Essentially...we are at a point with Zen, and modern intel processors, that 10% is almost unnoticeable to the average user. You will see a bigger increase in performance relatively by going with an AMD Zen CPU and adding a solid state drive for the difference in cost, than you would by actually going intel.
 
It's time to stop expecting Zen to compete with the top binned Intel parts, and instead look for them to compete at the median frequencies where the bulk of CPU sales are. And by compete I mean also take price/performance into consideration. DX12/Vulkan is becoming more prevalent by the month and the bulk of games are developed with 8 core consoles in mind anyway. Go to a retail store today and look at the speeds on the "Performance Desktop" and "Gaming desktop" configurations. You might be surprised. There is a huge difference between the clock speeds of an i5-6400 and an i7-6700k.

Also one of the biggest complaints against Bulldozer should become a non-issue. The whole power supply disparity. No longer will you need an extra beefy supply for the FX chips.

All in all, Zen should be a heck of a lot more marketable than Bulldozer was.
 

Compare an i3 with a same clocked pentium,the i3 also has 30-40% more IPC depending on what bench/prog/game you run but single thread speed is still the same.
40% increase could come purely from full FPU's/SMT.
Until ZEN get's released speculations run wild.
 
It may be true that amd is making a 40% leap in one generation but we have to factor how long it took them to get there. It would sound marvelous to say a nation went from hot air balloons to space travel in one advancement but if that took the nation 100yrs to do so? Well surely I'd hope they made a leap like that. Something to show for the ages they did nothing.

Intel has one of two choices, either go full speed ahead and bury amd so hard they may as well pack it up and go home. Or they can release incremental upgrades since they're already leading in performance. Either way people will complain that it's either unfair or no advancement. Intel also hasn't hit a wall with ipc, the newer skylake cpu's perform better (slightly) than haswell and they're clocked slower. That says just the opposite, that ipc is improving enough to be able to lower the clocks rather than pushing faster ones.

In terms of pricing I would expect amd's prices to match their performance levels. That's more or less how it's been and how just about any typical business model is run. You don't want to over charge customers and turn them away. At the same time you're selling product to make money and with amd's pile of debt they need all the money they can get. If it were your product why would you sell it for less than the market would bear? No one would. Amd's prices have been low out of necessity rather than generosity. If zen can start competing with say haswell, I would be reasonable and expect haswell prices.

Amd is playing catch up in many ways and isn't even bringing much in the way of innovation to the table. In the past they could claim things like fastest clocks yet or highest core count yet or something. Intel will have 10core cpu's on their enthusiast boards. They've already been bringing ddr4 to the enthusiast class for some time now and already brought it to mainstream boards. A few months ago earlier in 2016 amd finally brought usb3.1 and m.2 to their boards, just as they were about to get replaced by zen/am4. Granted things like smt is new compared to what fx has had but not new at all. Intel has been beating them to the punch on many fronts.

I'm hoping amd pulls through this time around in terms of performance, bulldozer was a massive letdown. Honestly even though I find it interesting how the tech works, I really don't care about the magic behind the curtain as much as I do the actual results. Whether they use 2 levels of cache, 3, 4, hybrid hbm, so long as it gets the job done. All the marketing hype means squat, how many fps? How much time is it going to take to encode a given video compared to the competition? That's the sort of data we can only get once it shows up on store shelves.

The same applies to intel as well. Both companies have a history of getting hopes up with all the hype. Internal early release 'demos' mean little as well since there hardly seems to be much info given as to exactly how they managed to achieve the results. Program an in house test or demo to show how fast a cpu is but then come to find out in real programs that we're actually using the difference is much less spectacular.

 
a few question tho.

first. that 40% leap is on single threaded performance right? if it is. wouldn't that make them comparable to today Intel? (not saying it will match it but even slower but it will not bottleneck anything for the next few years?)

2nd. yes Intel have 10 core but the price is 2279.99. and let say the 8 core AMD zen come out to be 359.99. or 399.99. wouldn't you go for AMD?

3nd. the reason why the FX is dead it caz you can't upgrade it anymore. but the Zen will be running on AM4 motherboard. so you have years to come of upgrading cpu?
 


You are confusing IPC with raw performance.

i3 has hyperthreading, so it technically does more work, but IPC is a different animal, and relates only to the capabilities of the core itself at 100% utilization. Same architecture chips all have the same IPC, as long as they have the same cores on the chip. SMT allows them to do more work, but is unrelated to IPC.
 


Just FYI...Intel is doing all they can already to "bury" AMD. Their engineers are legitimately running into limitations of their uarch and manufacturing process...

