A first look at AMD’s Kaveri APU for notebooks

Five months ago, AMD pulled the curtain back on Kaveri, its latest full-fat processor for desktops and laptops. The first variants of this CPU were targeted at desktop systems, but today, AMD is unleashing some new Kaveri flavors aimed squarely at notebooks.

Like their desktop brethren, these chips combine AMD’s Steamroller CPU cores with the same Graphics Core Next architecture featured in the newest Radeons. Mobile Kaveri chips also fully implement AMD’s Heterogeneous Systems Architecture, which aims to simplify the development of applications that share work across the CPU and GPU cores. Various other goodies round out the mobile Kaveri package, including a TrueAudio DSP block, better video decoding, PCIe 3.0 support, and improved power management. The power management changes involve better monitoring of temperatures and activity counters, which all works to help the processor reach higher speeds via Turbo.

Architecturally speaking, there’s no difference between the mobile and desktop versions of Kaveri. They’re based on the exact same silicon. The key differences are in binning, we’re told, and in the way the parts are “optimized and fused” to fit inside different thermal envelopes. While the TDPs for desktop Kaveri processors range from 45W to 95W, those for mobile Kaveri models max out at 35W and go as low as 15W.

AMD has put Kaveri on a crash diet, in other words, allowing the chip to fit a whole new wardrobe: everything from full-sized notebooks to ultra-thin laptops, including systems that might be called ultrabooks… if Intel hadn’t trademarked the name.

A few weeks ago, we attended an AMD press event in San Francisco, where we benchmarked one of the very first prototype notebooks based on Kaveri. We also got briefed on the upcoming mobile Kaveri lineup, which will comprise 10 models in all. Let’s start by dissecting the lineup, and then we’ll delve into our benchmarks.

Kevin Lensing, AMD’s Senior Director of Mobility Solutions, introducing Kaveri for notebooks.

The lineup

The mobile Kaveri family is split into three clans. The first includes standard-voltage processors with 35W TDPs:

Model Modules/

Integer

cores

Base/

Turbo

speeds

Total

L2 cache

capacity

Graphics

CUs

Graphics

clock

Max

DDR3

speed

PCIe

lanes

TDP
FX-7600P 2/4 2.7/3.6 GHz 4 MB 8 686 MHz 2133 MHz 16 Gen3 35 W
A10-7400P 2/4 2.5/3.4 GHz 4 MB 6 654 MHz 1866 MHz 16 Gen3 35 W
A8-7200P 2/4 2.4/3.3 GHz 4 MB 4 626 MHz 1866 MHz 16 Gen3 35 W

The fastest offering of the bunch, the FX-7600P, has all of its CPU and graphics units enabled. The CPU cores run at up to 3.6GHz thanks to Turbo Core. That’s actually not all that far off from the 4GHz peak speed of the A10-7850K, the fastest desktop Kaveri, which has a 95W TDP. The 7850K’s base speed is much higher, though, at 3.7GHz, and its peak graphics frequency is 720MHz.

Note the first two letters of the FX-7600P’s model number. Until now, the FX brand has applied solely to desktop processors without built-in graphics. AMD told us that, since it’s delivering an “enthusiast-level experience on mobile,” it now wants to “parlay” the “enthusiast goodwill” it’s accumulated in order to leverage the brand into its mobile lineup—or something along those lines. The marketing-speak was a little fuzzy, but the gist is that the FX name has gone freelance, and AMD will apply it to whatever processors it sees fit.

Competitively speaking, AMD expects the FX-7600P to fight it out with Intel processors ranging from the Core i5-4300M to the Core i7-4600M—all dual-core, quad-thread, 37W offerings.

The FX label also makes an appearance in the ultra-low-voltage Kaveri series:

Model Modules/

Integer

cores

Base/

Turbo

speeds

Total

L2 cache

capacity

Graphics

CUs

Graphics

clock

Max

DDR3

speed

PCIe

lanes

TDP
FX-7500 2/4 2.1/3.3 GHz 4 MB 6 533 MHz 1600 MHz 8 Gen2 19 W
A10-7300 2/4 1.9/3.2 GHz 4 MB 6 533 MHz 1600 MHz 8 Gen2 19 W
A8-7100 2/4 1.8/3.0 GHz 4 MB 4 514 MHz 1600 MHz 8 Gen2 19 W
A6-7000 1/2 2.2/3.0 GHz 1 MB 3 533 MHz 1600 MHz 8 Gen2 17 W

AMD maintains that the FX-7500 is powerful enough to compete with the Core i7-4500U, one of Intel’s fastest ultrabook processors. According to AMD, these two chips perform identically in PCMark, but the FX is speedier in graphics and GPU computing applications.

The FX-7500’s 19W TDP does exceed the i7-4500U’s 15W envelope by a few watts. However, because of its lower thermal density, the FX-7500 can apparently squeeze into systems with the exact same cooling solutions as the i5-4500U. The 19W figure isn’t set in stone, either. AMD says ultra-low-voltage Kaveri variants can be configured to fit inside TDPs from 15W to 25W. We’d expect some performance to be sacrificed at the lower end of that spectrum, though.