The myth that "Intel could bury AMD at any time, if they wanted" is not a reality. If AMD did not exist, Intel would still be selling pentium 4 processors at 500 MHz, but asking for $1000 per chip for the top SKUs because no competition. Make no mistake, Intel is a business, and they are ruthless about being a business. If they can do something to get a leg up (including illegal activities, see AMD v Intel anticompetitive lawsuit), they will. You need to be aware that Intel is not some space age technology company, and AMD are not some backwoods group of guys working in shacks trying to figure out Intel's greatness. The fact that AMD is even in the same ballpark as Intel with 10% of the same R&D budget is a testament to the talent level of people AMD has working for them.



In terms of pricing I would expect amd's prices to match their performance levels. That's more or less how it's been and how just about any typical business model is run. You don't want to over charge customers and turn them away. At the same time you're selling product to make money and with amd's pile of debt they need all the money they can get. If it were your product why would you sell it for less than the market would bear? No one would. Amd's prices have been low out of necessity rather than generosity. If zen can start competing with say haswell, I would be reasonable and expect haswell prices.
Click to expand...

This is actually a pragmatic stance, and reflects reality. Zen will be priced competitively in regards to what they expect to be the competitive options from Intel.

Amd is playing catch up in many ways and isn't even bringing much in the way of innovation to the table. In the past they could claim things like fastest clocks yet or highest core count yet or something. Intel will have 10core cpu's on their enthusiast boards. They've already been bringing ddr4 to the enthusiast class for some time now and already brought it to mainstream boards. A few months ago earlier in 2016 amd finally brought usb3.1 and m.2 to their boards, just as they were about to get replaced by zen/am4. Granted things like smt is new compared to what fx has had but not new at all. Intel has been beating them to the punch on many fronts.
Click to expand...

While true at this point...I would also point out, when AM3+ launched, it was the most advanced platform available, and prompted the competition to upgrade their own platform to add the cutting edge features...so it does go both ways.

I'm hoping amd pulls through this time around in terms of performance, bulldozer was a massive letdown. Honestly even though I find it interesting how the tech works, I really don't care about the magic behind the curtain as much as I do the actual results. Whether they use 2 levels of cache, 3, 4, hybrid hbm, so long as it gets the job done. All the marketing hype means squat, how many fps? How much time is it going to take to encode a given video compared to the competition? That's the sort of data we can only get once it shows up on store shelves.
Click to expand...

This is the relevant part. Real world performance...once people see how close it really is, I think there will be less concern.

The same applies to intel as well. Both companies have a history of getting hopes up with all the hype. Internal early release 'demos' mean little as well since there hardly seems to be much info given as to exactly how they managed to achieve the results. Program an in house test or demo to show how fast a cpu is but then come to find out in real programs that we're actually using the difference is much less spectacular.
Click to expand...

Yes, synthetic benchmarks are a bit dishonest, in the sense that synthetic benches can be structured to show specifically one thing, and very specifically not other things.
 


The ~20% IPC gain between ivy bridge and skylake is only valid for legacy workloads or modern workloads that don't benefit from the new SIMD units and companion tech. When you use software that benefits from the new hardware units and instructions found in latter microarchitectures then the IPC increase is much much larger even on a single generational jump...

I had a couple of new Haswell systems set up so decided to run the same benchmark code on those too. I was stunned by the performance of the Haswell processors! I tested a few other systems and found that Haswell gives about a 70% speedup over Ivy Bridge at the same core clock speed. I was surprised by that because what I was hearing about Haswell was that it was generally only about 10% (or less) faster than Ivy Bridge. It looks like the new AVX2 instructions on Haswell are really effective.
Click to expand...

^^^ I bolded the relevant part.

But let us assume that the gain between ivy bridge and skylake was 20% because most PC users don't care about the new instructions neither have software that uses the new instructions. It is fair to say that AMD is in the same situation, and a pair of days ago you acknowledge yourself that Excavator is about 20% faster clock-for-clock than Piledriver.

Those incremental gains are the result of tweaking/evolving an existent microarchitecture. Steamroller can be seen like an improved Piledriver and Excavator as an improved Steamroller...

The 40% gain introduced by Zen has a completely different origin. Zen is a new, from scratch, architecture. Zen is not based in the Bulldozer lineage. Moreover, the Bulldozer family is a speed-demon microarchitecture. Speed demons are optimized to get highest possible clocks at the expense of lower IPC. On the other hand Zen is a brainiac design; brainiacs aim to obtain highest possible IPC at the expense of lower clocks. The difference between speed-demons and brainiacs is a classic distinction in computer architecture. Getting 40% higher IPC for the transition from speed-demon to brainiac is not anything special or remarkable. In fact so early as 2014 I predicted Zen was going to have ~40% higher IPC than PD, and I wasn't far according to AMD official claims.