As for battery life, AMD expects to be “neck and neck” with the competition on that front—except, it says, in cases where Intel works “very closely” with system builders to maximize power efficiency.

We should note that the FX-7500 requires less power than AMD’s previous flagship for ultra-thin notebooks, the 25W A10-5745M. AMD’s own numbers indicate that these two offerings perform similarly in PCMark, but the FX-7500 is about 8% faster in 3DMark and 5% speedier in Basemark CL.

At the tail end of the ultra-low-voltage series, the A6-7000 is the most pared-down mobile Kaveri variant by far. Half of its CPU cores are disabled, as are three quarters of its L2 cache and five of its eight graphics compute units. Those sacrifices make it the lowest-power unit in the bunch, which could count for something, but we wouldn’t expect stellar performance.

Rounding out the mobile Kaveri lineup are three ultra-low-voltage chips for professional applications:

Model Modules/

Integer

cores

Base/

Turbo

speeds

Total

L2 cache

capacity

Graphics

CUs

Graphics

clock

Max

DDR3

speed

PCIe

lanes

TDP
A10 Pro-7350B 2/4 2.1/3.3 GHz 4 MB 6 533 MHz 1600 MHz 8 Gen2 19 W
A8 Pro-7150B 2/4 1.9/3.2 GHz 4 MB 6 533 MHz 1600 MHz 8 Gen2 19 W
A6 Pro-7050B 1/2 2.2/3.0 GHz 1 MB 3 533 MHz 1600 MHz 8 Gen2 17 W

The models in this series have the same specs as the FX-7500, A10-7300, and A6-7000. One might reasonably expect the Pro variants to have the same driver-level certification for professional design apps as AMD’s FirePro graphics products, but curiously, that’s not the case. AMD says it could offer such drivers, and it’s “had discussions” with partners about the matter. Right now, however, drivers certified for pro apps are “not a standard part of the Pro series.”

Hmm.

The Kaveri whitebook and the test

AMD seems to be making a habit of holding events where the press can benchmark its latest mobile silicon. The company invited us to Austin to test Mullins back in April, and this time, it hosted an event in San Francisco where we could test a 15.6″ whitebook based on the FX-7600P.

We would have loved to test an ultra-thin laptop based on the FX-7500, but alas, AMD didn’t have one of those at the event. The FX-7600P whitebook was surprisingly slim and light for that form factor, however, as the photo above shows. Along with the 35W Kaveri APU, the machine came loaded with 8GB of RAM and a 256GB Samsung XP941 solid-state drive. Its 15.6″ display had a 1920×1080 resolution and multi-touch input.

As far as we’re aware, the whitebook doesn’t directly correspond to an upcoming system from an AMD partner. If a similar machine were to hit stores, however, AMD says it might retail for something like $699.

So, what did we compare this thing against?

As it turns out, laptops based on Intel’s standard-voltage mobile processors are relatively rare these days. Systems featuring the company’s 15W, ultrabook-focused CPUs are much more commonplace. Faced with such a dearth of competitors, we wound up turning to the nice folks at Zotac, who supplied us with one of their Zbox ID92 mini PCs. The Zbox is based on a Core i5-4570T, which is technically a desktop processor. However, it has a 35W power envelope and specs nearly identical to those of the mobile Core i7-4600M—one of the chips AMD identified as a direct rival to the FX-7600P.

Like the Core i7-4600M, the Core i5-4570T we tested is a Haswell chip with two cores, four threads, a 2.9GHz base speed, a 3.6GHz Turbo speed, 4MB of cache, and Intel HD Graphics 4600. The only major difference is in graphics clock speeds. Where the i5-4570T has base and Turbo graphics clocks of 200MHz and 1.15GHz, respectively, the i7-4600M runs its IGP between 200MHz and 1.3GHz. The i7-4600M also has a slightly higher TDP of 37W instead of 35W.

In non-graphics benchmarks, the i5-4570T should give us a good sense of what Intel’s competition to the FX-7600P looks like. Just keep in mind that, in our graphics benchmarks, the i5-4570T may be slower than its mobile counterparts.

We also included, as a reference point, the Trinity whitebook AMD sent us a couple of years back. That system is based on the A10-4600M, which was the fastest 35W flavor of Trinity at the time. AMD released higher-clocked Richland chips last year, but Trinity and Richland are based on the same silicon, and Richland didn’t raise clock speeds by very much at 35W. The A10-4600M will give us some good historical context for the FX-7600P’s performance.

Oh, and one last thing: we outfitted both the Zotac and Trinity systems with 4GB of DDR3-1600 memory, while AMD set up the Kaveri machine with 8GB of DDR3-1866 RAM. DDR3-1866 may not be the fastest speed Kaveri supports, but it’s still faster than DDR3-1600—and the faster the memory, the better the integrated graphics should perform. As for the disparity in memory capacities, we don’t expect it to have a significant impact on our benchmarks, but keep it in mind nonetheless.

The results

Productivity apps

Let’s see. Kaveri comes out very slightly ahead in 7-Zip decrompression, and it has a clear edge in data encryption using TrueCrypt’s Twofish scheme. The Core i5-4570T, meanwhile, has a nice lead in 7-Zip compression, TrueCrypt’s AES scheme, and x264 encoding. I’d say that’s a win for Haswell overall.

Web browsing

Mobile Kaveri also falls behind in these two web benchmarks, which we used to measure JavaScript performance in Google Chrome. We saw a similar competitive breakdown when stacking up Kaveri against Haswell on the desktop. No question about it, AMD is at a disadvantage here.

3D rendering

Kaveri loses another round in Cinebench 3D rendering. As in the other tests, the FX-7600P offers a nice performance increase over Trinity, but it’s not quite quick enough to catch up to Intel.

The results—continued

GPU computing

AMD is keen to emphasize that its APUs feature powerful integrated graphics capable of doing serious GPU computing work. We didn’t have a chance to test any apps that harness Kaveri’s HSA capability, but we did run LuxMark and Musemage, which use the GPU to speed up 3D ray tracing and image editing, respectively.

For LuxMark, we ran the LuxBall HDR test. For Musemage, we simply fired up the built-in benchmark.

The Core i5-4570T comes out ahead in LuxMark’s CPU-only benchmark, but Kaveri has the upper hand in all of the other tests. Again, keep in mind that Intel’s 35W mobile chips have higher graphics clock speeds than the i5-4570T. Based on our desktop results, however, it seems doubtful that even a higher-clocked Haswell IGP would match Kaveri’s built-in Radeon in Musemage and LuxMark.

Graphics and gaming

Well, those 3DMark scores are awfully close. Might we see a similar equivalency in actual games?

To answer that question, we ran three titles from our Steam library: BioShock Infinite, Dirt Showdown, and Tomb Raider. We ran these at 1920×1080 with antialiasing disabled. “Medium” detail presets were used for BioShock and Dirt, and the “Low” preset was selected for Tomb Raider. Since testing time was limited, we used each game’s built-in benchmark.

The answer seems to be “nope.” Kaveri is the clear winner in these titles. Heck, even the two-year-old Trinity chip comes out ahead of the Core i5.

This is probably a good time to point out that the FX-7600P seems to have enough muscle to maintain playable frame rates in DiRT Showdown and Tomb Raider at 1080p—which, you know, ain’t bad for a mobile processor with integrated graphics. Performance in BioShock Infinite wasn’t quite as impressive, but that’s my fault. I should have tested at the “Low” detail preset rather than “Medium.” Unfortunately, by the time I’d realized my mistake, I was already on the plane home. Such are the dangers of out-of-town benchmarking sessions…

Battlefield 4 and Mantle

Since Kaveri’s integrated GPU is based on the Graphics Core Next architecture, it supports AMD’s Mantle API. We were curious to see how much, if at all, Mantle could help performance on this chip. We fired up Battlefield 4 to find out.

Testing was conducted at 1366×768 using the “Low” detail preset. These are the same settings we used to benchmark BF4‘s Mantle renderer on the 65W, desktop version of Kaveri.


Mantle gives the FX-7600P a small boost in frames per second, but according to our 99th-percentile-frame-time metric, its performance is no better than with Direct3D. The Mantle renderer also spends a little more time on frames that take longer than 50 ms to process, though the numbers there are much better than those we saw earlier this year. Mantle doesn’t induce huge frame-latency spikes like it used to.

Subjectively speaking, enabling Mantle does appear to improve input responsiveness and general smoothness in a noticeable way. Perhaps this has to do with what we’re seeing in the percentile graph above: the Mantle renderer spends less time than the Direct3D one on most frames, but it spends more time on the last 2% or so. We could also be seeing the effects of a shorter submission queue for frames, which would lead to less overall latency between input and output. That may be the more likely explanation.

We’re talking about subtle gains, for sure, and not everybody may want a general smoothness gain at the expense of more high frame times. Still, Mantle feels smoother, and it helps make BF4 fairly playable on this mobile incarnation of Kaveri.

Conclusions

So, that was our first brush with Kaveri in a mobile setting. What did we learn?

At 35W, Kaveri offers somewhat lackluster CPU performance compared to Haswell, but its graphics and GPU computing capabilities are excellent. This is, of course, the same picture our data has painted practically every time we’ve looked at a desktop or notebook AMD APU in recent years. These results should come as no surprise to folks familiar with the current state of the CPU industry.

What’s missing from our results is an indication of whether Kaveri can compete on battery life—something we couldn’t test during our benchmarking session, which was limited to just a few hours. We’re especially curious to know how the 19W ultra-low-voltage Kaveri chips fare against Intel’s 15W ultrabook processors. AMD may claim equivalency, but that’s something we’ll need to verify for ourselves.

Yes, it’s really too bad AMD didn’t give us a 19W Kaveri system to play with. As I noted earlier, notebooks with standard-voltage Intel processors seem to be getting harder to come by, while slim and light machines with 15W CPUs are all over the place. If AMD is the least bit competitive at 19W, then I think notebook vendors will happily follow the same template already set on the Intel side. That, in turn, might mean the 19W versions of Kaveri will be the most popular ones.

When we broached the subject with Kevin Lensing, AMD’s Senior Director of Mobility Solutions, he conceded that notebooks in general are “getting thinner.” He also told us we should indeed see AMD in a greater number of “mainstream thin notebooks” thanks to Kaveri. Interestingly, though, Lensing added that AMD won’t be chasing the $899 and higher price points occupied by more upscale ultrabooks, since the market for those is smaller.

In any case, it does seem like the real contest will take place in lower-cost ultra-thin systems—and we still don’t quite know what to expect there.

Comments closed
    • maxxcool
    • 6 years ago

    “”The marketing-speak was a little fuzzy, but the gist is that the FX name has gone freelance, and AMD will apply it to whatever processors it sees fit.””

    Pathetic.

      • tipoo
      • 5 years ago

      The FX-60 hangs its head in shame at its successors.

    • tbone8ty
    • 6 years ago

    Waiting for a nice gaming laptop with the top FX chip and dual graphics. Now that amd has frame pacing under control should make for a decent gaming experience.

    • Unknown-Error
    • 6 years ago

    As usual no power numbers from AMD! So we can all safely disregard this “AMD Announcement”.

    • deruberhanyok
    • 6 years ago

    Gigabyte: I can has Kaveri FX Brix? Because that looks tasty. Damn tasty.

    • Chrispy_
    • 6 years ago

    The A10-7300 looks like a real keeper:

    4 threads at up to 3.2GHz and 384 GCN Shaders, yet only 19W TDP.

    If that’s not an ultracompact do-everything chip, I don’t know what [i<]is....[/i<]

      • UnfriendlyFire
      • 6 years ago

      The question is how much will it throttle when it’s fully loaded?

      Notebookcheck reported that when they compared a standard voltage Intel chip (37W) compared to a ULV chip (17W-15W) at the same clock rate, the ULV was able to perform as well as the standard voltage chip until the IGP was stressed. Such as running a video or a game.

      Then the ULV’s CPU throttled while the standard voltage CPU was able to maintain its turbo (assuming no thermal constraints).

        • Chrispy_
        • 6 years ago

        indeed, but Intel’s idea of a TDP is different to AMD’s

        AMD will be playing the “superior IGP” card to hide the fact that their CPU architecture is pretty lame, and the last thing that will look good for them is if the CPU throttles back so much that it bottlenecks their one advantage.

        If it’s 19W, but tailorable from 15-25W then I’d expect that configurable range to cover everything from throttling both CPU and GPU to running both at 100% without throttling.

        • rechicero
        • 6 years ago

        That makes me think if comparing a laptop with a miniPC makes sense. The miniPC probably has better airflow and I guess it’ll be able to keep turbo more time (or not).

    • Andrew Lauritzen
    • 6 years ago

    Btw in the final review I’d love to see the 19W Kaveri compared to the 28W Iris 5100. I imagine given the comments on TDP comparisons below that might be in the same ballpark of power usage of the CPU part itself and the graphics comparison would likely be more interesting.

    Ultimately I guess you have to test battery use/power draw to figure out where the real perf/W stuff goes though, but just wanted to note my request to see that 28W SKU included 🙂

    • ptsant
    • 6 years ago

    I don’t quite understand the point of the “Pro” variants. Same watts, same frequencies. Why? Did I miss something? Does it enable ECC or some other “pro” feature like virtualization or secure technology (TPM or whatever)?

      • clocks
      • 6 years ago

      I was wondering the same thing. Unless there are typos, what’s the point of them?

      • NeelyCam
      • 6 years ago

      If you don’t know, you’re not Pro enough to buy one.

      • gc9
      • 6 years ago

      [quote<][b<]Introducing AMD PRO[/b<] ... Designed especially with business in mind, AMD PRO A-Series components offer outstanding stability and longevity, and enable open, industry-standard DASH manageability. ... [url=http://www.amd.com/en-us/press-releases/Pages/advanced-mobile-apus-2014jun04.aspx<]AMD press release[/url<] [/quote<] [quote<]... Aimed at the commercial marketplace, the APUs feature integrated virtualization software and chip-level security with remote data management. Longer life cycles of 24 months and lower operational costs are also being used to entice businesses and AMD is clearly hoping to match up with Intel's vPro suite of features. ... [url=http://www.notebookcheck.net/AMD-announces-Kaveri-A-Series-APUs-for-notebooks.118205.0.html<]NotebookCheck[/url<] [/quote<] [quote<] ... The Pro series targets business customers with a message of commercial stability and management. AMD guarantees that these APUs will remain available for an extended period of time, so enterprise customers won't need to worry about validating new hardware for a couple years. ... [url=http://www.anandtech.com/show/8119/amd-launches-mobile-kaveri-apus/2<]AnandTech[/url<] [/quote<]

        • UnfriendlyFire
        • 6 years ago

        Aka, longer driver support?

        Also, the HP Elitebook 700s that have the PRO APUs also have TPM enabled.

          • NeelyCam
          • 6 years ago

          [quote<]Aka, [s<]longer[/s<] driver support?[/quote<] ftfy (Sorry, I couldn't resist...)

    • rootheday3
    • 6 years ago

    One other comment – while the graphics performance seen on this top end sku is decent, I am guessing that the design wins (if any) will include a lot of the other skus in the list… most of which take a pretty hefty cut on Compute Units, frequency, or both.

    For example, the top couple 19W parts lose 23% on graphics frequency and 25% of CUs. Combined (assuming linear scaling), this means ~40% loss of graphics perfofmance.

    Its also a question about whether, for the 19W parts, games that light up both CPU and GPU will be fighting for TDP. This means that CPU may not be able to turbo up and perf hit may be significant.

    • mczak
    • 6 years ago

    I think the model data isn’t quite correct. Everybody else has the A10-7400P with 6 CUs, not 8 (making the FX-7600P the only fully enabled chip).
    And other sites also feature a A6-7000 model (same data as the A6 Pro-7050B, each of the Pro chips has an exactly equivalent non-pro model).

      • Cyril
      • 6 years ago

      Yep, the A10-7400P has 6 CUs. Fixed.

      The A6-7000 wasn’t in the spec sheet AMD sent me. Lemme look into it.

      Edit: Just updated the article with the A6-7000 specs. I expect the PCIe lane arrangement is the same as for the other ULV models, but I’m waiting on AMD to confirm.

    • maxxcool
    • 6 years ago

    Moar Mantkle!1! Look at the HUGE amazing 2 frame advantage!!!! :/

      • Phartindust
      • 6 years ago

      Yeah who needs it when you are already crushing intel’s igp, but I’ll take it for the smoother gameplay.

    • rootheday3
    • 6 years ago

    Cyril-

    Was your Zotac Haswell system with 4GB in single channel or dual channel? If it was single channel, that would hurt the iGPU perf by ~30% perf or more… won’t be enough to close the GPU perf gap completely but it would help HD 4600 some.

      • Cyril
      • 6 years ago

      It was dual-channel. I used the same RAM as for the Trinity whitebook.

    • chuckula
    • 6 years ago

    These look a whole lot better than mobile Trinity and the GPU seems to scale pretty nicely too.
    The big question mark is still battery life though, so hopefully TR will get a review unit and give us some numbers.

    If AMD can actually get the OEMs to use these chips in some nicer notebooks, they could do pretty well for themselves and I think lower-power desktop and notebook is where Kaveri becomes much more interestly.

    • ermo
    • 6 years ago

    The FX and A10 versions of this might make sense for, say, cheap-ish 14″ 1600×900 systems with Crucial MX100 drives. 1080p seems like pushing it on the graphics side.

    At that resolution, they’d have decent gaming chops, weigh around 1.6-2kg and still have a large enough battery that they’d be able to last quite a while during light (non-gaming) use, while simultaneously having a nice full size keyboard due to the 16:9 form factor width.

    Buying the top FX version and having it be able to switch between a 19W and a 35W power-profile on the fly via a software tweaking tool might also make sense. Want long run-times and low thermal load on your lap? Set it to the 19W profile. Plug it in and game away on a desk? Set it to 35W.

    And all that at a reasonable price point.

    Of course, the version I’d be interested in would be the CrossFireX-enabled 1080p version. And since that won’t fly on Linux due to abysmal driver support, I’m going to have to pass.

    EDIT: Looking at AMD’s Mobile GPU specs, it would appear that the [url=http://www.amd.com/en-us/products/graphics/notebook/r7-m200#<]Radeon R7 265M[/url<] w/ up to 6 CUs (384 Stream Processsors) and DDR3 RAM is the best bet for a balanced CrossFireX solution alongside the FX and A10 variants.

      • Phartindust
      • 6 years ago

      “Buying the top FX version and having it be able to switch between a 19W and a 35W power-profile on the fly via a software tweaking tool might also make sense. Want long run-times and low thermal load on your lap? Set it to the 19W profile. Plug it in and game away on a desk? Set it to 35W.”

      Now that’s an interesting thought. I wonder if that much of a range could be accomplished by just using turbo, and thus scale dependent on load.

      • LostCat
      • 6 years ago

      On Mantle games Crossfire is irrelevant, you don’t need both GPUs the same.

    • ronch
    • 6 years ago

    Now IF ONLY we’ll actually see these things on store shelves…

      • Shouefref
      • 6 years ago

      I already posted on TR that it’s very difficult to find AMD’s cpu’s/apu’s/whatever-u’s, but it seems some people don’t want to believe me.

        • derFunkenstein
        • 6 years ago

        I was wandering the electronics area at Wal Mart a couple nights ago waiting for my wife to finish something. There were two AMD notebooks there, out of probably 8 Windows machines. One was an E2, the other was a Richland A10-5000 series. Not much for choice.

      • Phartindust
      • 6 years ago

      From Hothardware:

      “And for the record, AMD has already announced several design wins out at Computex 2014, with machines from Acer, ASUS, Dell, HP, Lenovo, Samsung and Toshiba to name more than a few. In fact, we’re pretty impressed with the traction AMD has at launch for Kaveri mobile and it’s a good sign that their new solution is compelling enough for OEMs to immediately develop products based on the architecture.”

        • ronch
        • 6 years ago

        If only these laptops came out in January or February of this year.. Oh well. Better late than never, I guess. I surely would’ve opted for these if I am in the market for a laptop right now, assuming they’re priced right.

      • Shahnewaz
      • 6 years ago

      I used AMD’s own shopping site to find AMD powered laptops.
      [url<]http://shop.amd.com/us/Pages/ShopAMD.aspx[/url<] I was lucky enough that the country could be changed to Canada.

    • Shouefref
    • 6 years ago

    “At 35W, Kaveri offers somewhat lackluster CPU performance compared to Haswell, but its graphics and GPU computing capabilities are excellent. ”

    But what about office applications (Office office and other offices) ?
    I have the feeling TR only deals with gaming applications.
    Would AMD’s APU’s also feel lackluster for office and database applications?
    And considering its graphics and CPU capabilities, what would it do for Adobe’s programs, like InDesign?

      • gc9
      • 6 years ago

      The closest in previews so far is only overall PCMark8 area scores, not individual programs (e.g., see [url=http://www.anandtech.com/show/8119/amd-launches-mobile-kaveri-apus/3<]Anandtech[/url<] or [url=http://www.pcper.com/reviews/Processors/AMD-Kaveri-Mobile-Preview-AMD-FX-7600P-Performance/PCMark8-and-GPU-Performance-Co<]PCPer[/url<]). The Applications test requires the actual programs to be installed. [quote<] [b<]Home test[/b<] [i<]Common home computing tests[/i<] ... Home includes workloads for web browsing, writing, gaming, photo editing, and video chat. ... [b<]Creative test[/b<] [i<]Tests for experts and professionals[/i<] ... web browsing, photo editing, video editing, group video chat, media transcoding, and gaming workloads. ... [b<]Work test[/b<] [i<]Simple office productivity tests[/i<] ... basic office work tasks, such as writing documents, browsing websites, creating spreadsheets and using video chat. ... [b<]Applications test[/b<] [i<]Advanced and Professional Editions[/i<] ... system performance using popular applications from the Adobe Creative Suite and Microsoft Office. ... ... [b<]Advanced Edition[/b<] [i<]The Complete Benchmark[/i<] ... [list<][*<]Benchmark with Adobe & Microsoft applications.[b<]*[/b<][/*<][/list<] ... [sub<][b<]*[/b<] You must have the relevant Adobe and Microsoft applications installed in order to run the PCMark 8 Applications test. PCMark 8 is compatible with Adobe Creative Suite 6, Adobe Creative Cloud and Microsoft Office 2010 or later. ...[/sub<] [b<]Professional Edition[/b<] [i<]The industry standard[/i<] [list<][*<]Licensed for business use.[/*<][/list<] ... [url=http://www.futuremark.com/benchmarks/pcmark8<]PCMark 8[/url<] [/quote<]

        • Shouefref
        • 6 years ago

        Thank you.

        PCMark8. I try not to forget that.

      • xeridea
      • 6 years ago

      Lackluster single threaded, it fares much better with multiple threads. For office applications that support OpenCL, the APUs get a huge boost, far surpassing just CPU, and this would increase if there was HSA support in the future.
      If heavy singlethreaded workloads are something you do on your laptop, its not the best choice, it is pretty well rounded though.
      Edit: look at HSA speed of OpenOffice from this review… 5x faster than Intel: [url<]http://www.extremetech.com/computing/174632-amd-kaveri-a10-7850k-and-a8-7600-review-was-it-worth-the-wait-for-the-first-true-heterogeneous-chip/5[/url<] Edit2: This review has a lot more benchmarks. Still the APU (ignoring the OC), gets huge benifit from OpenCL/HSA, Intel benifits are slim, or not supported. [url<]http://www.hardcoreware.net/amd-kaveri-a10-7850k-apu-overclocked/3/[/url<]

        • NeelyCam
        • 6 years ago

        Why the downthumbs? The links support what is being said

          • chuckula
          • 6 years ago

          Anybody pulling out the OpenCL in office apps canard deserves the downthumbs… especially when the only evidence of OpenCL being used anywhere is in a hacked-up Libreoffice demo that does more to prove how poorly optimized the base Libreoffice code is than anything else.

          Oh, and the “fares much better with multiple threads” is showing a 4 “core” mobile part losing by smaller margins to dual-core competitors in heavily-threaded tests. That’s not exactly earth-shattering.

          The issue with fanboys: You could have just said “Oh, the IGP is better, plays games better!” and left it alone. You had to take the bridge too far by pretending that AMD’s propaganda overcomes the real and glaringly obvious deficiencies with these chips by pretending that they somehow dominate Intel in performance when a two year old could see that’s not the case.

            • NeelyCam
            • 6 years ago

            [quote<]Oh, and the "fares much better with multiple threads" is showing a 4 "core" mobile part losing by smaller margins to dual-core competitors in heavily-threaded tests. That's not exactly earth-shattering.[/quote<] So what? It doesn't have to be earth-shattering. The value proposition from AMD is still "mediocre CPU, solid GPU". And what I see is that OpenCL and HSA improved "office apps" performance quite significantly, optimized or not. If you wanna use the "well it's unoptimized code" argument to put down HSA, can you show me a better-optimized office app, and how Intel and AMD CPUs peform there? Even better - if you can find a non-Libreoffice result for an Intel CPU doing the same task as AMD is doing with Libreoffice HSA-enabled, that would be great. Without showing that Intel can do so much better in an "optimized" office app than AMD can with Libreoffice, your argument is weak.

            • Andrew Lauritzen
            • 6 years ago

            Nah I’m with chuckula on this one. Demonstrating a speedup on shitty code is hardly interesting in an architectural discussion. The burden of proof is squarely on the people claiming that such things aren’t possible to do on the CPU, etc.

            Frankly I’ve been doing “GPGPU” stuff since the start and I still have severe doubts that I couldn’t achieve whatever they are doing on the CPU as well, if not do better. It’s typical “the GPU is 1000x better than my scalar single-threaded pointer chasing code when I port it to CUDA” BS that has no bearing on a hardware/architecture discussion.

            You can even call BS right from the numbers themselves. Unless you’re using the texture sampler (… in an office app?) or maybe reduced-precision transcendentals (apples-to-oranges) you shouldn’t be able to get 5x over the CPU on these SKUs, even on AMD’s somewhat weak SIMD implementation. Definitely not over Haswell.

            Even OpenCL implementations themselves are still highly questionable right now. One look at results across various implementations is enough to demonstrate that they have very little to do with the hardware they are running on. It’s a complete crap-shoot at the moment.

            • dragontamer5788
            • 6 years ago

            I can agree to your criticisms over LibreOffice benchmarks.

            But Corel Aftershot is an older, more optimized program and the HSA results on those benchmarks demonstrate a clear advantage to HSA implementations. Yes, even over Haswell. Other OpenCL benchmarks includes Winzip (Haswell wins barely), but WinZip’s performance greatly improves with HSA, almost matching Haswell.

            There is a bit of a hype-train around HSA and AMD, but lets not ignore the facts. HSA does improve AMD’s speeds and makes it a more competitive chip against Intel.

            • Andrew Lauritzen
            • 6 years ago

            I dunno, some googling seems to indicate that one of the things they did for OCL Aftershot stuff was kernel fusion. That’s definitely a case that if they aren’t doing it on the CPU today (which usually people don’t because “JIT is hard” n stuff…) then the comparison is again pretty meaningless. Show me similar code in ISPC as the OCL stuff and let’s see what the performance is.

            It may not be the case as the details are light, but I’m not even convinced by what I see there. I’ve just been involved in too many of these GPU optimization situations to be fooled by the standard tricks. Obviously it’s great that they are optimizing the code, but mapping that back onto an argument about relative hardware efficiency requires a lot more diligence.

            Anyways I obviously don’t know the details here so maybe there’s something real, but whenever you’re not running the same code you have to convince me that your comparison is legitimate, not the other way around.

      • maxxcool
      • 6 years ago

      In a word yes. This would be a terrible DB processor.

    • Alexko
    • 6 years ago

    “At 35W, Kaveri offers somewhat lackluster CPU performance compared to Haswell, but its graphics and GPU computing capabilities are excellent. This is, of course, the same picture our data has painted practically every time we’ve looked at a desktop or notebook AMD APU in recent years. These results should come as no surprise to folks familiar with the current state of the CPU industry.”

    Kaveri’s single-threaded performance is lacking, but in multi-threaded benchmarks it does pretty well. Granted, Intel has better 35W (or rather 37W) CPUs with 4 cores, but those aren’t in these benchmarks.

    And more to the point, this marks a significant change from the previous situation where AMD was well behind in every CPU benchmark.

      • UnfriendlyFire
      • 6 years ago

      Tom’s Hardware pitted the FX-7600p against the i7-4702MQ (quad-core CPU):

      [url<]http://www.tomshardware.com/reviews/fx-7600p-kaveri-apu,3842-3.html[/url<]

      • Phartindust
      • 6 years ago

      Yes, progress is being made. I’m glad to see it to, and hope it continues, since this will lead to a better competitive landscape. We all win when that happens.

    • Andrew Lauritzen
    • 6 years ago

    Curiously, how much does Kaveri include of the chipset stuff in the 19W TDP? Haswell-U’s 15W includes the CPU, on-package PCH and FIVR so it may not be realistic to assume that we’re talking about a 4W difference (25% more) at a system level, which is what matters for battery life.

    Anyways looking forward to seeing a full review! Cheers for the preview info.

      • derFunkenstein
      • 6 years ago

      Nothing, I believe. Same A-series chipset as the bigger boys.

    • UnfriendlyFire
    • 6 years ago

    I think HP has some interesting designs. Well, they sorta took a decent Intel laptop (Elitebook 840) and used an AMD APU instead, based on the fact that the 745 G2 has exactly the same dimensions, weight, battery options, and look, compared to the 840:

    Elitebook 745 G2: [url<]http://www8.hp.com/emea_africa/en/products/laptops/product-detail.html?oid=6732430#!tab=specs[/url<] 14" 1080p, goes up to Pro A10-7350B (2.1/3.3 GHz) 3-cell 50 WHr with the option of installing a 6-cell 60 WHr battery slice that fits underneath the laptop, bringing the total battery capacity to 110 WHr, if you're willing to pay the price. Elitebook 725 G2: [url<]http://www8.hp.com/emea_africa/en/products/laptops/product-detail.html?oid=6732446&jumpid=ba_r329_hhoaffiliate&aid=38293&pbid=TnL5HPStwNw&aoid=35252&siteid=TnL5HPStwNw-Zf2AoJGtzfZ6FWWZPwqWCA#!tab=features[/url<] 12.5" 1080p, goes up to Pro A10-7350B 3-cell, 46 WHr only, no battery slice option These are the only English version of the product pages I could find. There's also a Norwegian one, but no US/UK page. And Googling the laptop models will bring up product documentation and tech support. The question is how long will it take for HP to allow customers to start buying the laptop?

      • derFunkenstein
      • 6 years ago

      I really like that the 14″ model has 1600×900 as an option. For my eyes 14″ 1080p is too small, but that looks to be just right.

      • UnfriendlyFire
      • 6 years ago

      Looks like TR reported about this: [url<]https://techreport.com/news/26566/kaveri-apus-land-in-three-hp-elitebook-laptops[/url<]

      • Phartindust
      • 6 years ago

      Looks like you can get the 12.5″ 725 starting today.

      [url<]http://shopping1.hp.com/is-bin/INTERSHOP.enfinity/WFS/WW-USSMBPublicStore-Site/en_US/-/USD/ViewProductCompare-Dispatch[/url<] Hp is charging $100 for the 1080p screen upgrade as expected. BUT $200 for a 180gb SSD? Think I'd get the Hdd, buy a better SSD and install it my self. Say one of these depending which form of SSD Hp is using. [url<]http://www.tomshardware.com/reviews/ssd-recommendation-benchmark,3269-3.html[/url<]

    • xeridea
    • 6 years ago

    [quote<]However, because of its lower thermal density, the FX-7500 can apparently squeeze into systems with the exact same cooling solutions as the i5-4500U.[/quote<] And because Haswell has Crisco for an internal thermal medium. They say next gen they are fixing that though.

      • exilon
      • 6 years ago

      Mobile chips don’t have an integrated heatspreader.

        • auxy
        • 6 years ago

        Even more than that, the TIM on Haswell chips is very good. The problem is the air gap, not the TIM itself, which is of very high quality.

    • MadManOriginal
    • 6 years ago

    I can’t wait to see the first notebook with these.

    See you in 2015!

      • windwalker
      • 6 years ago

      Indeed.
      Whitebook probably means no design win as of yet so it’s going to take a while.

      In other non-news AMD is still hilariously bad at naming products.

        • shank15217
        • 6 years ago

        Thats a shame because there is a market for them. Intel hasn’t been catching up to amd on the graphics side as many have predicted and they have taken some extreme measures in that effort, but AMDs cpu performance remains lackluster, not sure if they should even bother to release their excavator core before moving on to K12.

      • Hattig
      • 6 years ago

      Brickbook more like. AMD should offer their in-house designs to third party sellers to rebrand, in the same way they do with their GPUs.

      The current OEMs don’t do AMD any favours. We’ll see bulky 15.6″ 1366×768 systems with Kaveri in, and the OEMs will wonder why they don’t sell next to slimmer, nicer systems that just happen to have Intel inside.

      Also AMD should just ship out the systems to review sites. Doing it in house will just lead to substandard testing and unanswered questions about real world battery life.

      And why not also provide a 19W system, for like-for-like testing? That’s what people are interested in these days, the competition is 15W-25W notebooks, not 35W notebooks.

      And whilst I know you had to use that ZBox, it should be pointed out that the Zotac box might have a similar CPU, but it doesn’t have a similar chassis, and that means different cooling capabilities, and thus the processor will be running under different conditions (able to turbo for longer, etc).

      But it’s a good improvement by AMD. And $699 is a good price for a mid-range 1080p gaming notebook with SSD.

        • shank15217
        • 6 years ago

        yea, OEMS are the biggest enemy for amd. Every time an OEM bothers to make a good and laptop, it flies off the shelf but so few do it. For example HP made a killing with their amd netbook, and chrome is would be awesome on one of these. The GPU can drive some really high resolutions.

Pin It on Pinterest

Share This