The transition from Bulldozer lineage to Zen lineage is similar to what happened before when Intel switched from Netburst (speed-demon) to Core (brainiac).

Zen is not out still and you are already introducing Zen+? I don't agree with your claims about Zen+. But since this is a thread about Zen, let us left that for another occasion.

Pricing is not known at this point. It is possible that we can obtain "a solid state drive for the difference in cost". It is also perfectly possible that AMD goes completely crazy with Zen, like when priced original the FX-9000 series.
 


Yes, Intel is a business and AMD is as well...

The myth that Intel would still be selling pentium 4 processors at 500 MHz, if AMD didn't exist is too extended, but just a myth. AMD has not been competitive for a long while, but Intel has been improving its products in all the categories. The evolution of the Core architecture up to the current Skylake iteration is a consequence of Intel fighting others in the server space. E.g. the 18-core Haswell Xeons were released because Sun/Oracle and IBM have competitive server products not because AMD Opterons are competitive (AMD barely has 2% of market).

Silvermont and the rest of Atom line were developed because Intel is fighting Apple, Qualcomm, and other players in the mobile space.

The Xeon-D family has been developed to fight APM, Cavium, and other ARM companies in the microserver market.

The Phi Xeon line has been developed to fight Nvidia in the HPC market.

And so on...
 


Just because Zen will come with 8 cores at 359 doesn't mean it's anywhere close to the speed of an 8 core or 10 core intel extreme edition.

The IPC count on the Broadwell-E and soon to be Skylake-X CPUs is so much greater than the mainstream arena which is why it's so expensive (and worth the price).

It's basially comparing apples to oranges. Zen's 8 core CPUs I'm pretty sure are going to be targeted at the mainstream market, meanwhile the intel 8 and 10 core CPUs are targeted at enthusiast/power users. However if those 8 core Zen CPUs are indeed enthusiast class, then they'll be priced at $700 or more. (I'm just theorizing prices here since there still isn't much ground info on Zen yet.)
 
When talking about IPC, remember to also state the instruction set used (instructions per clock, remember?). It's different when we discuss AVX, AVX2 and SSEx (pun not intended)!

If we talk about "general available software", then keep it at "AVX1" level. If you want to put Pentiums and Celerons, then just use SSE3/4, cause they don't have AVX enabled (!) and will default to that codepath.

Back to the 40% statement. I would imagine is under a regular floating point operation metric, probably using AVX1. If they are talking Integer performance, then AMD is gonna be screwed again in the consumer market.

Cheers!
 
Hey, it's fully possible Zen will clock at 6Ghz and defeat every Intel chip ever made, but it's also possible it will match performance of the Intel Ivy Bridge design from 4 years ago. We don't know much about its performance yet and as far as I know even AMD isn't sure because the test chips they've gotten back don't clock anywhere near the speeds they want them to.
 


No, it is not. People already has engy samples and know the performance baseline for Zen. Also, AMD knows very well where it will land right now. I bet they are now just tweaking final specs for the CPU itself that have nothing to do with it's actual innards (as in uArch) and are already working on Zen+ (let's call it that for now).

We are speculating with whatever information we can find and comparing it to our collective knowledge and understanding of things.

Cheers!
 
True it's all just a guessing game. Most of the speculation I've seen from people in general so far has been highly optimistic from the consumer standpoint. Well amd could have a ton of cores and smt and higher ipc and then sell them dirt cheap and force intel to drop prices. If amd manages to compete directly with intel's chips they'll compete on price also.

So we may see some slightly improved fx version 2 chips in terms of performance still around the $160-180 range or we may see amd pull through and deliver near i7 performance - but at the $280-300 mark. I wouldn't expect amd to pull out i7 performance in zen and then turn around and sell them for $185. Of course that would be wishful thinking as a consumer and a great deal but so would a $5k ferrari. They don't exist.

I think amd fans will stick with amd and intel fans will stick with intel. At the end of the day they'll be priced accordingly and no major advantages in terms of price to performance. Hopefully amd has been doing their homework and not being overly hopeful themselves. That's what happened with bulldozer, they hyped it for so long and began before engineering samples were available. Then the end results fell flat, if they had truly known how it would perform in the real world they wouldn't have hyped it like they did.

I'm seeing some of the same telltale signs here, zen's wildly inaccurate perceived performance charts have been around far longer than they've had es chips to play with. Rather than tease the crowd and speculate knowing the cards in their hand, they're rolling the dice and hoping they don't flub again.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom